Test and review: AMD Kaveri in Dual Graphics configuration. Iron experiment: playing in Full HD resolution on graphics built into the processor

Perhaps the key trend in today's consumer electronics market is the search for new forms of goods. The release of devices “tailored” for narrow groups of buyers, the expansion of the functionality of existing devices and the release of familiar devices in fundamentally new form factors - all these phenomena can be combined under this heading. And the logic of the manufacturers is not difficult to understand: the period of consumer euphoria and market saturation has long passed, and now most consumers prefer to choose even such popular devices as smartphones and tablets based on a combination of characteristics and functionality - let alone personal computers, which are found in almost every home today .

Of course, it cannot be said that supply catastrophically exceeds demand, however, it can be noted that many users have already decided which device best suits their tasks, and are in no hurry to change the hardware they already have and perform its functions to modern analogues. Those who decide to update their fleet of equipment do so based on more rational motives than the desire to get everything at once: most often, the purchase is preceded by an accurate definition of tasks and a careful selection of hardware that can solve them as efficiently as possible.

Some people only need Internet access, music and video playback. Some people require high computing power and disk subsystem performance, but the graphics part is not at all important, others are tired of the quality of modern games and want to return the experience of the golden classics - the number of options is unlimited. Will a PC in its traditional form be able to cope with these tasks? Yes, definitely. But a much more interesting question is the feasibility of using a universal platform where only a certain part of its resources is used.

The iron manufacturers themselves understand the relevance of this issue and respond to market demands in the ways outlined in the first paragraph of this article. There is a need to organize home server- get a case in which the number of seats for data drives is limited only by the height of the chassis itself. You need an ultra-compact PC for surfing the Internet and other undemanding tasks - here is a ready-made platform with a soldered-in processor and passive cooling. If you want to get serious about design and modeling on your home computer, modern video cards can not only process graphics in games, but also act as computational accelerators, significantly facilitating the process of rendering scenes and applying filters.

Returning to the topic of this article - the development of integrated graphics is also a consequence of the processes outlined above. If earlier no one considered integrated chips seriously - it would be nice not to interfere with office workers doing reports in text editors - then today it is already clear that due to various reasons For many home PC owners, a discrete video card is unnecessary. Therefore, modern integrated chips must not only provide a certain minimum level of performance in applications that are not overloaded with graphics - their tasks also include correct work with modern web browsers not shy about using GPU resources, and playing high-definition video content, and even games, which the role of an HTPC or living room platform not only does not exclude, but even facilitates.

In other words, regardless of the tasks performed on the computer, integrated graphics should not act as a “weak link”. Therefore, this article intends to examine the performance of modern solutions of this type from two perspectives: how the integrated cores of Haswell and Kaveri processors are ahead of their direct predecessors, and whether they can be considered as an alternative to entry-level discrete video cards.

Meeting the participants

Solutions of the previous model range, namely Intel HD Graphics 4000 And Radeon HD7660D/HD8670D have already been discussed in detail by the author in a number of previous articles, and repeating what was said earlier does not make much sense. In addition, the architectural features and performance of these solutions have long been studied by users and can only be of interest as a “starting point” for comparison with their modern counterparts. Therefore, let's move on to getting to know the new products.

Intel HD Graphics 4600

It’s also worth making a reservation - this article does not discuss mobile graphics, which may well serve as a topic for a separate study, but exclusively desktop solutions, so the choice of HD Graphics 4600 looks quite justified - this particular graphics core is the most productive solution in the current line of Intel central processors . Yes, the company promises to equip the upcoming Devils Canyon processors with HD Graphics 5000 series chips, but for now these certainly interesting cores remain the prerogative of exclusively mobile processors.

The HD Graphics 5000 line (namely, the Iris Pro (HD 5200), Iris (HD 5100) and the HD 5000 chips) would be of interest primarily because their key feature is the second computing unit, which proportionally increases the number of rasterization units and pixel pipelines and computing cores, and also allows you to distribute the load between two nodes. Add to this the increased cache size and some tricks to solve the problem of RAM speed insufficient for the needs of integrated graphics... but unfortunately, due to the reasons stated above, users of desktop systems are forced to be content with only HD Graphics 4600, the architecture of which turns out to be much simpler.

Unlike the older solution, this graphics core does not provide revolutionary changes. Essentially, the HD 4600 is an evolutionary development of the HD 4000, using the same architecture and layout principle, but offering a larger number of execution units. The HD 4600 has 20 shader processors, two rasterization units and four texture modules - thus, solely according to the passport data, the new product should be a quarter ahead of its predecessor.

The remaining innovations in the HD 4600 do not directly affect graphics performance, but are also worthy of attention. Thus, the chip received support for DirectX11.1, OpenCL 1.2 and OpenGL 4.0 instructions, and also supported Direct Compute 5.0 and Intel Quick Sync technology. Among the useful innovations, noteworthy is the ability to connect up to three monitors to the integrated core, combining them into a single working space- previously this was a distinctive feature of discrete graphics.

AMD R7 Graphics

Unlike Intel, which takes advantage of the scalable architecture of its graphics cores and produces more powerful solutions mainly due to an increase in the number of execution units, AMD has made a long-awaited revolution. As you know, Devastator graphics cores in Richland and Trinity processors were based on the outdated VLIW4 architecture, which underlay the HD6000 series video cards. Currently, the corresponding chips are preserved only in the ultra-budget price segment, giving way to the more progressive GCN architecture, so transferring the graphics part of the APU to it even looks like a slightly belated decision.

So, the graphics part of Kaveri processors is based on an updated version of the Graphics Core Next architecture, which makes it similar to such chips as Hawaii (R9 290 line) and Bonaire (HD 7790 and R7 260 line). Accordingly, support for all proprietary technologies such as improvements in the accuracy of the original LOG/EXP operations and MQSAD optimization to speed up motion estimation algorithms, as well as hardware multimedia processors that are more relevant for HTPC (including True Audio) are included.

The design features of this architecture have already been discussed earlier, so the equipment of the chip is of interest. In its top version, the Kaveri integrated graphics contains 8 computing units (or 512 shader processors), which exceeds the same indicators of the Oland XT chip underlying the Radeon R7 250 video card, and strangely makes this chip similar to the Radeon HD 7750 (Cape Verde Pro) , although it is still impossible to put an equal sign between them. Another similarity is the presence of only one geometry engine in the integrated R7 Graphics, but there are not four raster operations units, like Cape Verde, but only two, like Oland. As readers could see earlier, the R7 250, equipped with fast gddr5 memory, is not greatly hindered by this circumstance, but the integrated graphics, which are forced to borrow part of the RAM that is slow by the standards of video cards... in general, is unlikely to benefit.

In contrast, a positive factor is the presence of eight asynchronous computing engines (ACE), whose role is to distribute tasks between computing units and access to a common second-level cache. Increasing the number of these blocks had a good effect on the performance of the energy-efficient Kabini/Temash platforms, as well as the graphics part of the Playstation 4 (which also has 8 ACEs), so this solution allows us to hope for an effective distribution of the computing load between the blocks.
The remaining innovations, as in the case of HD Graphics 4600, do not directly affect performance, but significantly improve the consumer characteristics of the product. The Universal Video Decoder (UVD) hardware accelerates video playback in H.264, VC-1, MPEG-2, MVC and MPEG-4 formats. The updated version, which received version 4, is essentially no different from the previous one, but AMD claims greater resistance to decoding errors.

The Video Codec Engine (VCE) decoder is an analogue of technologies such as Intel Quick Sync and Nvidia NVEnc. Although third-party observers claim that competitor decoders are still ahead of this AMD solution, the manufacturer’s attention to this aspect cannot but rejoice.

True Audio technology, which is new for integrated graphics, also arouses some curiosity, given that audio is by no means the last thing for HTPCs. In theory, when using this technology, sound processing is taken over by the three cores of the Tensilica HiFi2 EP Audio audio processor integrated into R7 Graphics. Moreover, sound can be output using this technology not only via HDMI or Display Port, but also through a three-millimeter jack - thus, True Audio does not replace a sound card, but complements it, processing sound through sets of effects and algorithms, access to which is provided by True Audio API, a kind of analogue of the Mantle API, only for sound. Unfortunately, the connection to software is a significant disadvantage of this technology: if Mantle is already used not only in Battlefield 4, then the only game with True Audio support so far is the new one Thief.

Test stand and testing methodology

Intel HD Graphics 4000 and Radeon HD8670D were expectedly chosen as rivals for HD Graphics 4600 and AMD R7 Graphics. In addition, the testing included discrete video cards discussed in the previous material - GeForce GT 640 and R7 250, which can be considered the minimum level for gaming accelerators.

The test bench configurations were selected as follows. The components common to all test platforms were:

CPU cooling system: Thermalright AXP-100;
Thermal interface: Gelid GC-Extreme;
RAM: Kingston KHX1866C9D3K2/8G;
Disk subsystem: SSD Kingston SH103S3/120G;
Optical drive: LiteOn iHAP122;
Frame: CoolerMaster 690 II Regular. The standard fans were replaced with two Termalright X-Silent 140 at 650 rpm on the front panel and side wall;
Reobas: Xilence FCP;
Power unit: Corsair CX 750M.

For platform LGA 1155 The following components were selected:

Motherboard: AsRock Z77 Pro3;
CPU: Intel Core i5-3570K.

For platform LGA 1150:

Motherboard: MSI Z87-G43;
CPU: Intel Core i5-4670K.

For platform socket FM2/FM2+:

Motherboard: Asus A88XM-Plus;
CPU: AMD A10-6800K/AMD A10-7850K.

All test processors worked in normal mode, since their performance is obviously sufficient for integrated graphics. The RAM also worked in normal mode - 1600 MHz with timings 9-9-9-27 for HD Graphics 4000 and HD Graphics 4600, and 2133 MHz with timings 10-11-10-30 for Radeon HD8670D and R7 Graphics. The graphics cores themselves were tested in two modes: at standard frequencies and in maximum overclocking mode.

All tests were carried out under Windows 7 Professional with installed package Service updates Pack 1. The following driver versions were used:

AMD: Catalyst 14.4;
Nvidia: ForceWare 335.23;
Intel: 15.33.18.64.3496;

Synthetic tests were carried out with standard settings, tests in games - with average graphics settings corresponding to the level of the tested video cards and graphics cores. Three resolutions were used for tests: 1366x768, 1680x1050 and 1920x1080 pixels. The settings are described in more detail in the graphs.

Synthetic tests

Traditionally, a test package opens the line of synthetics 3DMark 2013. In this version, Futuremark followed the lead modern trends, and from a hardcore benchmark for high-end PCs, its most famous product is gradually turning into universal system for testing platforms of varying degrees of mobility. Therefore, of the three benchmarks, we are only interested in one - Fire Strike, which is still capable of bringing even premium-segment hardware to its knees.

This time there were practically no surprises in this test - the integrated graphics were arranged in the order that corresponds to the performance of these solutions “on paper”. The only interesting thing was that the Kaveri integrated graphics is confidently ahead of the GeForce GT640 here, although looking ahead, in real applications this situation does not always occur.

Next in line is the benchmark Unigine Heaven, which has not received updates for a long time, but still remains quite demanding on video card performance.

But in this test the results are much more interesting. The confident advantage of Haswell integrated graphics over Ivy Bridge is natural, but the gap is much more impressive than in 3DMark. However, the interesting thing is that the HD 4600 is fighting almost on an equal footing with the Radeon HD8670D - a very good result for Intel, and more than a noticeable increase compared to the previous generation. However, the same words can be applied to AMD: the integrated Kaveri graphics here are also noticeably faster than the Richland graphics core. But the competition with the GeForce GT640 no longer ends with a quick victory: Kaveri is faster in nominal terms, but loses in overclocking - obviously, the slow memory is starting to take its toll.

New development from Unigine - benchmark Valley- takes us from fantastic skies to real earth and pleases the eyes of Russian enthusiasts with native pines, birches and glades overgrown with chamomile and fireweed, not forgetting to load and warm up the video cards well.

This benchmark is traditionally more loyal to Nvidia products, so Kaveri can only overclock the GeForce GT640, and that advantage cannot be called noticeable. But what’s curious is that the overclocked Radeon HD 8670D turns out to be slightly faster than the R7 Graphics at nominal frequencies. As for Intel graphics, Haswell is again faster than its predecessor, but it is no longer able to compete with AMD solutions.

Game tests

Batman: Arkham City

The second part of the adventures of the Dark Knight, defender of Gotham. The game does not borrow the surroundings of Christopher Nolan's extremely successful film trilogy and uses a comic book style, which does not prevent it from delivering an excellent plot, a successful combination of action and stealth elements, detective riddles and so on. The modified (once again) Unreal Engine 3 engine is responsible for the graphics part, and nowadays the game can only be called demanding at the highest graphics settings.

In low resolutions with medium settings you can comfortably play even on HD4000, but Haswell still turns out to be faster, especially in terms of the minimum FPS level. Integrated AMD graphics come in a tight group; the balance of power here is approximately the same as Intel solutions, although the level of performance is, of course, much higher. R7 Graphics lags behind GeForce GT640, but the gap between them is not so critical.

With an increase in resolution, the integrated Haswell graphics still provide a minimally comfortable level of FPS, while its predecessor is no longer able to cope with the load. However, all Intel solutions here are noticeably behind AMD products, whose performance can be described as comfortable. The gap between R7 Graphics and GeForce GT640 remains at the same level.

It will be possible to play Full HD on integrated Intel graphics only if the settings are further reduced, but the rest of the test participants maintain the same balance of power as in previous modes.

Battlefield 4

A game that doesn't need any special introduction. Another reincarnation of a recognized leader in the team shooter genre, another new word in graphics, another raid of hordes of fans on the forums - everything is as usual. However, the value of this game lies in the new version of the Frostbite engine, which is already taking away the title today." universal machine world" from the Unreal Engine - in any case, the Bioware studio is creating its new games on this engine, which will be followed by other developers in the future.

Battlefield 4 was created in close collaboration with AMD, so the results should not be surprising. Already in low resolution, of all Intel chips, only the overclocked HD Graphics 4600 produces a more or less smooth framerate, although this is not enough for comfortable gaming. But the Richland and Kaveri graphics cores perform very well here - no matter how strange the idea of ​​playing Battlefield on integrated graphics may seem, in practice it is possible - of course, subject to the selection of settings and resolutions.

However, with an increase in resolution, even the HD8670D approaches the minimum comfortable level - you can still play, but the not-so-smooth frame rate prevents you from achieving success in exterminating your opponents. But R7 Graphics is doing much better, which is explained by the Frostbite engine’s love for the GCN architecture. The integrated Kaveri graphics here outperforms the GT640 in nominal mode and shows a similar level of performance after overclocking.

In Full HD, the GeForce GT 640 is only slightly ahead of the HD8670D and noticeably inferior to the R7 Graphics, but here all the solutions mentioned provide only a minimally comfortable level of performance.

Dirt 3

The last part of the once famous series, which retained at least some connection with real racing competitions. From a gameplay point of view, you can make a lot of complaints about the game, but in terms of graphics, the eighth part of the series is quite good, and also does not have exorbitant system requirements - just what is needed for budget video cards and integrated solutions.

Even at low resolutions, HD Graphics 4000 does not provide the required performance, although its successor, HD Graphics 4600, holds up well. However, the lag between Intel graphics cores and AMD products requires no comment - in fact, their performance ends where the performance of the HD8670D and R7 Graphics begins. But even overclocking does not help the latter to reach discrete graphics in the form of GeForce GT640.

At a resolution of 1680x1050 pixels, the FPS counter stops going off scale when measuring the performance of discrete cards, but otherwise the picture changes little. Thus, from the Intel camp, only the overclocked HD 4600 demonstrates a noticeable result, and the R7 Graphics in overclocking mode finally manages to outstrip the GeForce GT640 at nominal frequencies, but the overall balance of power remains the same.

Full HD resolution with medium graphics settings becomes the final triumph of AMD's APU - you can even play on the HD8670D in nominal mode, and overclocking leaves room for higher settings.

Far Cry 3

The brand, once created and lost by Crytek, picked up by the publishing house Ubisoft, finally got rid of the shortcomings of the second part, returning players from the dull brown savannah to the tropical jungle. The plot (there is one in the game, and this is already pleasing against the backdrop of Crysis 3) delivers with its originality, the gameplay is a combination of shooter and role-playing parts, as well as full-fledged sandbox gameplay, and the graphics deliver on their own.

At low resolutions, integrated Haswell graphics demonstrate an impressive lead over its predecessor and provide the minimum required level of performance. Moreover, in overclocking the HD4600 manages to catch up with the Radeon HD8670D at nominal frequencies. But the R7 Graphics, although it outperforms its ancestor by an equally impressive number of frames, is still unable to catch up with the GeForce GT640, although increasing the frequencies allows it to come almost close.

But with an increase in resolution, you will have to forget about playing on integrated Intel graphics, and the Radeon HD8670D does not cope very well with a resolution of 1680x1050 pixels. But this mode sets a more serious task for the GeForce GT640, which allows the R7 Graphics to catch up with it after overclocking.

In Full HD, the game turns out to be even more demanding on the PC graphics subsystem. The Radeon HD 8670D cannot cope with this resolution even after overclocking, and the R7 Graphics and GeForce GT640 produce almost the same framerate, which is barely enough to make the gameplay more or less smooth.

Hitman: Absolution

A new part of the adventures of the hired killer known under the code number "47". The cold-blooded, completely devoid of emotions anti-hero, during the existence of the series, managed to form an entire army of fans around himself, the ranks of which could not be thinned even by the release of several frankly disastrous parts. However, Absolution is not one of the latter - it has a decent plot and level of production, complex gameplay and the necessary level of player freedom.

Like Battlefield 4, the game is very loyal to AMD video cards, so the results are not at all surprising. The graphics part of the Haswell processor is noticeably ahead of the HD 4000, but even overclocking does not allow it to approach the comfort limit. However, for other participants in the testing, Hitman graphics turns out to be a difficult task: Radeon HD8670D, R7 Graphics and GeForce GT640 are in an extremely dense group, only the R7 250 with gddr5 memory demonstrates a fundamentally different level of performance.

With increasing resolution, the balance of power does not change - the GeForce GT640 is located between the Radeon HD8670D and R7 Graphics, only the R7 250 takes performance to a new level.

In Full HD, R7 Graphics manages to win a convincing victory over the GT640, however, in this resolution with medium graphics settings, the integrated graphics are no longer able to provide an acceptable frame rate.

TES V: Skyrim

Not just another game in The Elder Scrolls series, but this time a worthy successor to Morrowind's laurels. Wiki delivers... Nords, mead, dragons, the harsh and dim beauty of northern landscapes, blooming at night with heavenly lights of various shades, as well as the presence of a sane central plot and a bunch of side quests. Technologically, the game does not bring any revelations, but it turns out to be quite demanding on PC resources, especially at maximum settings and with high-resolution textures.

Without being part of any marketing programs, which happens quite rarely these days, Skyrim is able to work adequately on a wide variety of hardware. Thus, in low resolutions you can comfortably play even on the HD Graphics 4000, and its successor, the HD Graphics 4600, demonstrates a fundamentally different level of performance, after overclocking it is almost equal to the Radeon HD8670D at nominal frequencies. It is noteworthy that the latter, as a result of overclocking, is equal to the R7 Graphics, and the Kaveri integrated graphics is ahead of the GeForce GT640.

What is especially interesting is that owners of integrated Intel graphics cores may not be limited to low resolutions; Skyrim also plays well at 1680x1050 pixels, although HD Graphics 4000 in this case needs to be overclocked. Otherwise, the balance of power does not change - Haswell is again on the heels of the unaccelerated Richland, and Kaveri is approaching the GT640.

In Full HD, the integrated graphics of Ivy Bridge are completely choking, but Haswell is still able to cope with the game, but it is no longer possible to catch up with AMD products. It is curious that at this resolution, AMD graphics cores of both generations demonstrate almost the same performance, and the overclocked R7 Graphics is only equal to the GeForce GT640.

Sleeping Dogs

An unexpected hit in the style of GTA, which had been in development for a long time and SUDDENLY launched in November 2012. Immersing the player in the frightening, but in its own way attractive atmosphere of the Hong Kong underworld, literally saturated with the spirit of John Woo films, the game adds a significant amount of martial arts and Asian flavor to the standard mechanics, which looks fresh and original. The game is a multi-platform project, but the PC version with high-resolution textures is very demanding on the graphics subsystem.

Already at low resolutions, integrated Intel graphics are out of work, while AMD graphics cores provide fairly high performance. The R7 Graphics is even ahead of the GeForce GT640 in nominal terms, although overclocking gives the Nvidia product the victory in this competition.

With an increase in resolution to 1680x1050 pixels, the GeForce GT 640 loses its ardor, stopping between the overclocked Radeon HD8670D and R7 Graphics at nominal frequencies. At the same time, AMD's new product is noticeably ahead of its predecessor, providing a much more comfortable gaming experience.

In Full HD, the gap between the two AMD solutions increases even more, but the GeForce GT640 unexpectedly has the strength to pursue the R7 Graphics.

Tomb Raider

Not just another part of the franchise, known, perhaps, even to people far from computer games, but its complete reboot, made in a much more realistic style. The main character is no longer a goddess of war with fantastic forms, but just yesterday’s student, who for the first time encountered real danger and was forced to fight for survival, and the adventure elements no longer include dashing shooting from pistols with endless ammunition. For which the screenwriters should be thanked. Technically, the game is again a cross-platform project, although the PC version is equipped with many improvements.

Once again, even at low resolutions, Intel graphics are no competition, although the overclocked Haswell graphics core demonstrates acceptable speed. It is noteworthy that the gap between R7 Graphics and its predecessor is not as noticeable here as in the case of Sleeping Dogs, and the new AMD product only manages to get closer to the GeForce GT640.

With increasing resolution, Intel products lose their last ambitions, but this mode is already becoming a test for other test participants. R7 Graphics here demonstrates a noticeable advantage over the Radeon HD8670D, but after overclocking it is only slightly ahead of the GeForce GT640 at nominal frequencies.

In Full HD the situation is repeated again - the R7 Graphics turns out to be faster than the Radeon HD8670D, but the GeForce GT640 is ahead in overclocking mode.

World of Tanks

A game whose name in context Google search appears first when you search for "world of", and that's it. Perhaps one of the first MMO projects that was able to satisfy the needs of users tired of the adventures of long-eared and green-skinned people. At the same time, it is highly popular among history buffs, reenactors, modellers and others involved, which only benefits the player community, reducing the percentage of students and interesting characters. It is distinguished by historical accuracy, a realistic damage model, a rich fleet of equipment, but the gameplay has a fairly low barrier to entry. The first versions of the game had modest system requirements, but as a result of recent innovations, the load on PC hardware has increased many times over.

If at the time of patch 8.11 it was quite comfortable to play on HD Graphics 4600 (which the author, it must be said, did), then with the release of update 9.0, integrated Intel graphics are no longer useful even in low resolutions. At the same time, AMD products, which the BigWorld engine vehemently hates and strives to destroy in every possible way, demonstrate a more than sufficient level of performance. It is worth noting that the overclocked Radeon HD8670D is on par with the R7 Graphics at nominal frequencies, and in general the Kaveri’s advantage over its predecessor here does not look very convincing. On the other hand, even the overclocked Radeon HD8670D manages to outperform the GT640 - considering the engine's love for Nvidia products, this is a good result.

Increasing the resolution only makes the indicated trends more clear. The integrated AMD graphics provide sufficient performance for gaming, but the R7 Graphics does not show a clear advantage over the Radeon HD8670D. At the same time, the integrated Richland and Kaveri cores show the same level of performance as the GeForce GT640.

In Full HD, the overall balance of power remains unchanged, but the GT640 is already confidently ahead of the Radeon HD8670D and only slightly behind the R7 Graphics, which here finally manages to break away from its predecessor.

World of Warcraft: Mists of Pandaria

The great and terrible MMORPG, which has existed perhaps longer than some game studios have been in business, holds the record for relatively honest taking of money from the population. The WoW graphics engine has always been distinguished by excellent optimization: for example, the author of this article, during patches 1.3, managed to play the subject on a GeForce 2 MX 400 installed on his work computer. The video card was already an antiquity back then, but nevertheless it played the game at a resolution of 800 x 600 pixels. A similar situation is observed now: with proper selection of settings, you can play passably even on Intel HD Graphics 2000, but in order to set the parameters to the maximum, you will need almost top-end hardware.

Confirming the above, at low resolutions even the HD4000 provides acceptable performance, although the integrated Haswell graphics are noticeably faster. Unlike WoT, here R7 Graphics immediately gains an advantage over the Radeon HD8670D and, moreover, is ahead of the GT640.

However, at a resolution of 1680x1050 pixels the picture turns out to be different: the R7 Graphics here is not so much ahead of the Radeon HD8670D, and it can only catch up with the nominal GT640 after overclocking.

In Full HD the situation is almost exactly the same, except that the gap between R7 Graphics and Radeon HD8670D slightly increases.

conclusions

As you'd expect, new generations of integrated graphics cores have taken performance up another notch. This is especially noticeable in the example of the Intel HD Graphics 4600, the performance increase of which is observed in literally all test applications, and at times even allows it to compete with the integrated AMD graphics of the previous generation, which the HD Graphics 4000 could not achieve under any circumstances. It is likely that HD Graphics 5200/5100 will be able to match and even surpass the Radeon HD8670D, but their appearance in Devils Canyon processors is still close, but still future. And these processors are not designed to work with integrated graphics, and Haswell and Haswell Refresh will still be equipped with HD Graphics 4600.

In general, the idea of ​​increasing the execution units, carried out by reducing the technical process and reducing the overall power consumption of the chip, worked well - performance has increased so much that when assembling a budget PC, it may be more profitable not to purchase a video card like the GeForce GT630, but to buy a more productive one (including graphics parts) of the processor. And users who primarily need the performance of computing cores do not have to think about buying a discrete video card.

As for R7 Graphics in Kaveri processors, the opposite is true. AMD made the long-awaited technological revolution, transferring integrated graphics to the current GCN architecture, but the performance did not make a revolutionary leap. Yes, Kaveri's integrated graphics are fast - in fact, they're the fastest integrated graphics solution available today, and even the HD Graphics 5200 hardly has a chance to compete with it. But at the same time, the increase in performance compared to the Radeon HD8670D is by no means amazing: yes, games run faster, yes, those resolutions are available in which the predecessor is completely blown away, but budget gaming video cards like the R7 250 still provide much greater performance. On the other hand, no one has canceled the obvious advantages of integrated graphics. Supplied under the same cover with the processor, unlike a discrete video card, it does not require additional expenses, does not take up much space, allowing you to assemble the system in an ultra-compact case, and allows you to get rid of an excess heat source, which in the latter case can be a significant argument.

Thus, what conclusion can be drawn from this article? In fact, despite the increased overall performance, the balance of power between Intel and AMD in the integrated graphics segment has not changed at all compared to the days of Ivy Bridge and Richland/Trinity. Intel continues to focus on the cores, and AMD's APU concept remains uncompetitive where it is intended. Just as Richland used to be the best offer for a home multimedia PC without discrete graphics, Kaveri now occupies this position. The video and audio playback capabilities of the new products are a little wider, but the implementation of these capabilities directly depends on the software, and therefore cannot be considered an advantage in the full sense of the word.

It cannot be said that AMD produces weak video cards, especially in the inexpensive segment. The performance of video cards is often sufficient for most tasks. Especially if these are not highly demanding tasks, such as video rendering or working with 3D graphics. To better determine the level of performance, we should consider two AMD Radeon R7 200 Series graphics cards.

The table describes the characteristics of the AMD Radeon R7 200 Series, namely, a comparative analysis of two video cards from this series is presented.

Video card settings	Radeon R7 240
	Oland XT
Core frequency	780 MHz
Graphics memory type	DDR3
Memory amount	2 GB
Memory frequency	1600 MHz
Technical process	28
Threaded Processes	320
Rendering units	8
	20
	128 bit
Transistors	1040 million	1040 million
Heatsink	30 W
Support	DirectX 12

It is worth considering that the base frequency of the R7 240 core is 730 MHz, and 780 MHz is the frequency after overclocking. The video card parameters indicate the memory type is DDR3, but there is also an option with GDDR5 memory. The comparison will use DDR3 since it is the most common type at the moment.

Radeon R7 200 Series Review

AMD Radeon R7 200 Series belongs to the category of budget and affordable video cards. However, it is made quite well. The video cards discussed in this review are from Gigabyte.

Radeon R7 240 review

The model received 2 GB of DDR3 video memory. It also has original factory overclocking. The assembly itself is of high quality, even though this is a budget segment.

On the top graphics card There is a cooling cooler with a large radiator. This decision is due to the strong heating of AMD cards. The radiator is made of aluminum, and the fan itself protrudes slightly. The length of the entire video card is 19.5 cm.

The results in Metro Last Light are quite good. The cores operated at 900 MHz. The video card was loaded at 90-100%, while the average temperature did not exceed 46 degrees. The coolers worked at 33%, and the rpm reached 2 thousand. The cooler made virtually no noise.

Radeon R7 250 review

The external design of the graphics card is no different from the younger model. It also has an electrically insulating coating made of blue PCB and is 19.5 cm wide. The radiator is the same bulky as that of the AMD Radeon R7 240.

The cards in question differ exclusively in memory chips and power phases. The Radeon R7 250 has three-phase power, unlike the two-phase R7 240.

Test results for Metro Last Night are similar. The video card worked stably at 90-100%, and did not get particularly hot. The temperature did not exceed 46-47 °C.

The only difference is the number of revolutions per minute. The fan operated at a speed of 1200 rpm, which is two times less than the speed of the Radeon R7 240. The FPS indicator remained stable around 30-40 frames.

How to overclock a Radeon R7 200 Series video card

First, you will need to install the following utilities: MSI Afterburner, 3DMark, TechPowerUp GPU-Z, FurMark.

1. Launch MSI Afterburner and click on the settings button (gear).
2. Select the “User interface” tab and set the desired language in the settings.
3. Click on the “Settings” button and in the “Monitoring” tab bring up the following parameters: GPU core frequency, GPU1 memory frequency, frame rate, GPU1 temperature.
4. For each of the selected parameters, set the “Show in Overlay Display” option and save the changes.
5. Click on the “Settings” button again and in the “Basic” tab, check the boxes for “Unlock voltage control” and for “Unlock voltage monitoring.”
6. Launch the FurMark program and select the desired screen resolution, as well as the maximum available anti-aliasing.

Now the most important stage is overclocking the AMD Radeon R7 200 Series video card. We start by overclocking the video memory. First, we increase the memory frequency by 100 MHz and save the setting. Then we run the video card in FurMark. We repeat this procedure until the first artifacts appear.

If your computer freezes during testing, you should restart it immediately. After the reboot, we set the parameters at which there are no artifacts.

Finally, we check the card in 3DMark to avoid glare, spots and other defects.

The situation is the same with video core overclocking. We set the “Power Limit” parameter to the maximum, and then increase the core frequency by 10 MHz. We carry out tests in programs that were used to overclock the memory.

If artifacts appear, then we increase the voltage on the core. We repeat the procedure until the desired result is achieved.

Game testing results

In GTA V, both video cards show good results. At low settings Both video cards produced graphics in the region of 35-40 FPS. At the initial frequencies, the R7 240 DDR3 outperforms slightly and produces 10-15 FPS more. Such indicators are achieved not only due to the high performance of video cards, but also due to good level optimization of GTA V.

In the game War Thunder, at base frequencies, video cards produce a stable 35 FPS. And the Radeon R7 240 is 13 FPS ahead of the GT 730. The situation after overclocking is even better. Both video cards from AMD are not only on par with the GeForce GT 730 DDR3 and GeForce GT 730 GDDR5 type, but are also several percent ahead of them. It is worth noting that the graphics settings were set to average values.

Well, the last game is Dota 2. Both cards from AMD work stably around 45 FPS. In heavily loaded scenes, the number of frames dropped to 25-30 FPS. At base frequencies, the Radeon R7 240 outperformed the GeForce GT 730 by 25 FPS.

The situation with the R7 250 is a little worse. The lack of overclocking of the video memory frequency greatly affects the performance gain. That's why FPS indicator The Radeon R7 250 is slightly lower than the GeForce GT 730 (GDDR5). Tests were carried out at minimum graphics settings.

Overall, AMD Radeon R7 200 Series gaming tests show satisfactory results. Video cards are capable of running quite modern games, albeit at low settings. A comparative analysis showed that in most cases, video cards from AMD are ahead of video cards from Nvidia. But you need to take into account that video cards are in the budget segment.

Analysis of geometric average results, purchase attractiveness and measurement of energy consumption

Introduction

The purpose of the review is to determine the optimal processor for working with the Radeon R7 260X 2048 MB video card.

The list of tested CPUs included:

• Core i7-4770K;
• Core i5-4670K;
• Core i5-4570;
• Core i5-4440;
• Core i5-4430;


• Core i3-4340;
• Core i3-4130;


• FX-8350 BE;
• FX-6350 BE;
• FX-4350 BE;


• A10-6800K;
• A8-6600K;


• A10-5800K;
• A8-5600K;


• Athlon II X4 760K;
• Athlon II X4 740.

The material is an integral part of the “Database of Game Configuration Test Results” project. From it, the data required for this work was sampled. I ask our readers to take into account the fact that a huge and painstaking work has been done, so the author’s hard work should be respected.

This section of articles is for reference only; there are no comments, since each reader will be able to independently obtain the information he needs.

Test configuration

Tests were carried out on the following stand:

• Video card: Radeon R7 260X 2048 MB - 1100/6500 @ 1200/7200 MHz (Sapphire);
• Motherboard #1: GigaByte GA-Z87X-UD5H, LGA 1150, BIOS F7;
• Motherboard #2: GigaByte GA-990FXA-UD5, AM3+, BIOS F12;
• Motherboard #3: ASRock FM2A85X Extreme4, FM2, BIOS 1.70;
• System CPU cooling: Corsair Hydro Series H100 (~1300 rpm);
• RAM: 2 x 4096 MB DDR3 Geil BLACK DRAGON GB38GB2133C10ADC (Spec: 2133 MHz / 10-11-11-30-1t / 1.5 V), X.M.P. - off;
• Disk subsystem: 64 GB, SSD ADATA SX900;
• Power unit: Thermaltake Toughpower 1200 Watt (standard fan: 140 mm inlet);
• Frame: open test bench;
• Monitor: 30" DELL 3008WFP (Wide LCD, 2560x1600 / 60 Hz).

Processors:

• Core i7-4770K - 3500 @ 4500 MHz;
• Core i5-4670K - 3400 @ 4500 MHz;
• Core i5-4570 - 3200 MHz;
• Core i5-4440 - 3100 MHz;
• Core i5-4430 - 3000 MHz;


• Core i3-4340 - 3600 MHz;
• Core i3-4130 - 3400 MHz;


• FX-8350 BE - 4000 @ 4700 MHz;
• FX-6350 BE - 3900 @ 4700 MHz;
• FX-4350 BE - 4200 @ 4700 MHz;


• A10-6800K - 4100 @ 4700 MHz;
• A8-6600K - 3900 @ 4700 MHz;


• A10-5800K - 3800 @ 4500 MHz;
• A8-5600K - 3600 @ 4400 MHz;


• Athlon II X4 760K - 3800 @ 4500 MHz;
• Athlon II X4 740 - 3200 @ 4100 MHz.

Software:

• Operating system: Windows 7 x64 SP1;
• Video card drivers: AMD Catalyst 13.10 Beta.
• Utilities: FRAPS 3.5.9 Build 15586, AutoHotkey v1.0.48.05, MSI Afterburner 3.0.0 Beta 14.

Testing tools and methodology

For a more clear comparison of processors, all games used as test applications were launched at a resolution of 1920x1080.

Built-in benchmarks, FRAPS 3.5.9 Build 15586 and AutoHotkey v1.0.48.05 utilities were used as performance measurement tools. List of gaming applications:

• Assassin's Creed 3 (Boston port).
• Bioshock Infinite (Benchmark).
• Crysis 3 (Welcome to the Jungle).
• Far Cry 3 (Chapter 2. Hunter).
• GRID 2 (Benchmark).
• Hitman: Absolution (Benchmark).
• Medal of Honor: Warfighter (Somalia).
• Sleeping Dogs (Benchmark).
• Tom Clancy's Splinter Cell: Blacklist (Attack of the American base).
• Tomb Raider (Benchmark).
• Total War Rome II (Benchmark).
• World of Tanks (Airfield).

Measured in all games minimum And average FPS values. In tests in which there was no possibility to measure minimum FPS, this value was measured by the FRAPS utility. VSync was disabled during testing.

Component Specifications

Overclocking processors

The processors were overclocked as follows. The stability of overclocking was checked using the OSST 3.1.0 “Perestroika” utility by running the CPU for half an hour on the maximum matrix with a forced 100% load. I agree that overclocking the tested CPUs is not absolutely stable, but for any modern game it is one hundred percent suitable.

With maximum overclocking for all AMD processors, the memory controller frequency was raised to 2400-2800 MHz.

Core i7-4770K

Regular mode. Clock frequency 3500 MHz, base frequency 100 MHz (100x35), DDR3 frequency – 1600 MHz (100x16), supply voltage 1.08 V, DDR3 supply voltage – 1.5 V, Turbo Boost – enabled, Hyper Threading – enabled.

The processor was overclocked to a frequency of 4500 MHz. To do this, the multiplier was raised to 45 (100x45), DDR3 frequency - 2133 MHz (100x21.33), supply voltage - up to 1.25 V, DDR3 supply voltage - 1.5 V, Turbo Boost - off, Hyper Threading - off.

Core i5-4670K

Regular mode. Clock frequency 3400 MHz, base frequency 100 MHz (100x34), DDR3 frequency – 1600 MHz (100x16), supply voltage 1.07 V, DDR3 supply voltage – 1.5 V, Turbo Boost – enabled.

The processor was overclocked to a frequency of 4500 MHz. To do this, the multiplier was raised to 45 (100x45), the DDR3 frequency was 2133 MHz (100x21.33), the supply voltage was up to 1.25 V, the DDR3 supply voltage was 1.5 V, Turbo Boost was turned off.

Core i5-4570

Regular mode. Clock frequency 3200 MHz, base frequency 100 MHz (100x32), DDR3 frequency – 1600 MHz (100x16), supply voltage 1.07 V, DDR3 supply voltage – 1.5 V, Turbo Boost – enabled.

Core i5-4440

Regular mode. Clock frequency 3100 MHz, base frequency 100 MHz (100x31), DDR3 frequency – 1600 MHz (100x16), supply voltage 1.06 V, DDR3 supply voltage – 1.5 V, Turbo Boost – enabled.

Core i5-4430

Regular mode. Clock frequency 3000 MHz, base frequency 100 MHz (100x30), DDR3 frequency – 1600 MHz (100x16), supply voltage 1.06 V, DDR3 supply voltage – 1.5 V, Turbo Boost – enabled.

Core i3-4340

Regular mode. Clock frequency 3600 MHz, base frequency 100 MHz (100x36), DDR3 frequency – 1600 MHz (100x16), supply voltage 1.05 V, DDR3 supply voltage – 1.5 V, Hyper Threading – enabled.

Core i3-4130

Regular mode. Clock frequency 3400 MHz, base frequency 100 MHz (100x34), DDR3 frequency – 1600 MHz (100x16), supply voltage 1.04 V, DDR3 supply voltage – 1.5 V, Hyper Threading – enabled.

FX-8350 BE

Regular mode. Clock frequency 4000 MHz, system bus frequency 200 MHz (200x20), DDR3 frequency - 1866 MHz (200x9.33), core supply voltage 1.28 V, DDR3 supply voltage - 1.5 V, Turbo Core and APM - included.

The processor was overclocked to a frequency of 4700 MHz. To do this, the processor multiplier was raised to 23.5 (200x23.5), the core supply voltage was up to 1.54 V, the DDR3 supply voltage was 1.5 V. The DDR3 frequency was 2133 MHz (200x10.67), Turbo Core and APM were turned off.

FX-6350BE

Regular mode. Clock frequency 3900 MHz, system bus frequency 200 MHz (200x19.5), DDR3 frequency - 1866 MHz (200x9.33), core supply voltage 1.28 V, DDR3 supply voltage - 1.5 V, Turbo Core and APM - included.

The processor was overclocked to a frequency of 4700 MHz. To do this, the processor multiplier was raised to 23.5 (200x23.5), the core supply voltage was up to 1.53 V, the DDR3 supply voltage was 1.5 V. The DDR3 frequency was 2133 MHz (200x10.67), Turbo Core and APM were turned off.

FX-4350BE

Regular mode. Clock frequency 4200 MHz, system bus frequency 200 MHz (200x21), DDR3 frequency - 1866 MHz (200x9.33), core supply voltage 1.33 V, DDR3 supply voltage - 1.5 V, Turbo Core and APM - included.

The processor was overclocked to a frequency of 4700 MHz. To do this, the processor multiplier was raised to 23.5 (200x23.5), the core supply voltage was up to 1.52 V, the DDR3 supply voltage was 1.5 V. The DDR3 frequency was 2133 MHz (200x10.67), Turbo Core and APM were turned off.

A10-6800K

Regular mode. Clock frequency 4100 MHz, system bus frequency 100 MHz (100x41), DDR3 frequency – 2133 MHz, core supply voltage 1.31 V, DDR3 supply voltage – 1.5 V, Turbo Core and APM are included.

A8-6600K

Regular mode. Clock frequency 3900 MHz, system bus frequency 100 MHz (100x39), DDR3 frequency – 1866 MHz, core supply voltage 1.3 V, DDR3 supply voltage – 1.5 V, Turbo Core and APM are included.

The processor was overclocked to a frequency of 4700 MHz. To do this, the processor multiplier was raised to 47 (100x47), the core supply voltage was raised to 1.5 V, the DDR3 supply voltage was 1.5 V. The DDR3 frequency was 2133 MHz, Turbo Core and APM were turned off.

A10-5800K

Regular mode. Clock frequency 3800 MHz, system bus frequency 100 MHz (100x38), DDR3 frequency – 1866 MHz, core supply voltage 1.32 V, DDR3 supply voltage – 1.5 V, Turbo Core and APM are included.

A8-5600K

Regular mode. Clock frequency 3600 MHz, system bus frequency 100 MHz (100x36), DDR3 frequency – 1866 MHz, core supply voltage 1.31 V, DDR3 supply voltage – 1.5 V, Turbo Core and APM are included.

The processor was overclocked to a frequency of 4400 MHz. To do this, the processor multiplier was raised to 44 (100x44), the core supply voltage was up to 1.45 V, the DDR3 supply voltage was 1.5 V. The DDR3 frequency was 2133 MHz, Turbo Core and APM were turned off.

Athlon X4 760K

Regular mode. Clock frequency 3800 MHz, system bus frequency 100 MHz (100x38), DDR3 frequency – 1866 MHz, core supply voltage 1.31 V, DDR3 supply voltage – 1.5 V, Turbo Core and APM are included.

The processor was overclocked to a frequency of 4500 MHz. To do this, the processor multiplier was raised to 45 (100x45), the core supply voltage was up to 1.45 V, the DDR3 supply voltage was 1.5 V. The DDR3 frequency was 2133 MHz, Turbo Core and APM were turned off.

Athlon II X4 740

Regular mode. Clock frequency 3200 MHz, system bus frequency 100 MHz (100x32), DDR3 frequency – 1866 MHz, core supply voltage 1.29 V, DDR3 supply voltage – 1.5 V, Turbo Core and APM are included.

The processor was overclocked to a frequency of 4100 MHz. To do this, the bus frequency was raised to 114 MHz (114x36), the core supply voltage was up to 1.42 V, the DDR3 supply voltage was 1.5 V. The DDR3 frequency was 2127 MHz, Turbo Core was enabled and APM was disabled.

Let's move directly to the tests.

The performance of the new A10-7850K APU was compared to that of its direct competitor, the Core i5-4440, Intel's similarly priced offering based on the latest Haswell design. Along the way, we compared the operating speed of the flagship Kaveri model with the older modification Richland, A10-6800K. Also included in the test results are the performance indicators of the A8-7600 we reviewed earlier: this processor, compared to the A10-7850K, has a lower clock speed and is equipped with a stripped-down graphics core built on the basis of 384 shader processors.

As a result, the set of test equipment took the following form:

• Processors:
  • AMD A10-7850K (Kaveri, 4 cores, 3.7-4.0 GHz, 2x2 MB L2, Radeon R7 Series);
  • AMD A10-6800K (Richland, 4 cores, 4.1-4.4 GHz, 2x2 MB L2, Radeon HD 8670D);
  • AMD A8-7600 (Kaveri, 4 cores, 3.3-3.8 GHz, 2x2 MB L2, Radeon R7 Series);
  • Intel Core i5-4440 (Haswell, 4 cores, 3.1-3.3 GHz, 4x256 KB L2, 6 MB L3, HD Graphics 4600).
  • CPU cooler: Noctua NH-U14S.
• Motherboards:
  • ASRock FM2A88X Extreme6+ (Socket FM2+, AMD A88X);
  • Gigabyte Z87X-UD3H (LGA1150, Intel Z87 Express).
• Memory: 2x8 GB DDR3-2133 SDRAM, 9-11-11-31 (G.Skill F3-2133C9D-16GTX).
• Graphic cards:
  • AMD Radeon HD 7750 (2 GB/128-bit GDDR5, 900/4500 MHz);
  • AMD Radeon R7 250 (2 GB/128-bit GDDR5, 1000/4600 MHz);
  • NVIDIA GeForce GTX 780 Ti (3 GB/384-bit GDDR5, 876-928/7000 MHz).
• Disk subsystem: Crucial m4 256 GB (CT256M4SSD2).
• Power supply: Corsair AX760i (80 Plus Platinum, 760 W).

Testing was performed in the operating room Microsoft system Windows 8.1 Enterprise x64 using the following set of drivers:

• AMD Chipset Drivers 13.12;
• AMD Catalyst Graphics Driver 14.1 beta 1.6;
• Intel Chipset Driver 9.4.0.1027;
• Intel® Iris and HD Graphics Driver 15.33.8.64.3345;
• Intel Management Engine Driver 9.5.0.1345;
• Intel Rapid Storage Technology 12.9.0.1001;
• NVIDIA GeForce 332.21 Driver.

⇡ Performance with discrete graphics

First of all, we test processors on platforms with a high-performance discrete graphics card installed. This configuration allows you to compare x86 performance of different architectures and provides information on how suitable certain CPUs are for working as part of high-performance systems where external video cards are of the highest price range installed in mandatory. In this case, the graphics core of the processors cannot be used, and it is deactivated.

It should be emphasized that in the context of studying the A10-7850K, such testing has a direct practical meaning. AMD has abandoned further development of its FX series processors, so the role of CPU for systems with discrete graphics will gradually shift to Kaveri or their successors.

Futuremark PCMark 8 2.0

Traditionally, we primarily use the PCMark 8 2.0 integral test to measure performance, which simulates various types of typical system load. Three scenarios are considered: Home - normal home use of a PC, Creative - use of a PC for entertainment and for working with multimedia content, and Work - use of a PC for typical office work.

If you read our previous material about Kaveri processors, then the results presented will not come as a surprise to you. Yes, the computing performance of Steamroller cores is low, so the quad-core Kaveri lags far behind the younger quad-core Haswell. This was quite expected, so the fact that the A10-7850K lags not only behind Haswell, but also behind the A10-6800K of the Richland generation can be much more surprising. Obviously, Steamroller's microarchitectural improvements are absolutely not enough to compensate for the lower clock speed of this processor. As a result, the old APU model turns out to be 3-4 percent faster than the new one.

It's funny that, justifying the rather high price set for the A10-7850K, AMD itself refers to the high performance of this processor in PCMark 8. The fact is that AMD means the results with OpenCL acceleration enabled, but in the case of using a discrete video card it it is impossible to use, which leads to the sad picture shown in the diagrams above.

Application Performance

Adobe Photoshop CC tests performance when processing graphic images. The average execution time of a test script is measured, which is a creative reworking of the Retouch Artists Photoshop Speed ​​Test, which involves typical processing of four 24-megapixel digital camera images.

In Autodesk 3ds max 2014 we test the final rendering speed. Measures the time it takes to render a single frame of the standard Space_Flyby scene from the SPEC benchmark using mental ray at 1920x1080 resolution.

Maxon Cinebench R15 measures the performance of photorealistic 3D rendering in the CINEMA 4D animation package. The scene used in the benchmark contains about 2 thousand objects and consists of 300 thousand polygons.

Archiving speed testing is measured in WinRAR 5.0. Here we test the time spent by the archiver to compress a directory with various files with a total volume of 1.7 GB. In this case, the maximum degree of compression is used.

To test the speed of video transcoding into the H.264/AVC format, we use the widely used x264 codec version r2358. To evaluate performance, we use the original 1080p@50fps AVC video file from the x246 FHD Benchmark 1.0.1, which has a bitrate of about 30 Mbit/s.

The gap between the A10-7850K and the similarly priced Core i5-4440 ranges from 30 to 70 percent. In other words, choosing Kaveri family processors for use in systems with a discrete video card makes no sense at all. Even the cheaper A10-6800K, which belongs to the previous generation of APUs, is often able to offer higher scalar computing performance.

Gaming performance

We tested in games using Full HD resolution and high quality settings. Our high-performance GeForce GTX 780 Ti discrete graphics card allows you to see significant differences in processor speed even in this case. Settings used:

• Batman - Arkham Origins: 1920x1080 resolution, Anti-Aliasing = MSAA 4x, Geometry Details = DX11 Enhanced, Dynamic Shadows = DX11 Enhanced, Motion Blur = On, Depth of Field = DX11 Enhanced, Distortion = On, Lens Flares = On, Light Shafts = On, Reflections = On, Ambient Occlusion = DX11 Enhanced, Hardware Accelerated Physx = High.
• Civilization V: Brave New World: resolution 1920x1080, Antialiasing = 4xMSAA, High-Detail Strategic Vie = On, GPU Texture Decode = On, Overlay Detail = High, Shadow Quality = High, Fog of War Quality = High, Terrain Detail Level = High , Terrain Tesselation Level = High, Terrain Shadow Quality = High, Water Quality = High, Texture Quality = High. DirectX 11 version of the game is used.
• F1 2013: resolution 1920x1080, Ultra Quality, 4xAA, DirectX11. The Texas track and the version of the game with support for AVX instructions are used.
• Metro: Last Light: 1920x1080 resolution: DirectX 11, High Quality, Texture Filtering = AF 16X, Motion Blur = Normal, SSAA = On, Tesselation = On, Advanced PhysX = On. During testing, the D6 scene is used.

The results obtained in gaming tests once again confirm everything said above. The computing performance of the A10-7850K is no better than that of the A10-6800K. The Richland generation processor, although based on the Piledriver microarchitecture rather than the Steamroller, has a 10 percent higher clock speed and more aggressive turbo technology. This is quite enough to provide more frames per second in games when using a discrete video card.

Therefore, it is not surprising that the A10-7850K is not comparable in gaming performance to the Core i5-4440. Intel's quad-core processor produces much higher performance in games, so the Socket FM2+ platform is completely unsuitable for high-performance gaming systems. However, this hardly came as a surprise to anyone: we encounter low gaming performance of AMD processors every time we talk about the Bulldozer microarchitecture or its successors.

Steamroller vs Piledriver

The results obtained in computational tests make one wonder how much more advanced the Steamroller microarchitecture is actually than its predecessor. AMD claimed performance gains at constant clock speeds of 15-20 percent. But practical results clearly indicate that the implemented improvements often do not compensate for the 10% reduction in clock speed. Therefore, we decided to see how much faster Kaveri would be than Richland, provided they were clocked at the same frequency.

The following table shows the results of tests conducted with the A10-7850K and A10-6800K processors, the frequency of which was forced to 4.0 GHz.

	Kaveri 4.0 GHz	Richland 4.0 GHz	The Steamroller Advantage
PCMark 8 2.0, Home	2937	2873	+2,2 %
PCMark 8 2.0, Work	2825	2796	+1,0 %
PCMark 8 2.0, Creative	2990	2894	+3,3 %
WinRAR 5.0, seconds	204,8	197,3	-3,7 %
Photoshop CC, seconds	150,3	157,5	+4,8 %
3ds max 2014, seconds	248	339	+36,7 %
x264 (r2358), fps	15,1	12,92	+16,9 %
Cinebench R15	336,8	310,8	+8,4 %
Metro: Last Light, 1920x1080 SSAA HQ	45,8	43,1	+6,3 %
Civilization V, 1920x1080 4xAA HQ	56,3	53,7	+4,8 %
F1 2013, 1920x1080 4xAA UHQ	72,5	75,8	-4,4 %
Batman: Arkham Origins, 1920x1080 4xAA UHQ	75	71,1	+5,5 %

The relationship between the performance of Steamroller and Piledriver appears to be very uneven. At best, the advantage of the new microarchitecture exceeds 35 percent, and at worst, it loses up to 4 percent. The average value of Kaveri's superiority over Richland in performance at the same clock frequency is about 7 percent.

The nature of the results obtained allows us to make an unambiguous conclusion that, first of all, the superiority of Steamroller over Piledriver is revealed on multi-threaded algorithms that use integer instructions. In other words, the division of the general instruction decoder module into a dual-core instruction decoder module, carried out in Steamroller, along with other optimizations, made it possible to increase the efficiency of integer execution devices. Therefore, tasks like 3D rendering or video transcoding received a very noticeable increase in execution speed. In the same case, when applications actively use the still shared block of operations with real numbers or SIMD instructions, the performance gain is noticeably smaller.

The drop in performance observed in some cases seems to be associated with a deterioration in the speed characteristics of the memory controller, which in Kaveri creates a problem O higher call latency than Richland.

Kaveri 4.0 GHz		Richland 4.0 GHz

The reasons for this effect are probably that the Kaveri memory controller is designed to be universal at the architectural level and, in addition to two DDR3 channels, has two additional channels supporting GDDR5 memory. Currently available processor models have this functionality blocked, but its potential presence, as tests show, somewhat slows down the operation of the entire memory subsystem.

⇡ Performance of the integrated graphics core

Gaming Performance

The fact that the traditional computing performance of the A10-7850K is not as high as we would like does not mean anything. You just shouldn’t consider this processor as a possible basis for a system equipped with a discrete video card - it’s completely unsuitable for this. Its strength lies elsewhere: Kaveri can make it possible to do without any video card at all. Its integrated Radeon R7 family graphics core aims to offer gaming-worthy performance.

When it comes to the A10-7850K's integrated graphics capabilities, AMD highlights that it's faster than the graphics cards found in 35 percent of gaming PCs (according to Steam).

Thanks to this, this hybrid processor can provide a fairly high level of graphics performance (more than 30 frames per second in Full HD resolution) not only in most network games, but also in popular single-player games.

However, we decided to start testing the graphics performance of the video core of the A10-7850K processor with the traditional 3DMark Professional Edition 1.2 benchmark. The results of this hybrid processor were compared with the performance of not only the integrated graphics A10-6800K, A8-7600 and Core i5-4440, but also the discrete video accelerators Radeon HD 7750 and Radeon R7 250.

The superiority of the A10-7850K graphics core over all other integrated graphics options is obvious. Thanks to the new GCN 1.1 architecture and the number of shader processors increased to 512, the APU in question is noticeably faster than both the older Richland and Haswell. In fact, the A10-7850K truly offers the highest-performing integrated graphics on the desktop right now.

However, despite this, the A10-7850K still does not reach the performance of the Radeon HD 7750 and Radeon R7 250 graphics cards. The problem of graphics built into the APU has been known for a long time: the insufficiently high bandwidth of the memory subsystem limits its performance. Therefore, the A10-7850K not only noticeably lags behind the Radeon HD 7750 with 512 shader processors, but also loses even to the Radeon R7 250, whose number of shader processors is limited to 384. Discrete video cards are equipped with GDDR5 with a bandwidth of over 70 GB/s, which is used in the platform Socket FM2+ dual-channel DDR3-2133 memory can only offer 34 GB/s bandwidth.

However, let's see what happens in real games.

In the multiplayer shooter Battlefield 4, the integrated graphics of the A10-7850K processor, as promised by AMD, is capable of providing a comfortable number of frames per second in Full HD resolution even at medium quality settings. The superiority over the older Richland is 16-18 percent, and over Haswell it reaches 70 percent. However, those who like to play with high image quality will still have to reduce the resolution somewhere to the 720p level. Unfortunately, the A10-7850K graphics cannot offer performance levels comparable to those of the Radeon HD 7750 and Radeon R7 250: these video cards are 35-40 percent faster.

The popular shooter Crysis 3 has high demands on the performance of the graphics accelerator, and here we are faced with the fact that the A10-7850K cannot produce acceptable performance in Full HD even with minimal image quality. Obviously, owners of gaming systems based on the A10-7850K will have to reduce the resolution in some cases. For example, in the same Crysis 3, 30 frames per second with average image quality can only be obtained in 720p resolution. It should be noted that the Radeon HD 7750 and Radeon R7 250 video cards are free from this problem.

The F1 2013 racing simulator does not have high requirements for graphics performance, therefore, having a platform based on the A10-7850K, it can be played in Full HD even with high image quality. The advantage of senior Kaveri over Richland here is 25-30 percent.

Another graphics-intensive game besides Crysis 3 is the shooter Metro: Last Light. Having a configuration based on the A10-7850K without a discrete video accelerator, you won’t be able to comfortably play it in Full HD resolution even with minimal settings, and with average quality the resolution will have to be lowered to 720p. The hundred-dollar discrete graphics cards Radeon HD 7750 and Radeon R7 250 offer 30-40 percent higher performance and do a good job displaying Metro: Last Light at a resolution of 1920x1080, which is not available on the A10-7850K. In other words, talking about Kaveri as a processor whose built-in graphics engine is capable of providing the ability to install Full HD resolution in any games is completely inappropriate.

In the third-person action adventure Tomb Raider, the graphics performance of the A10-7850K is quite good. At a resolution of 1920x1080, medium image quality can be set, with an advantage over Richland of 7-15 percent. The GT2 graphics core of the Haswell processor lags behind the A10-7850K graphics by an impressive 50-75 percent, making any Intel desktop offerings a poor option for use in gaming systems that rely on graphics cores built into the CPU.

By the way, I would like to draw your attention to one curious point: the A10-7850K demonstrates only slightly higher performance than the A8-7600, despite the fact that the number of shader processors in the older APU is one third more. This is another illustration of the fact that the performance of integrated AMD cores It was not at all limited by their graphic resources, but by memory bandwidth. Therefore, the fact that the Radeon HD 7750 and Radeon R7 250, equipped with 128-bit GDDR5 memory, produce 35-40 percent higher FPS should not be surprising.

AMD specifically emphasizes that integrated systems built on its processors can be a good choice for fans of online free-to-play games. Our tests in the multiplayer combat aviation arcade simulator War Thunder fully confirm this. Owners of configurations with the A10-7850K processor will be able to comfortably play this game in Full HD resolution when choosing High Quality Images. Other AMD processors also look good here. Intel's Haswell with the GT2 graphics core is not able to provide such a level of performance.

At the same time, the most popular multiplayer game World of Tanks places higher demands on graphics subsystem performance. For getting comfortable frequency frames in a resolution of 1920x1080, owners of the A10-7850K will have to reduce the quality to medium. And by the way, the older Kaveri does not provide noticeable advantages over Richland - probably the reason lies in the high processor dependence of this game. However, be that as it may, the A10-7850K hybrid processor is quite a worthy choice for the system of a dedicated tank fan. However, discrete graphics cards with a price of about $100, here, as in other cases, allow you to get 30-35 percent higher performance.

⇡ Effect of memory frequency

The fact that external video cards with a similar graphics core configuration to the A10-7850K have noticeably higher performance, and also that the difference in practical graphics speed between the A10-7850K and A8-7600 reaches only 5-10 percent, clearly indicates the main bottleneck in graphics performance - the speed of the memory subsystem. It is clear that to improve the performance of Kaveri's built-in graphics, faster memory is needed. AMD planned to provide Kaveri with support for faster types of SDRAM than DDR3, but something went wrong, and the final versions of desktop processors, although they switched to the new Socket FM2+ platform, turned out to be compatible only with traditional DDR3 SDRAM.

This means that you can increase the speed of the memory subsystem in Kaveri only by using faster DDR3 modules. Formally, these processors support modules with frequencies up to DDR3-2133, and it was with this memory that we conducted tests. However, as practice has shown, DDR3-2400 can also be installed in systems with A10-7850K. We will talk about what kind of performance gain can be obtained in this case below. And at the same time, let's see how much the A10-7850K will lose in speed if the system with it is equipped not with DDR3-2133, but with slower modules.

The above diagrams hardly require detailed comments. They very clearly indicate how important fast memory is for Kaveri. Switching from DDR3-2133 to DDR3-2400 allows you to get a noticeable increase in performance - about 5 percent. If, in a system with an A10-7850K, you use not DDR3-2133, but, for example, consumer DDR3-1600, then the loss in gaming performance will reach up to 20 percent. In other words, when assembling an inexpensive gaming system with the A10-7850K, you clearly shouldn’t skimp on memory.

⇡ Mantle software interface

Like the Volcanic Islands generation of graphics cards, Kaveri processors, based on the same GCN architecture, support the new Mantle GUI. This name has long excited the minds of owners of new AMD video cards, since the introduction of this interface promises a fairly serious increase in performance in games. The situation is similar with Kaveri: the introduction of Mantle could be another way to more fully unlock the potential of the built-in graphics core. Being well aware of the hardware intricacies of APUs, Mantle offers a specially optimized layer between the game engine and the hardware resources of the computing and graphics cores. Similar low level software interface has long been used in game consoles, and there he shows very good results. That's why widespread implementation Mantle in modern games can increase the appeal of Kaveri for budget gamers.

For systems built on Kaveri processors, Mantle not only implements various low-level optimizations, but also distributes the load created by the graphics driver more evenly across the x86 processor cores. However, it should be borne in mind that Mantle is most effective when gaming performance is limited by the speed of the processor's computing resources, and in configurations using integrated video cores, the situation is usually the opposite: the bottleneck is GPU power and memory bus bandwidth. However, at the time of introducing Kaveri, AMD talked about the possible performance increase that can be obtained through the proprietary API - this increase in real games allegedly reaches 45 percent.

AMD currently has a beta driver version 14.1 ready to support Mantle, and there is a game, Battlefield 4, that can use this software interface. Naturally, we tested how enabling Mantle affects frame rates when a gaming system with integrated graphics, built on the A10-7850K processor, is used to run Battlefield 4.

There is no sign of 45 percent growth here. The increase in frames per second in Battlefield 4 on a system based on the A10-7850K does not exceed a few percent. As you know, activating Mantle gives the maximum increase in systems with a weak processor and a powerful graphics card, and in the case of the A10-7850K, the ratio of performance of computing cores and GPU is the opposite.

At the same time, turning on Mantle in a system based on the A10-7850K also has a noticeable negative effect. You just need to look not at the average, but at the minimum FPS.

The minimum FPS when using Mantle drops noticeably compared to DirectX, that is, AMD's proprietary software interface worsens the smoothness of the game without any prerequisites. Perhaps the problem lies in the fact that the Mantle driver is currently in beta. I would like to believe that AMD will still make some changes to it that can correct the low minimum FPS and further increase the speed of Battlefiled 4 through Mantle on systems built on the company’s APUs.

⇡ Dual Graphics technology

Every time it comes to testing integrated processor graphics, AMD presents its unique trump card - Dual Graphics technology. This technology, promoted since Llano, allows the creation of asymmetric CrossFire configurations using the graphics core built into the processor. Kaveri was not spared either. The integrated video core of the A10-7850K processor, belonging to the Radeon R7 series, can be “paired” with any discrete video card of the same Radeon R7 family installed in PCI slot Express. Previously, it was believed that certain restrictions were imposed on the architecture of such video cards, but in fact there are no limits: any Radeon R7 graphics card with GCN architecture can work with the A10-7850K in Dual Graphics mode.

Moreover, with the release of Kaveri and the release of the Catalyst driver version 14, AMD finally managed to solve the long-standing problem with tiaring(frame breaks) of the output image, which directly affected Dual Graphics configurations. Now Dual Graphics technology works much better and does not cause any unpleasant artifacts, so it can be considered as one of the ways to increase graphics performance.

To get a feel for how Dual Graphics works on a Kaveri-based system, we tested the performance of a combination of the A10-7850K and a Radeon R7 250 graphics card with GDDR5 memory.

Dual Graphics technology promises the maximum performance increase if the performance of the processor graphics and discrete video card is approximately the same. Therefore, AMD calls the Radeon R7 240 the most profitable pair for the A10-7850K. The Radeon R7 250 is more expensive and faster, so the graphics built into the processor does not help it too much: the increase in performance compared to a single video card ranges from 35 to 45 percent.

At the same time, Dual Graphics technology has not lost its limitations, which in many cases call its usefulness into question. As you can see from the results, it does not always give a positive effect. There are a huge number of games that not only do not receive a boost from Dual Graphics, but, on the contrary, begin to produce lower frame rates. This is due both to the lack of necessary driver optimizations and to the fact that in some cases Dual Graphics is not enabled at all at the software level. For example, this technology can only accelerate games running through DirectX 10/11, but not DirectX 9. In other words, the scalability that Dual Graphics can offer is completely unimpressive.

⇡ Heterogeneous performance

Along with gaming applications, the graphics core of Kaveri processors can also be used to speed up calculations by ordinary applications. general purpose. As already mentioned, with the release of Kaveri, AMD is introducing the HSA architecture, which makes the shader clusters of the graphics core independent structural units and thereby simplifies the programming and use of parallel shader processors for calculations. However, the implementation of HSA and the OpenCL 2.0 framework tailored for this architecture is a matter of the distant future, while AMD cannot even offer the driver necessary to enable this technology. But support for OpenCL 1.1 in Kaveri, as in other varieties modern processors with integrated graphics performs excellently, and OpenCL-enabled applications can offload some of their computational work to shader pipelines through this programming interface.

Base software products, capable of harnessing the heterogeneous capabilities of hybrid processors, is growing steadily and today includes an impressive number of popular programs.

The upcoming implementation of HSA should expand this list, however, it is worth noting that not all algorithms can be accelerated through the use of parallel graphics core processors. AMD names image recognition, biometric analysis, augmented reality systems, audio and video encoding, editing, and transcoding tasks, as well as multimedia data search and indexing as applications where the use of hybrid APU capabilities can make practical sense.

Ideally, we wouldn't want to resort to separate performance tests on tasks that use OpenCL. It would be much better if support for heterogeneous processors appeared in common applications, including those we use for routine testing. However, this is not yet the case: hybrid computing is not implemented everywhere, and in the vast majority of cases, OpenCL acceleration is used only to implement some specific functions, and in order to see it, it is necessary to come up with special tests. Therefore, the study of heterogeneous performance became a separate and independent part of our material.

The first and most famous OpenCL performance test is the Luxmark 2.0 benchmark, which is built on the basis of the LuxRender renderer, which uses a physical model of light propagation. To evaluate the heterogeneous performance of processors, we use the scene medium difficulty Sala, and we render it using both graphics and x86 cores.

As you can easily see, connecting the computing resources of graphics cores to the work leads to a serious increase in performance, but does not change too much qualitatively. Intel processors, like AMD APUs, are quite capable of offering similar functionality: their modern modifications support OpenCL 1.1 fully and without any restrictions. Therefore, when using the power of the graphics core, the older Kaveri maintains its lag behind the quad-core Haswell. It is not as catastrophic here as in tasks that rely only on x86 cores, but nevertheless, the A10-7850K does not look like a full-fledged competitor to the Core i5-4440.

Another test that actively uses the resources of graphics cores is SVPMark 3. It measures system performance when working with the SmoothVideo Project package, aimed at improving the smoothness of video playback by adding new frames to the video sequence that contain intermediate positions of objects.

In the diagram you can see the performance of processors both without using the resources of their graphics cores, and after enabling GPU acceleration. Interestingly enough, not only Kaveri, but also Haswell receives a noticeable acceleration. Thus, using OpenCL increases the performance of the A10-7850K by 48 percent, and the Core i5-4440 is accelerated by 33 percent. If we take into account that the Core i5 can offer four x86 cores with higher specific performance, ultimately the heterogeneous performance of the A10-7850K and Core i5-4440 is set at approximately the same level.

One of the most significant achievements of the APU concept, indicating its acceptance by the market software, was the appearance of OpenCL support in the popular WinZIP archiver. Therefore, we could not ignore the measurement of archiving speed in WinZIP 18. For testing purposes, the folder with the unpacked Adobe Photoshop CC distribution package was compressed.

WinZIP well illustrates the thesis that not all algorithms can be accelerated by transferring the load to graphics cores. Although WinZIP formally has OpenCL support, in reality, parallel graphics cores are activated only when compressing files larger than 8 MB. Moreover, there is no particular speed gain from this, so the difference in performance of hybrid processors with OpenCL enabled and disabled is minimal. Accordingly, the Intel quad-core Haswell shows higher performance here in all cases.

Formal support for OpenCL has also appeared in the popular graphics editor Adobe Photoshop CC. True, in fact, the heterogeneous capabilities of the APU are used only in the operation of several filters. Specifically, AMD recommends measuring performance during the Smart Sharpen operation, which is what we did with the 24-megapixel image.

The increase in the speed of the Smart Sharpen filter, which can be obtained when the graphics part of modern processors is involved in the work, is impressive. This operation starts to run 90 percent faster on a system with an A10-7850K, and 45 percent faster on a system with a Core i5-4440. In other words, using the Smart Sharpen filter as an example, we can see the good computing performance of the Kaveri graphics core, but it still does not allow the A10-7850K to outperform the similarly priced quad-core Haswell. And by the way, even with OpenCL acceleration enabled, the older Richland outperforms the A10-7850K due to the higher clock speed of its computing and graphics cores.

Part of the high-resolution video transcoding operations can also be transferred to the GPU. To check what kind of speed increase can be obtained in this case, we used the OpenCL-supporting utility MediaCoder 0.8.28. Performance evaluation is carried out using the original 1080p@50fps file in AVC format from the x246 FHD Benchmark 1.0.1, which has a bitrate of about 30 Mbps.

Here, Kaveri’s performance can be increased quite slightly by using the graphics core for calculations. But Intel's Core i5-4440, which has support for a special technology for Quick Sync video transcoding, increases its speed significantly when the computing resources of the graphics core are enabled. In fact, AMD processors have a similar technology for hardware video encoding - VCE. However, for some reason, none of the common utilities for video transcoding support this engine. Let's hope that with the introduction of a new and more flexible version of this VCE 2 engine in Kaveri, the situation can finally change.

Another example of a popular application that supports OpenCL is a professional editing and editing program Sony video Vegas Pro 12. When performing video rendering in it, the load can be distributed across heterogeneous resources of hybrid processors.

Involving the graphics core of Kaveri processors in the computing work allows you to get a very significant increase in video rendering speed. However, this still does not allow AMD's older APU to catch up with the competing Core i5-4440. Modern Intel processors have much more powerful x86 cores, so even with OpenCL enabled, the A10-7850K seriously falls short of Haswell speed. In addition, Intel processors also support OpenCL and are accelerated when connected to computing work graphics core resources. The speed increase is not as impressive as that of AMD's APU, however, it is clearly not worth writing off.

At AMD's request, we included Futuremark PCMark 8 2.0 in this part of testing. This benchmark can use OpenCL acceleration when simulating normal user activity in common tasks. And then we can get an idea of ​​the performance that hybrid processors will show in the ideal case, when all common applications receive effective support for heterogeneous computing.

It's clear why AMD uses PCMark 8 2.0 results in all of its marketing materials. Thanks to its strong graphics core, the A10-7850K wins in all three scenarios: Home, Creative and Work. This clearly indicates that, subject to proper heterogeneous application optimization, Kaveri processors can be much better than Intel CPUs. In other words, the APU concept being developed by AMD really has great potential, which the introduction of HSA technology should help fully reveal.

⇡ Energy consumption

Power consumption is another traditionally sore point for AMD processors. At least for their productive modifications, which do not have artificially low frequencies to meet the requirements of economical thermal packages. With the release of Kaveri processors, AMD expected to slightly improve the current situation and even slightly reduced the calculated heat dissipation indicators for the older models of the A10 line. Not only the new 28 nm process technology, but also lower clock frequencies should have helped improve energy performance. In other words, the specific productivity per each watt expended should have increased.

How does this work in practice? The following charts show the total consumption (without monitor) of systems using integrated processor graphics, measured at the outlet where the power supply is plugged into. test platform. All energy-saving technologies available in processors are activated. The load on the processor cores is created by the 64-bit version of the LinX 0.6.5 utility with support for the AVX instructions, and the graphics cores are loaded by the Furmark 1.12 utility.

The idle consumption of modern processors is close to zero, so the figures shown in the graph above relate more to the platforms as a whole than to the APUs under study. Therefore, it is not surprising that, regardless of which processor is installed in the Socket FM2+ platform, the consumption is approximately the same. A Haswell-based system consumes less - this is due to the energy-saving technologies available in modern Intel logic sets.

At full load on the x86 cores, it suddenly turns out that the A10-7850K has become even more power-hungry than the previous flagship of the Richland generation, the A10-6800K. The consumption of the new processor is 9 W higher, even though its operating frequencies are noticeably lower. Accordingly, it is impossible to talk about any competition in terms of efficiency with Intel quad-core processors.

With a graphical load the situation is somewhat different. The graphics core of Kaveri processors has noticeably better efficiency than Richland graphics. However, it is necessary to mention one nuance: Kaveri can dynamically control the frequency of its graphics core, and under high load it is automatically reduced. Apparently, in this case we were faced with a consumption limit, since during testing of the A10-7850K and A8-7600, their GPU frequency periodically decreased from the standard 720 MHz to 650 MHz, and at times even to 550 MHz.

Kaveri also demonstrates low consumption with parallel load on all cores at the same time. However, in this test we were faced with intelligent frequency control of not only the GPU, but also the computing cores. As it turns out, under high graphics load, Kaveri not only resets the frequency of its GPU, but also limits the frequency of the processor cores to 3 GHz. As a result, with a simultaneous high load on all the resources of the hybrid processor, its consumption is not too large, but this, naturally, also affects performance.

⇡ Overclocking

The older Kaveri model, A10-7850K, is formally one of the overclocking models with unlocked multipliers - this is clearly indicated by the letter K at the end of the model number. But in this case, this is more a tribute to tradition than a real strength of the new products. The new 28nm SHP (Super High Performance) technology used to manufacture Kaveri does not at all contribute to the emergence of untapped frequency potential in these APUs. And even from a theoretical point of view, new hybrid processors should perform even worse than their predecessors, which also did not have good overclocking capabilities.

This has been confirmed in practice. The maximum frequency at which the A10-7850K, on ​​the one hand, remained stable, and on the other, did not reduce its speed due to exceeding the maximum temperature, was 4.4 GHz. The supply voltage on the processor had to be raised to 1.375 V.

It should be emphasized that overclocking the A10-7850K is not such a trivial procedure due to intelligent algorithms for dynamic frequency control depending on temperature and load. Increasing the processor multiplier above the nominal value seems very easy at first glance and rarely causes stability problems. But when testing under load, it often turns out that the processor, in order to maintain its performance, arbitrarily resets the frequency of individual cores significantly below the values ​​​​set in the motherboard BIOS. Unfortunately, this intelligence cannot be turned off in any way, so when considering overclocking results, among other things, you need to pay special attention to checking the real frequencies of all four processor cores. Such spontaneous “braking” of the processor, unfortunately, does not make it possible to significantly increase its supply voltage.

Along with the traditional processor part, you can also overclock the graphics core built into the APU. By increasing the voltage on the processor north bridge to 1.375 V, we were able to achieve GPU stability by increasing its frequency in the motherboard BIOS to 960 MHz.

However, in reality, overclocking graphics in the A10-7850K makes little practical sense. Firstly, it is not the frequency that limits GPU performance, but the memory bus bandwidth. Secondly, when the GPU frequency increases, it again has to deal with overly intelligent autonomous frequency control. Increasing the frequency of the graphics core leads to the fact that in reality, under 3D load, it begins to systematically drop to lower values, and the gaming performance observed in practice practically does not increase.

In other words, AMD tried to make Kaveri processors with predictable power consumption and heat dissipation, and this required the introduction of real frequency control technologies that do not play well with overclocking. This means that Kaveri is not suitable for overclocking experiments.

⇡ Conclusions

Overall, Kaveri turned out to be a very controversial product, and opinions about it can vary dramatically depending on the angle from which you look at the new product. We already talked about this when we looked at the A8-7600 modification, and we should repeat it now, based on the results of our acquaintance with the A10-7850K.

The new processor is incredibly interesting because it develops the concept of heterogeneous computing and introduces HSA technology, which allows software developers to easily move on to writing algorithms that execute on graphics core computing clusters. It seems that a little more - and AMD will ensure that new applications will work on its processors no worse than on Intel CPUs. For this, Kaveri has all the necessary resources and, most importantly, the enormous theoretical computing power hidden in the graphics core.

However, not all so simple. So far, there are not many even simple applications optimized for OpenCL, and the efficiency of existing implementations of heterogeneous computing leaves much to be desired. In addition, on parallel graphics core computers may be transferred not just any algorithms. As a result, emphasizing that in theory Kaveri-based systems can be very productive, we are forced to state a real and noticeable lag behind the older A10 model we reviewed from the competing quad-core Core i5 in the vast majority of computing tasks. Moreover, this situation is now observed not only in applications executed exclusively on x86 cores, but also where OpenCL support has already been implemented.

Another thing is games. Here AMD is doing very well, even though the speed of the GPU built into the A10-7850K is strictly limited by the memory bus bandwidth. Despite this, configurations built on this processor and using the capabilities of the integrated graphics core can rightfully be considered full-fledged entry-level gaming systems. Most modern games can be run on the A10-7850K in Full HD resolution, and many of them, such as popular network projects, work quite well even with a choice of medium or high image quality. Desktop Haswells cannot offer such gaming performance in principle, at least until Intel decides to transfer older modifications of its GT3/GT3e graphics cores to desktop processor models.

As a result, at the moment the A10-7850K can only be recommended as the basis for inexpensive desktop computers for undemanding gamers. This processor is of little interest to enthusiasts, primarily due to its limited x86 performance. However, if AMD moderates its ambitions and lowers prices, pitting the A10-7850K against the competitor’s dual-core processors rather than quad-core ones, we will be ready to reconsider our position.

« Why is this integration needed? Give us more cores, megahertz and cache!“- the average computer user asks and exclaims. Indeed, when a computer uses a discrete video card, there is no need for integrated graphics. I admit, I lied about the fact that today a central processor without built-in video is harder to find than with it. There are such platforms - LGA2011-v3 for Intel chips and AM3+ for AMD “stones”. In both cases, we are talking about top solutions, and you have to pay for them. Mainstream platforms, such as Intel LGA1151/1150 and AMD FM2+, are universally equipped with processors with integrated graphics. Yes, “built-in” is indispensable in laptops. If only because in 2D mode mobile computers battery life lasts longer. On desktops, integrated video is useful in office builds and so-called HTPCs. Firstly, we save on components. Secondly, we again save on energy consumption. However, in Lately AMD and Intel seriously say that their integrated graphics are for all graphics! Also suitable for gaming. This is what we will check.

We play modern games on the graphics built into the processor

300% increase

For the first time, graphics integrated into the processor (iGPU) appeared in Intel Clarkdale solutions (first generation Core architecture) in 2010. It is integrated into the processor. An important amendment, since the very concept of “embedded video” was formed much earlier. Intel did it back in 1999 with the release of the 810 chipset for Pentium II/III. At Clarkdale, integrated HD Graphics video was implemented as a separate chip located under the heat-distributing cover of the processor. The graphics were produced according to the old 45-nanometer technical process at that time, the main computing part was produced according to 32-nanometer standards. The first Intel solutions in which the HD Graphics unit “settled” along with other components on one chip were Sandy Bridge processors.

Intel Clarkdale - the first processor with integrated graphics

Since then, on-chip graphics for mainstream LGA115* platforms has become the de facto standard. Generations Ivy Bridge, Haswell, Broadwell, Skylake - all have integrated video.

Graphics integrated into the processor appeared 6 years ago

Unlike the computational part, the “embedding” in Intel solutions is progressing noticeably. HD Graphics 3000 in Sandy Bridge K-series desktop processors has 12 execution units. HD Graphics 4000 in Ivy Bridge has 16; HD Graphics 4600 in Haswell has 20, HD Graphics 530 in Skylake has 25. The frequencies of both the GPU itself and RAM are constantly increasing. As a result, the performance of embedded video increased by 3-4 times over four years! But there is also a much more powerful series of “embedded” Iris Pro, which are used in certain Intel processors. 300% interest over four generations is not 5% per year.

Intel Integrated Graphics Performance

In-processor graphics is one segment where Intel has to keep up with AMD. In most cases, the Reds' decisions are faster. There is nothing surprising in this, because AMD develops powerful gaming video cards. Here it is in integrated graphics desktop processors The same architecture and the same developments are used: GCN (Graphics Core Next) and 28 nanometers.

AMD hybrid chips debuted in 2011. The Llano family of chips was the first to combine integrated graphics and computing on a single chip. AMD marketers realized that it would not be possible to compete with Intel on its terms, so they introduced the term APU (Accelerated Processing Unit, processor with a video accelerator), although the idea had been hatched by the Reds since 2006. After Llano, three more generations of “hybrids” came out: Trinity, Richland and Kaveri (Godavari). As I already said, in modern chips the integrated video is architecturally no different from the graphics used in Radeon discrete 3D accelerators. As a result, in 2015-2016 chips, half of the transistor budget is spent on iGPUs.

Modern integrated graphics take up half the usable CPU space

The most interesting thing is that the development of APUs influenced the future... of game consoles. So the PlayStation 4 and Xbox One use an AMD Jaguar chip - eight-core, with graphics based on GCN architecture. Below is a table with characteristics. The Radeon R7 is the most powerful integrated video the Reds have to date. The block is used in AMD A10 hybrid processors. Radeon R7 360 is an entry-level discrete video card, which, according to my recommendations, can be considered a gaming card in 2016. As you can see, the modern “integration” in terms of characteristics is not much inferior to the Low-end adapter. It cannot be said that the graphics of game consoles have outstanding characteristics.

The very appearance of processors with integrated graphics in many cases puts an end to the need to buy an entry-level discrete adapter. However, today integrated video from AMD and Intel is encroaching on the sacred - the gaming segment. For example, in nature there is a quad-core Core i7-6770HQ (2.6/3.5 GHz) processor based on the Skylake architecture. It uses integrated Iris Pro 580 graphics and 128 MB of eDRAM memory as a fourth-level cache. The integrated video has 72 execution units operating at a frequency of 950 MHz. This is more powerful than the Iris Pro 6200 graphics, which uses 48 actuators. As a result, the Iris Pro 580 turns out to be faster than these discrete video cards, like the Radeon R7 360 and GeForce GTX 750, and also in some cases imposes competition on the GeForce GTX 750 Ti and Radeon R7 370. What else will happen when AMD transfers its APUs to the 16-nanometer process technology, and both manufacturers eventually start using them together with integrated graphics memory HBM/HMC.

Intel Skull Canyon - a compact computer with the most powerful integrated graphics

Testing

To test modern integrated graphics, I took four processors: two each from AMD and Intel. All chips are equipped with different iGPUs. So, AMD A8 (plus A10-7700K) hybrids have Radeon R7 video with 384 unified processors. The older series - A10 - has 128 more blocks. The flagship also has a higher frequency. There is also the A6 series - its graphics potential is completely sad, since it uses the “built-in” Radeon R5 with 256 unified processors. I did not consider it for games in Full HD.

AMD A10 and Intel Broadwell processors have the most powerful integrated graphics

Concerning Intel products, then the most popular Skylake Core i3/i5/i7 chips for the LGA1151 platform use the HD Graphics 530 module. As I already said, it contains 25 actuators: 5 more than the HD Graphics 4600 (Haswell), but 23 fewer than the Iris Pro 6200 (Broadwell). The test used the youngest quad-core processor - Core i5-6400.

	AMD A8-7670K	AMD A10-7890K	Intel Core i5-6400 (review)	Intel Core i5-5675C (review)
Technical process	28 nm	28 nm	14 nm	14 nm
Generation	Kaveri (Godavari)	Kaveri (Godavari)	Skylake	Broadwell
Platform	FM2+	FM2+	LGA1151	LGA1150
Number of cores/threads	4/4	4/4	4/4	4/4
Clock frequency	3.6 (3.9) GHz	4.1 (4.3) GHz	2.7 (3.3) GHz	3.1 (3.6) GHz
Level 3 cache	No	No	6 MB	4 MB
Integrated Graphics	Radeon R7, 757 MHz	Radeon R7, 866 MHz	HD Graphics 530, 950 MHz	Iris Pro 6200, 1100 MHz
Memory controller	DDR3-2133, dual channel	DDR3-2133, dual channel	DDR4-2133, DDR3L-1333/1600 dual channel	DDR3-1600, dual channel
TDP level	95 W	95 W	65 W	65 W
Price	7000 rub.	11,500 rub.	13,000 rub.	20,000 rub.
Buy

Below are the configurations of all test benches. When it comes to integrated video performance, it is necessary to pay due attention to the choice of RAM, since it also determines how many FPS the integrated graphics will show in the end. In my case, DDR3/DDR4 kits operating at an effective frequency of 2400 MHz were used.

Test benches
№1:	№2:	№3:	№4:
Processors: AMD A8-7670K, AMD A10-7890K;	Processor: Intel Core i5-6400;	Processor: Intel Core i5-5675C;	Processor: AMD FX-4300;
			Motherboard: ASUS 970 PRO GAMING/AURA;
		RAM: DDR3-2400 (11-13-13-35), 2x 8 GB.	Video card: NVIDIA GeForce GTX 750 Ti;
			RAM: DDR3-1866 (11-13-13-35), 2x 8 GB.
Motherboard: ASUS CROSSBLADE Ranger;	Motherboard: ASUS Z170 PRO GAMING;	Motherboard: ASRock Z97 Fatal1ty Performance;
RAM: DDR3-2400 (11-13-13-35), 2x 8 GB.	RAM: DDR4-2400 (14-14-14-36), 2x 8 GB.	RAM: DDR3-2400 (11-13-13-35), 2x 8 GB.
Motherboard: ASUS CROSSBLADE Ranger;	Motherboard: ASUS Z170 PRO GAMING;
RAM: DDR3-2400 (11-13-13-35), 2x 8 GB.	RAM: DDR4-2400 (14-14-14-36), 2x 8 GB.
Motherboard: ASUS CROSSBLADE Ranger;
RAM: DDR3-2400 (11-13-13-35), 2x 8 GB.
Operating system: Windows 10 Pro x64;
Peripherals: LG 31MU97 monitor;
AMD Driver: 16.4.1 Hotfix;
Intel Driver: 15.40.64.4404;
NVIDIA Driver: 364.72.

RAM support for AMD Kaveri processors

Such sets were chosen for a reason. According to official data, the built-in memory controller of Kaveri processors works with DDR3-2133 memory, however, motherboards based on the A88X chipset (due to an additional divider) also support DDR3-2400. Intel chips, coupled with the flagship Z170/Z97 Express logic, also interact with faster memory; there are noticeably more presets in the BIOS. As for the test bench, for the LGA1151 platform we used a dual-channel Kingston Savage HX428C14SB2K2/16 kit, which overclocked to 3000 MHz without any problems. Other systems used ADATA AX3U2400W8G11-DGV memory.

Selecting RAM

A little experiment. In the case of Core i3/i5/i7 processors for the LGA1151 platform, using faster memory to accelerate graphics is not always rational. For example, for the Core i5-6400 (HD Graphics 530), changing the DDR4-2400 MHz kit to DDR4-3000 in Bioshock Infinite gave only 1.3 FPS. That is, with the graphics quality settings I set, performance was limited precisely by the graphics subsystem.

Dependence of the performance of the integrated graphics of an Intel processor on the frequency of RAM

The situation looks better when using AMD hybrid processors. Increasing the speed of RAM gives a more impressive increase in FPS; in the frequency delta of 1866-2400 MHz we are dealing with an increase of 2-4 frames per second. I think that using RAM with an effective frequency of 2400 MHz in all test benches is a rational solution. And closer to reality.

Dependence of the performance of the integrated graphics of an AMD processor on the frequency of RAM

We will judge the performance of integrated graphics based on the results of thirteen gaming applications. I roughly divided them into four categories. The first includes popular but undemanding PC hits. Millions play them. Therefore, such games (“tanks”, Word of Warcraft, League of Legends, Minecraft - here) have no right to be demanding. We can expect a comfortable FPS level at high graphics quality settings in Full HD resolution. The remaining categories were simply divided into three time periods: the 2013/14, 2015 and 2016 games.

Integrated graphics performance depends on RAM frequency

The quality of graphics was selected individually for each program. For undemanding games, these are mainly high settings. For other applications (except Bioshock Infinite, Battlefield 4 and DiRT Rally) - low quality graphics. Still, we will test the built-in graphics in Full HD resolution. Screenshots describing all graphics quality settings are located in the screenshot of the same name. We will consider 25 fps to be playable.

Undemanding games	2013/14 Games	Games of 2015	Games of 2016
Dota 2 - high;	Bioshock Infinite - average;	Fallout 4 - low;	Rise of the Tomb Raider - low;
Diablo III - high;	Battlefield 4 - average;	GTA V - standard;	Need for Speed- low;
StarCraft II - high.	Far Cry 4 - low.		XCOM 2 - low.
		DiRT Rally - high.
Diablo III - high;	Battlefield 4 - average;	GTA V - standard;
StarCraft II - high.	Far Cry 4 - low.	"The Witcher 3: Wild Hunt" - low;
		DiRT Rally - high.
Diablo III - high;	Battlefield 4 - average;
StarCraft II - high.	Far Cry 4 - low.
Diablo III - high;
StarCraft II - high.

HD

The main purpose of testing is to study the performance of integrated processor graphics in Full HD resolution, but first, let's warm up on a lower HD. The iGPU Radeon R7 (for both A8 and A10) and Iris Pro 6200 felt quite comfortable in such conditions. But the HD Graphics 530 with its 25 actuators in some cases produced a completely unplayable picture. Specifically: in five games out of thirteen, since in Rise of the Tomb Raider, Far Cry 4, The Witcher 3: Wild Hunt, Need for Speed ​​and XCOM 2 there is no place to reduce the quality of graphics. It is obvious that in Full HD the integrated video of the Skylake chip is a complete failure.

HD Graphics 530 already merges in 720p resolution

The Radeon R7 graphics used in the A8-7670K failed in three games, the Iris Pro 6200 failed in two, and the built-in A10-7890K failed in one.

Test results in 1280x720 pixels resolution

Interestingly, there are games in which the integrated video of the Core i5-5675C seriously outperforms the Radeon R7. For example, in Diablo III, StarCraft II, Battlefield 4 and GTA V. Low resolution affects not only the presence of 48 actuators, but also processor dependence. And also the presence of a fourth level cache. At the same time, the A10-7890K outperformed its opponent in the more demanding Rise of the Tomb Raider, Far Cry 4, The Witcher 3 and DiRT Rally. The GCN architecture works well in modern (and not so modern) hits.