TechSpot: AMD A8-7600 APU review: Single chip PC gaming on a budget?

As the successor to last year's Richland APUs, Kaveri has been updated with new CPU cores based on AMD's Steamroller architecture. The Radeon R7 series GPU has also been integrated, though the 384 SPU version on most Kaveri APUs isn't much different than the A10-6700 and A10-6800's Radeon HD 8670D. Kaveri is AMD's fourth-gen APU while Steamroller is third-gen CPU technology that is supposedly 10% faster per-clock and per-core than Piledriver.

On the GPU side of things, AMD is moving away from the Cayman architecture featured in Richland, which was first seen back in 2010 with the Radeon HD 6000 series. In its place will be a Hawaii / GCN-based GPU, allowing for HSA Heterogeneous Computing.

A new memory controller is also included with Kaveri, supporting DDR3-2400 and hUMA (heterogeneous unified memory access), which gives the CPU and GPU simultaneous access to the same memory. There's PCI Express 3.0 support as well, providing up to 24 lanes for better CrossFire performance.

Read: AMD A8-7600 Kaveri APU Review

These articles are brought to you in partnership with TechSpot.

Report a problem with article
Previous Story

Motorola reveals plans for $50 smartphone, more device customization

Next Story

Press renders of Nokia Lumia Icon leak as launch date gets closer

20 Comments

Commenting is disabled on this article.

AMD still has a ways to go in single thread/core processing. These are still running 4/5 the single core speed of much lower end i3 processors.

On the GPU side, they run circles around Intel. However, just to be competitive CPU side, they are having to throw in twice as many cores, which means most software is still going to suffer.

I am still crossing my fingers for AMD to be able to take on the i3/i5 processors and bring GPU performance that Intel can't currently touch.

Mobius Enigma said,
AMD still has a ways to go in single thread/core processing. These are still running 4/5 the single core speed of much lower end i3 processors.

On the GPU side, they run circles around Intel. However, just to be competitive CPU side, they are having to throw in twice as many cores, which means most software is still going to suffer.

I am still crossing my fingers for AMD to be able to take on the i3/i5 processors and bring GPU performance that Intel can't currently touch.


With Mantle and TrueAudio I'm not particularly worried, and the HSA app benchmarks look pretty serious too. They might not be better for apps you get right this moment, but there's a lot to be said for it long term.

TheExperiment said,

With Mantle and TrueAudio I'm not particularly worried, and the HSA app benchmarks look pretty serious too. They might not be better for apps you get right this moment, but there's a lot to be said for it long term.

If you think a bit faster GPU access and offsetting audio is going to compensate for a 20% computational difference, someone is misleading you.

If you look at gaming now, 90% of the titles are CPU bottlenecked up to 1080p, not GPU. Thus TrueAudio or Mantle will do nothing to help them.

As for thread out to the GPU (HSA), Windows 8 already is doing more of this than people realize.

(Even going back to Vista, OS level software was using early DirectCompute on DX10 hardware, which has been extended in 7 and further in 8 where the OS is taking older software code and shoving it through the GPU when it can.)

A lot of WinRT is using or set up to use the GPU/CPU agnostically, and this may help AMD, but they are going to still need to meet Intel on per core processing.

With Windows 9 (newer concepts brought over from the XB1 development) , more of this happen, but by the time it and other software catches up to provide necessary gains, Intel's GPU adaptations will be closer to what AMD is producing.

The other problem facing AMD is NVidia is retooling for mobile and mobile GPU additions to Intel devices. This could mean discrete NVidia GPUs in tablets running close to the power range of Intel HD GPUs.

When it comes down to base level performance, single thread speeds are crucial. Even if you have 64 cores, if executing a single thread on one core is 20% slower, that software is still going to run 20% faster on a i3/i5 CPU. Even when dealing with 4-8 threads on an i5 class processor it is going to be 10-20% faster than the fastest 8 core AMD CPU.

I'm an not a fan of AMD or Intel on this subject. They both have done good and really disgusting things to the technology market.

Part of Intel's laziness in the past years is waiting out the ATI/AMD integration and 'just' staying ahead of them in core speeds. It is why their lower end mobile 'Atom' technology stood still while Microsoft literally begged them to move it forward over 5 years ago.

AMD also let their CPU performance tank in favor of their GPU technologies. This has kept them a bit ahead of NVidia, but at the cost of remaining competitive with Intel. It shouldn't have needed to be a trade off.

AMD had a better SoC technology they got from Microsoft and started with a better multi-core technology, and have fallen behind Intel with both. There is also no reason AMD should have let up advances on their APUs and let Intel get their HD 4xxx GPUs running as fast as AMD integrated GPUs.

All these technologies you mention will help AMD, but they will also help Intel. I see people also mention hUMA as a savior, but they forget or missed that Microsoft created the software version of this technology back in 2005/2006 and it has been in Windows NT since Vista. (As is how the WDM/WDDM and full video subsystem works in Vista and more specifically the evolution of how it works in Windows 8 today. Notice when AMD talks hUMA, they are very specific to talk about it in 'hardware', because engineers will quickly point out NT is already doing it in software.)


Mobius Enigma said,
If you think a bit faster GPU access and offsetting audio is going to compensate for a 20% computational difference, someone is misleading you.

There's more to Mantle than the considerable overhead reduction, but it'd make up for a lot more than 20%. Even a half clocked FX processor had a huge advantage over D3D11.

Also, as to HSAs potential, here's some early benchmarks - http://www.extremetech.com/com...t-true-heterogeneous-chip/5

You're not all wrong, but you're not all right either.

TheExperiment said,

There's more to Mantle than the considerable overhead reduction, but it'd make up for a lot more than 20%. Even a half clocked FX processor had a huge advantage over D3D11.

Also, as to HSAs potential, here's some early benchmarks - http://www.extremetech.com/com...t-true-heterogeneous-chip/5

You're not all wrong, but you're not all right either.

I should make that more clear. In CPU bound software, even a HUGE difference in Mantle is not going to make any difference, especially when the CPU is running 20% slower per thread.

As I agreed, HSA can make a lot of difference; however, there are a lot of caveats.

For example: The performance gain is in heavily ALU/FP operations, which still isn't lowering the per core latency of more complex instructions.

HSA is essentially just a more granular hardware (on die) level of what Windows NT is already partially doing, and like I mentioned will be getting expanded in Windows 9. (For example a common HSA benchmark is to decode JPGs, and Windows NT has been shoving this through the GPU since Vista.)

Even if Intel does nothing, by the next iteration of Windows, Intel is going to get a bump from the increased CPU/GPU scheduling features of NT. (Unless Microsoft backs off this and decides to leave it to hardware, but that currently isn't the plan.)

Intel also has other options, even with their current CPU designs.

They have also talked about creating a new dual functionality core that would be an extension of their Multimedia processing features and directly focus on offering an faster agnostic core.

Because it would only be handling code that potentially can run on both types of cores, it doesn't have to weighted down by a full CPU core instruction set and also doesn't have to run with only reduced instructions to get a higher speed. (Using very layman terms here so everyone can understand.)


The main thing about HSA is that it is a hardware variation of Windows has been moving toward since Vista was released, and it has potential, but it is highly biased for specific types of operations.


It also still doesn't fix the deficit AMD has with per core speeds that are significantly behind Intel's CPUs.

...and I really want AMD to close this gap and stop burning time with tricks that will be obsolete by the time they reach fruition. Intel also needs the competitive push.

AMD keeps putting time in things like Mantle, hUMA, HSA, and when you step back, it becomes clear that this is stuff that already exists in software (NT specifically) and they are just moving it to the silicon with the hopes of getting a slight bump.


While I don't feel like responding to Wall O Text right now and this will probably be my last post, you talk like GPU accel/GPGPU/HSA are the same thing. It seems like you already know the difference, and yet...

We'll just have to disagree on the potential for the techs, I view it as another AMD64 moment where AMD actually brought something legitimately excellent to the table and Intel will very likely follow suit.

I understand peoples doubts given how far behind AMD had fallen, but I am definitely impressed by the new tech and the only way to confirm either of our beliefs is to wait for the software.

Mobius Enigma said,

I should make that more clear. In CPU bound software, even a HUGE difference in Mantle is not going to make any difference, especially when the CPU is running 20% slower per thread.

As I agreed, HSA can make a lot of difference; however, there are a lot of caveats.

For example: The performance gain is in heavily ALU/FP operations, which still isn't lowering the per core latency of more complex instructions.

HSA is essentially just a more granular hardware (on die) level of what Windows NT is already partially doing, and like I mentioned will be getting expanded in Windows 9. (For example a common HSA benchmark is to decode JPGs, and Windows NT has been shoving this through the GPU since Vista.)

Even if Intel does nothing, by the next iteration of Windows, Intel is going to get a bump from the increased CPU/GPU scheduling features of NT. (Unless Microsoft backs off this and decides to leave it to hardware, but that currently isn't the plan.)

Intel also has other options, even with their current CPU designs.

They have also talked about creating a new dual functionality core that would be an extension of their Multimedia processing features and directly focus on offering an faster agnostic core.

Because it would only be handling code that potentially can run on both types of cores, it doesn't have to weighted down by a full CPU core instruction set and also doesn't have to run with only reduced instructions to get a higher speed. (Using very layman terms here so everyone can understand.)


The main thing about HSA is that it is a hardware variation of Windows has been moving toward since Vista was released, and it has potential, but it is highly biased for specific types of operations.


It also still doesn't fix the deficit AMD has with per core speeds that are significantly behind Intel's CPUs.

...and I really want AMD to close this gap and stop burning time with tricks that will be obsolete by the time they reach fruition. Intel also needs the competitive push.

AMD keeps putting time in things like Mantle, hUMA, HSA, and when you step back, it becomes clear that this is stuff that already exists in software (NT specifically) and they are just moving it to the silicon with the hopes of getting a slight bump.


Hsa does one thing important and is none of what you wrote... it eliminates the need of moving things from main memory or cpu address space to gpu memory. Comparing hsa with gpu acceleration is idiotic because hsa does nothing to change that, it is still gpu acceleration!! What it does is remove the memory copy ping pong overhead!

Mantle reduces cpu relevance. What you mentioned that games are cpu limited is true but with mantle reducing the api overhead .oxide showed that even a 2ghz downclocked fx was gpu constrained with mantle while serious fps decay happened with directx.

Wrapping it up gpu acceleration is the norm you imbecile. EVERY aero since w7 uses gpu accwleration as does ie and some other office pieces. That is not the point!!

The point is that mantle helps feeding the gpu faster by having less cpu overhead because of api and huma which is part of hsa helps feed the gpu by not having to pingpong data between the 2 addresspaces.

This means that low res gaming won't be cpu restricted and the gpu and cpu will comunicate much faster

Edited by Andrew, Jan 24 2014, 8:08am :

tcube said,

Hsa does one thing important and is none of what you wrote... it eliminates the need of moving things from main memory or cpu address space to gpu memory. Comparing hsa with gpu acceleration is idiotic because hsa does nothing to change that, it is still gpu acceleration!! What it does is remove the memory copy ping pong overhead!

Mantle reduces cpu relevance. What you mentioned that games are cpu limited is true but with mantle reducing the api overhead .oxide showed that even a 2ghz downclocked fx was gpu constrained with mantle while serious fps decay happened with directx.

Wrapping it up gpu acceleration is the norm you imbecile. EVERY aero since w7 uses gpu accwleration as does ie and some other office pieces. That is not the point!!

The point is that mantle helps feeding the gpu faster by having less cpu overhead because of api and huma which is part of hsa helps feed the gpu by not having to pingpong data between the 2 addresspaces.

This means that low res gaming won't be cpu restricted and the gpu and cpu will comunicate much faster

This type of crap response is what I get for keeping the topic in layman terms for a public forum. Maybe I also should never have used analogous constructs, as it seemed to confuse you.

Wrapping it up gpu acceleration is the norm you imbecile. EVERY aero since w7 uses gpu accwleration as does ie and some other office pieces. That is not the point!!

Funny thing, it was NOT my point either.

You are correct that I didn't revisit the hUMA aspects of HSA specifically.

I also didn't specifically revisit that Vista introduced two new technologies for GPU usage, that is 'beyond' GPU acceleration.

WDM/WDDM added kernel level GPU 'scheduling' - and wait for it...
Sharing of System/VRAM, aka GPU RAM Virtualization. Starting with Vista, the GPU memory pool is a shared allocation.

Neither of which have ANYTHING to do with your childlike reference to
'Aero w7 gpu accwleration'.

So I apologize for trying to keep the subject light and not specifically revisiting each item from my first point to prevent your confusion.

I however won't waste any more time with this. I get paid to teach similar concepts, and I don't have time to offer free information to be greeted with a childlike, petulant response.

Mobius Enigma said,
WDM/WDDM added kernel level GPU 'scheduling' - and wait for it...
Sharing of System/VRAM, aka GPU RAM Virtualization. Starting with Vista, the GPU memory pool is a shared allocation.
Just because you can, doesn't mean you want to do it. Vista did this for stability reasons, but it doesn't enable developers to work any more efficiently with the GPU; you're still paying a heavy tax everytime you want to talk to the GPU or exchange some data. With a true shared memory design, you're not paying that tax and you can basically be as chatty as you want with the GPU, pass *real* pointers around (i.e. not virtualized), etc. This should enable developers to offload to GPU many more tasks, in a much finer-grained way, than they previously could, which will mean great performance gains.

TheExperiment said,
I'll have my A10-7850K in a few weeks =)

I just finished a new PC build with an A10-7850K. At first it was slower than expected. Turns out that the only display driver currently available is a half-gig AMD beta driver, which can be downloaded from the AMD web site. With the driver installed the machine is quite fast. Video is set to share 1-GB of RAM (8-GB installed, with Windows 8.1). Use the fastest RAM you can, of course, when you order your new gear.

hUMA will be really good for integrated gfx, decent speed boosts to be gained and with mantle will push it ahead of other integrated GPU's so its all good

Lachlan said,

if Microsoft office would change to be optimized for HSA then AMD would have a huge hit on their hands


Well, MS is championing C++AMP so it's entirely possible.