AMD A10-5800K and A8-5600K: the second desktop APU! - BeHardware
>> Processors

Written by Guillaume Louel

Published on January 8, 2013

URL: http://www.behardware.com/art/lire/878/


Page 1

Introduction

A year after the launch of its first APUs, the Llanos, AMD is now bringing out the second generation of processors that combine a ‘fast’ graphics part with traditional x86 cores. Launched at a relatively high price, with modest clocks and a K10.5 architecture that had served its time, the Llanos didn’t really convince us in their first desktop incarnation.


Everything is new on the second generation however. Although the APUs are still manufactured on a 32nm process by GlobalFoundries, AMD has otherwise wiped the slate clean. As with the mobile version of these chips, launched in June, the K10.5 architecture used for the Phenom IIs has been replaced with Piledriver, a second version of the Bulldozer architecture. This isn’t however the only difference as the graphics part has also been modified.

Of course, the competition has also moved on in the last year and Intel has now brought out Ivy Bridge in place of Sandy Bridge. The main innovation on the Intel side – apart from the engraving / energy consumption aspect – concerns the more powerful graphics part. Will the new Virgo APUs give a decisive advantage over the Llanos? How will these new chips do against the competition from Intel? And how does the integrated graphics part perform compared to the entry level Nvidia and AMD solutions? All questions that we’re going to try to answer in this article!


Page 2
Llano, Trinity, Virgo: recap

The first generation of Llano APU represented the concretisation (at long last) of the Fusion strategy that AMD had talked about when buying the graphics manufacturer, ATI: integration of the GPU with the processor. The vision was delayed several times for lots of different reasons and finally Intel actually got there first at the beginning of 2011 with Sandy Bridge, the first chips in this category to include a graphics core on their die.

Engraved at 32nm by GlobalFoundries, Llano was based on the integration of x86 cores using the K10.5 architecture, namely the same architecture used for the Phenom IIs and the Athlon IIs. Two (major) cache concessions were however made: the L2 cache was limited to 1 MB per core and there was no L3 cache at all.


On the GPU side, Llano was based on the ‘Redwood’ GPU, more commonly known as the GPU used for the Radeon HD 5570s. It used AMD’s VLIW5 architecture. There was a Turbo mode for the CPU part on these chips, but not for the GPU part.

Trinity Desktop = Virgo!

In fact, behind codename Virgo is the desktop version of Trinity (see our review of the mobile version). As we said in that report, it looks as if AMD came up with a complete redesign for Trinity because Llano and Trinity/Virgo have nothing in common.


Or almost nothing. There are a few things. Llano and Trinity are for example both engraved at 32nm by GlobalFoundries. Also, neither has an L3 cache. The Virgo die measures 246 mm2 for a total of 1.3 billion transistors.

On the CPU side, the K10.5 x86 cores have been replaced with Piledriver modules. To recap, Piledriver is version 2 of the Bulldozer architecture, launched on the desktop side with the AMD FXs last year. The basic concept of Bulldozer and Trinity is the fusion of two cores in a single module so as to share a certain number of resources. Thus the module has two processing units for integers and a shared unit that is used for floating point calculations. And while each core has a level 1 cache, they share a level 2 cache (2 MB per module compared to 1 MB per core for Llano), as well as other resources (instruction decoders, prefetcher and so on).


AMD announced lots of changes to Bulldozer with its Piledriver architecture, or more exactly lots of little corrections pretty much everywhere. The details released by AMD are however relatively high level. First of all with respect to the instruction set, Piledriver supports FMA3s (Fused Multiply/Add on 3 operands, a = a * b +c ) in addition to FMA4 (a = b * c + d) that was already supported previously (Intel will use FMA3 as of next year), as well as the 16/32 bit floating point conversion instructions introduced by Intel in Ivy Bridge.

For the rest, the changes are mostly small touches at all levels, with the branch prediction mechanisms and schedulers announced as more efficient and gains have been announced for divisions. Overall, it’s quite difficult to judge the impact that these changes will have on performance and the absence of a level 3 cache stops us from carrying out a comparison at equal clocks with the AMD FX ‘Bulldozer’ (we’ll have to wait for the AMD FX ‘Piledriver' Desktops, the Visheras). To recap AMD announced a gain of less than 10% between Bulldozer and Piledriver. It's on the energy consumption and energy efficiency side that AMD is claiming the biggest gains, with a 10% to 20% energy economy improvement over the Bulldozer architecture.


On the GPU side it’s also all new as AMD has gone from a VLIW5 architecture to a VLIW4 one, as used on the Cayman GPUs (Radeon HD 6900) last year. The number of processing units depends on the model. Thus the A10-5800L, that we have tested today has 384 shader units, the A8-5600K has 256 and the A6 and A4 have 192 and 128 units respectively.


The final noteworthy change is the Turbo. AMD now has both a CPU and GPU Turbo, both of which can be used according to requirements.


Where AMD used these improvements to reduce the TDP on the mobile versions in comparison to Llano, it has made a different choice for the Virgo desktop version and has retained identical TDPs (100W and 65W depending on the model, both models tested have a TDP of 100W) but increased clocks. Thus the A10-5800K has a base clock of 3.8 GHz and a Turbo clock of 4.2 GHz (3.6/3.9 for the A8), whereas the first Llanos started at 2.9 and 2.6 GHz respectively. The A8-3870K, the current Llano high-end, is only clocked at 3 GHz. On the graphics side, the clock hasn’t however changed. We’re still at 800 MHz for the high-end model.


Page 3
AMD A10-5800K and A8-5600K, Gigabyte F2A85X-UP4

Above and beyond the changes inside, the new AMD APUs also bring in a new socket. Out with FM1 and in with a new, incompatible socket, FM2. AMD usually tends to make its sockets last so this development is unusual, especially as visually the two are pretty much identical, except for the placement of the socket notches. Note that AMD has told us that it is committed to using FM2 for at least the next generation.


On the left an FM1 processor and on the right an FM2. Note the socket notches in the middle.

One of the advantages obtained through this socket change concerns multi-screen support. As AMD also supports Eyefinity technology, on FM2 boards it will be possible to use three outs at the same time, an option that you don’t get with the competition (in practice, though in theory this is possible). Thus you will be able to combine DVI, DisplayPort and HDMI outs (you need two DP outs with Intel Ivy Bridge, which won't be the case with Haswell, the next generation).

Gigabyte F2A85X-UP4

Note also the arrival of a new chipset, the A85X to be added to the A75 and A55 launched with the FM1 platform. On the FM2 socket, there will be three chipsets sharing the following features:


Apart from the additional two Serial ATA 6 Gb/s ports, the main difference comes with support for two PCI Express (2.0) graphics ports in x8 mode for Crossfire usage. This addition seems to fly in the face of Dual Graphics, which is offered on the platform!


AMD supplied us with the Gigabyte F2A85X-UP4 for our test, a board which is relatively well equipped and which offers four video outs on its back panel (DP, Dual-DVI, HDMI and VGA) as well as more standard features such as four USB 3.0s (2 chipsets, 2 EtronTech EJ168As), 2 USB 2.0s and a Gigabit Ethernet. On the audio side, there are six assignable jacks and an optical S/PDIF running on a Realtek ALC 892 controller.


There are three physical PCI Express x16 slots, the first two connected to the processor (16/0 or 8/8) and the last to the chipset in x4.


Note finally there is a UEFI BIOS with its 3D mode, something that is now standard on Gigabyte boards!


Page 4
Dual Graphics, impact of memory

One of the particularities of the Llano platform is that it allowed you to combine a graphics card with the GPU on the APU to give a Crossfire mode solution. This is known as Dual Graphics. There were however certain restrictions. You had to use GPUs that were similar in terms of architecture and performance to obtain the best results.

Thus with Llano, you could pair high end models up with the HD 6670 or 6570 based on the VLIW5 architecture that was also used by the Llano GPU. As AMD hasn‘t launched any Radeon HD 7000s in this price range, these are still the cards that will be used, in spite of the move over to VLIW4 architecture for the GPU part of Virgo.

Installing a Dual Graphics system at the time of launch of the first generation of APU turned out to be a real headache and required all sorts of operations (the procedure is detailed here!). Thankfully things are easier now.

Firstly, on the BIOS side, motherboards now automatically detect the presence of an additional graphics card and our Gigabyte board was able to display the BIOS on the additional graphics card if the top screen had been plugged in. This may seem basic but to recap, on launch of Llano the video signal of the BIOS had to go via the IGP, which was especially problematic given that the screen had to be plugged into the graphics card to give the best level of performance in Dual Graphics mode!


In practice, you will now simply be able to add your graphics card, plug in the screen... and make a pit stop in the BIOS. In effect, with our motherboard, adding a secondary graphics card deactivated the APU GPU and the Dual Graphics option then no longer appeared in the AMD control panel. We therefore had to force the presence of the IGP and once this was done, a single click was all that was needed to activate Dual Graphics.

The impact of memory on graphics performance

We looked at the impact of the memory clock on games. As you know, the memory bandwidth is particularly important when it comes to 3D games and it was interesting to see if the increase in clock would allow us to obtain a clear gain on the APUs. Here we measured relative performance at 1920 x 1080 with our A10-5800K, with the memory clocked at 1600 MHz (9-9-9) and 2133 MHz (9-11-9). Note that the motherboards also support a clock of 1866 MHz even if we couldn't get this clock working during our tests.


The gains were variable but significant. Thus, although Civilization V gave a performance gain of just 7%, F1 2011, Battlefield 3 and Batman were up 17%. Crysis 2 showed an even bigger gain, though its original score was very low. Using faster memory will therefore have an impact on these platforms, though this isn’t necessarily in keeping with the price positioning of the APU offer!


Page 5
Memory bandwidth, the platforms

Mono-threaded bandwidth

We started with the memory bandwidth accessible via a single thread. The readings were taken with the Aida64 memory test.


Note a small but significant gain for the new arrivals in this bench, taking them above a bandwidth of 10 GB/s both in reads and writes. The Intel offer leads but the theoretical gain on the AMD side is appreciable.

Multithreaded bandwidth

Lets move on to the multithreaded benchmark included in Rightmark.


The read performance of the Virgos is pretty good though writes aren’t as impressive, something we’ve already seen on the mobile version of these APUs! It remains to be seen how all this translates in our practical tests.

Test platforms

As our tests cover a multitude of platforms on the processor side, we used the following mobos to carry them out:
  • Socket AM3+: Asus M5A97 EVO (Phenom II 955)
  • Socket FM1: Asrock A74 Pro4 (AMD A8-3870K)
  • Socket FM2: Gigabyte F2A85X-UP4 (AMD A10-5800K, A8-5600K)
  • LGA 1155: Asus P8Z77-V Pro (Intel Core i3 3220, 3225, Core i5 3330)


The rest of the configuration was the same on all the platforms:
  • BeQuiet BQT E6-400W power supply
  • 4 x 4 GB ofDDR3 1600 (CAS 9) memory
  • Windows 7 64 bit


So as to test the potential of the processors to use a graphics card, we added, where indicated a Radeon HD 7970.

Finally, we used the following models to compare the graphics performance of these APUs to the AMD and Nvidia entry level graphics card offer. As we often complain, there's a very big variation in the specs of entry level cards as both AMD and Nvidia pretty much allow their partners free reign. The models that we have chosen here are among the better ones, with close to official specs. Variations do exist however (particularly on the Nvidia side). We have therefore given the clocks of our models:
  • GeForce GT 620 (GPU 700 MHz, DDR3 600 MHz)
  • GeForce GT 630 (GPU 810 MHz, DDR3 800 MHz)
  • GeForce GT 640 (GPU 900 MHz, DDR3 891 MHz)
  • Radeon HD 6450 (GPU 625 MHz, DDR3 667 MHz)
  • Radeon HD 6670 DDR3 (GPU 800 MHz, DDR3 800 MHz)
  • Radeon HD 6670 GDDR5 (GPU 800 MHz, GDDR5 1000 MHz)
  • Radeon HD 7750 (GPU 820 MHz, GDDR5 1150 MHz)


Let’s move on to the tests!


Page 6
CPU: Cinebench, x264, Visual Studio

We began our performance tests by looking at processor-side performance.

Cinebench R11.5


We used Cinebench version R11.5 to measure 3D rendering performance. To recap, the software uses the Cinema 4D rendering engine.


[ Monothread ]  [ Multithread ]

The increase in clock plays a big part in the increase in performance with a single thread. Here the Virgos do pretty well even though we’re still some way down on what you get out of a Core i3 on a single thread. Once we moved on to multithreaded performance things changed somewhat. The Core i3 and its two cores were outdone by the A10 and its two modules/four cores. Llano remains in the lead nevertheless. The most expensive Core i5 uses its four cores to lay down the law however.
Staxrip - x264 b2197


Moving frontend, we used Staxrip to transcode a scene from Avatar via x264 in build 2216. We carried out a medium type 2 pass encoding on a 720p source, re-encoded at a bitrate of 6 Mbits/s. To recap, the second pass is the one that benefits most from multithreading.


[ 1st pass ]  [ 2nd pass ]

The A10-5800K finally succeeds in knocking the old Phenom II 955 off its perch in this test both on 1st and 2nd passes. The A10 outdoes the Core i3.

Visual Studio 2012


We opted for the 2012 version of Visual Studio. We compiled the latest version (1.8.1) of the source code of the 3D Obre engine (examples included). Parallel compilation was activated for each project in VS.


In what is a relatively challenging test, the A8-5600K can’t keep pace with Llano. The A10, Llano and Core i3 finish very close to each other though the Core i5 remains in the lead.


Page 7
CPU: 7-Zip, Bibble

7-Zip 9.20


We used version 9.20 of 7-Zip to compress a large volume of files using the LZMA2 algorithm.


The Virgo APUs do pretty well here, both outdoing the Intel Core i3. The A10 is even pretty close to the Core i5.

Bibble 5.2.3


Let’s bring our purely processor tests to an end with the photo processing software, Bibble. We processed a lot of 48 RAW photos and exported them as JPEGs.


Bibble is a little less of a happy hunting ground for the new AMD APUs and while the new arrivals do better than Llano, the modest dual core Core i3 outperforms them.

Let's move on to the CPU games tests.


Page 8
CPU: F1 2011, Crysis 2, Arma II

For all tests on this page, we added a Radeon HD 7970 to each of our platforms. The goal of these tests is to evaluate the potential of the processors to pilot a discrete graphics card. All the tests were carried out at 1920x1080.

F1 2011


Here we pushed the settings to max in DirectX 11.


F1 2011 is particularly sensitive to the model of processor used. The Core i5 shines here, standing out from the rest, and is followed by the Core i3. The AMD APUs don’t do particularly well in this test, with the A10 just about on a par with Llano.

Crysis 2


We used Ultra mode with patch 1.9. The DirectX 11/Tesselation pack was installed and used.


In Crysis 2 the Core i3 and the different AMD processors are all more or less on a par. The Core i5 significantly benefits from its two extra cores and is way out front.


Arma II


All options were pushed to a maximum, including viewing distance.


The AMD offer couldn't even stand up to the Core i3 here, with the Core i5 marking out even more of a lead!

We now move on to the OpenCL tests!


Page 9
OpenCL: DxO Optics Pro, WinZIP, Luxmark

AMD has been bigging up its OpenCL support for some time. Now Intel also has a driver for the Ivy Bridges. We wanted to look at what impact it has in three accelerated applications. There are still relatively few applications that use OpenCL, though the number is increasing. Transparent support of multiple platforms isn’t however yet entirely there…

DxO Optics Pro 7.5.4

We used version 7.5.4 of this photo processing application to carry out RAW to JPEG exports on a series of 48 files.


[ CPU ]  [ OpenCL ]

DxO Optics Pro only allows activation of OpenCL by default on a platform where the CPU is slower than the OpenCL acceleration (in theory…). A benchmark was run at launch of the application to measure the respective performance levels.

Note that while the Virgo APUs show a marked gain over Llano between the CPU and the OpenCL version, the CPU version is in the lead! The bench kicked in. We advise the bench designer to give advanced users an option here as this would make things simpler!

We looked at what an additional graphics card added to the mix. On the graphs below we give the result of the A10 for comparison. A10 + Graphics card indicates that the IGP was deactivated. A10 DG is where Dual Graphics has been turned on with the Virgo IGP and the Graphics Card then functioning in Crossfire.


The gains in this test were very variable. A few remarks:
- The HD 7750 was less effective than the HD 6670 GDDR5
- The GT 640 did best in this test
- The HD 6450 and GT 620 were not taken into account because of the built-in selection system

WinZIP 16.5

As highlighted by AMD at the launch of its Radeon HD 7000s, version 16.5 of WinZIP includes OpenCL support. We compressed the same set of files as that used in our 7-Zip test.


[ CPU ]  [ OpenCL ]

As we said before, having additional cores doesn’t change much in WinZIP and the Core i5 is on a par with the Core i3 here. Note a net gain on the CPU side for the Virgos compared to Llano. With OpenCL on, there are significant gains for the APUs, which move in front of the Intel CPUs. Note that here, the publisher of WinZIP seems to have an exclusive partnership with AMD: it’s impossible to activate OpenCL with an Intel processor, or as you’ll see further on, with an Nvidia graphics card!


The model of card used hardly makes any impact on performance! We hope that in the next version of this test, Corel will have the good taste to allow OpenCL on other cards!

LuxMark 2.0

The last of our OpenCL tests was LuxMark, a benchmark that allows you to compare the OpenCL CPU (the rendering of the OpenCL kernels carried out on the processor), GPU, and mixed CPU + GPU modes.


[ OpenCL CPU ]  [ OpenCL GPU ]  [ OpenCL CPU+GPU ]

The first bit of good news is that, unlike in our previous test, the Intel OpenCL driver no longer crashes when you launch Luxmark. Apart from with the A10, CPU performance is systematically above GPU performance. The cumulated version gives gains with a notably lower yield on the APUs, but this is still enough to allow the A10-5800K to position itself in front of the Core i5-3330.

Note for information that both the AMD and Intel OpenCL x86 drivers can be used on different brand processors. The AMD driver is the most efficient and when used on a Core i3 takes performance above 200 in CPU mode as against just 169 with the Intel driver! There is therefore plenty of room for optimising the Intel driver.


[ GPU ]  [ CPU + GPU ]

Adding a GPU to the A10 produces some very interesting results here! Firstly, and this is sufficiently unusual to be highlighted, Dual Graphics works at full speed here, both for the DDR3 and GDDR5 versions (the second has the advantage). The Radeon HD 7750 nevertheless still leads the field. Performance on the Nvidia cards are less impressive here and the difference between the GT630 and 640 almost nonexistent.

Note finally that adding processor performance reduces performance in some cases (HD 7750 and Dual Graphics HD 6670 GDDR5).

Let’s now move on to the games tests!


Page 10
GPU: F1 2011, Civilization V, Battlefield 3

We measured performance in six modern games on our platforms. We carried out these tests at 1920 x 1080 and 1280 x 720.

F1 2011

We started with the Codemasters game, F1 2011. We tested the game both in DirectX 9 and DirectX 11 modes. We used the Medium/Intermediary setting.


[ 1080p DirectX 9 ]  [ 720p DirectX 9 ]
[ 1080p DirectX 11 ]  [ 720p DirectX 11 ]

The Virgo APUs dominate when compared with the other integrated models. In all other cases you have to add a GT 640 or a Radeon HD 7750 to do better. Note that while Dual Graphics is efficient with the GDDR5 version of the 6670, with the same card equipped with DDR3 performances actually go down! The Intel offer trails by quite some distance on this title, even with the HD 4000, which barely outdoes the HD 6450 and GT 620.

Civilization V

We measured the graphics performance on a scene loaded at the end of a game. All options were set to minimum.


[ 1080p ]  [ 720p ]

Dual Graphics doesn’t contribute much here and the two HD 6670s on their own are close to the A10-5800K, which is also close to the GT 640. The HD 7750 shows its muscle in spite of the modest graphics.

Battlefield 3

We used the lowest setting in this title.


[ 1080p ]  [ 720p ]

The very demanding Battlefield 3 has all our APUs on their knees. Note, Dual Graphics makes a real difference here, outdoing the HD 7750 solution at 1920 x 1080


Page 11
GPU: Batman Arkham City, Crysis 2, Diablo 3

Batman Arkham City

We used the built-in benchmark with the options set at minimum.


[ 1080p ]  [ 720p ]

Dual Graphics contributes varying results depending on the titles. It does nothing here. The HD 7750 dominates easily and the GT 630 and 640 outperform the APUs.

Crysis 2

We measured performance in a scene in "High" mode… which is the lowest graphics mode available on this title! Marketing-speak again…


[ 1080p ]  [ 720p ]

Crysis 2 is particularly hard on all our cards at 1920 x 1080. Only the HD 7750 really allows you to play. The lack of smoothness is too much of an issue on all the other models. Note that Dual Graphics significantly slows down performance.

Diablo 3

Our last games test is with Diablo 3. The graphics engine is relatively modest and was designed for laptop PC platforms. Here we used the minimum graphics mode (the Low FX option was also ticked) and we measured performance during a movement on a fixed part of the map. Because of the random nature of the fight scenes, it isn't possible to measure the heaviest load scenes precisely. In practice, we noted 20 to 30% dips in performance on the scores given.


[ 1080p ]  [ 720p ]

Diablo 3 is a little more forgiving, given the low level of graphics used. The game can be played on the APUs and Dual Graphics does have an impact, including with the HD 6670 DDR3 solution. Alone, the A10-A5800K is a long way behind the Radeon HD 6670 GDDR5 and GeForce GT 640.


Page 12
GPU average, Energy consumption, overclocking

GPU average

While each individual score is of course of interest, we have calculated an average graphics performance for each of the solutions.


The first thing to note is that we can clearly see that cards such as the AMD Radeon HD 6450 or the Nvidia GeForce GT 620 are no longer worth buying, even when compared with the Intel HD 4000. However, you can still pay between €40 and €60 for these cards and they’re still selling like hotcakes!

The Radeon HD 7660D that you find within the A10-A5800K gives similar levels of performance to the Radeon HD 6670 DDR3 or the GeForce GT 630, which come in at between €60 and €70. To recap, coupled with DDR3-2133 instead of DDR3-1600, it would give another 15% or so.

When you combine the 6670 with GDDR5, performance levels take off and with a Radeon HD 7750 they’re doubled compared to the APU alone. On average, Dual Graphics gives a gain of 14.3% with a 6670 DDR3 and 18.2% with 6670 GDDR5: as we have seen, this is very variable depending on the game used and the impact can be negative.
Energy Consumption

We measured the energy consumption of our different configurations in multiple scenarios:
- At idle
- Prime95 (high CPU load)
- Furmark (high GPU load)
- Prime95 + Furmark (high CPU+GPU load)
- F1 2011 (typical game load)

Energy consumption was taken at the 220V socket. Here are the results:

[ Idle / Prime95 ]  [ Furmark / Prime95+Furmark ]  [ F1 2011 ]

There are several important lessons to be drawn. Firstly at idle, the Virgo APUs are on a par with the Intel Ivy Bridges, which is a pretty good thing. In CPU load, the APUs consume significantly more than a Core i3 3220, which was almost on a par with them in the tests. Processor load energy consumption is even higher than for the Core i5 3330, which was a long way ahead of the APUs in the tests. In terms of performance per Watt, a reading AMD has long cherished, the APUs (and indeed Piledriver in general) don’t do all that well.

Graphics energy consumption in Furmark is also higher than with the Intel solutions, but performance is also higher and may therefore justify the difference. On our last graph, typical energy consumption in F1 2011 at 1920 x 1080, once again the energy consumption of the APUs isn’t all that good. We have added, for information, the energy consumption of a Core i3/HD 6670 GDDR5 pairing in this title. While more expensive than the A10-5800K, this combination does give better performance for 20 Watts less.


The energy consumption of entry level graphics cards is quite high given the performance obtained.

Overclocking

The K in the name of the two processors that we have tested today indicates that their multiplier is unlocked. Good news!


Our first tests weren’t very fruitful however, largely because of the base voltage used by our A10-5800K: 1.416V. Under these conditions we obtained a clock of 4.4 GHz with air cooling by pushing voltage up to 1.46V, a score it will probably be possible to improve with a bit of tinkering.

Note also another positive point: our motherboard (and at least its Asus equivalent) allows you to import XMP timings automatically. While AMD trumpets its own AMP timings, in practice our card managed to load an XMP profile (correctly). This is good news and we hope that with DDR4, JEDEC will finally standardise this sort of profile!


Page 13
Conclusion

Arriving a year after the Llano APUs, the 2012 APUs, here represented by the A10-5800K and A8-5600K, correct a certain number of the problems that existed on the first generation. We note first of all, with satisfaction, an increase in performance, both on the processor and the graphics side. In both cases, the gain, while not extraordinary, is of benefit.

Firstly, looking at CPU performance, moving from the K10.5 architecture to Piledriver has been achieved relatively smoothly: the higher clock and bigger memory cache allow the new Virgo APUs to outdo Llano, something we would not necessarily have put money on after the launch of the FXs and the Bulldozer architecture. Compared to the Intel offer, if we look simply at x86 performance, the Virgo APUs are battling it out with the Core i3s, with an advantage in some multithreaded applications.

On the graphics side, the developments are not spectacular. Yes, the new APUs do outperform Llano and are significantly up on the current Intel offer (whether the HD 2500 or the HD 4000), but overall the level of performance is pretty low. Nevertheless, the budget end of the graphics card market has now been left behind.

With aggressive pricing of around €110 for the A8-5600K APU and €130 for the A10-5800K, AMD seems to have found a niche within which to offer its product, making it competitive with the Intel offer, the Core i3 3220 which is currently on sale for €130.


Of course, to achieve this level of performance AMD does use four cores instead of two. This wouldn’t matter so much except that it does entail a significantly higher energy consumption in load, even if in idle it stays low. This brings us to the crucial point which is who these chips are targeted at.

By offering equivalent or sometimes even better processor performance than the Core i3 3220 for an equivalent price but at much higher energy consumption, and graphics performance that is better than the competition though still not at a high enough level to interest gamers, the Virgo APUs suffer from the same problem as the Llanos did. Although in their mobile versions, the fact that they are often linked up to a low res screen allows them to offer a decent compromise that can make sense, on the desktop side, with the increase in 1080p screens, these APUs quickly run out of breath, including on less demanding titles. And, in practice, adding a graphics card doesn't change much. Only the HD 6670 GDDR5 version improves overall performance… when Dual Graphics works as expected! The variability we saw from one game to another, sometimes increasing, sometimes reducing performance levels, seems to make it too complex for the consumer targeted by these solutions. Who's left? AMD says that it is targeting that very vague segment known as ‘casual’ gamers, those who use small games on social networks… often written in Flash, sometimes HTML5, but never really accelerated in a big way by a graphics card!.

When it comes down to it, we think once again that the customers of these APUs will be the OEMs who, more than anything else, have an eye on price. In this respect, the Virgo APUs may make sense in machines which, in a very precise segment, could then offer higher performance than that given by those built with competitor products. This segment, it’s true, does represent a certain volume of sales (for good or bad!) and this may well explain AMD’s motivation in offering such products. This isn’t the logic of an individual buyer, unless they have very strict size criteria or CPU and GPU power needs corresponding to the balance offered by AMD here.

Let’s try and finish on a positive note all the same: the fact is that, here, AMD is offering, at a reduced price, processors that can be overclocked, have AES-NI instructions and have a graphics part that supports IOMMU (the equivalent of Intel’s VT-d). This is all very positive as these areas are reserved for much more costly models on the Intel side!


Page 14



Copyright © 1997-2014 BeHardware. All rights reserved.