AMD Fusion Details

A Phenom dual cored CPU and a RV7xx GPU.. Wha??

Im confused, how will a dual cored CPU on the same PCB of a GPU help performence, what workload will it take and how will it communicate with the GPU..

meh..
 
I think you missed the point, its more a replacement for integrated graphics without the need for a separate gpu.
 
If the power usage is low, I could see this being the basis for an incredible mobile part.

Knowing Phenoms & RV7s, this will NOT be low power usage but still interesting to watch. Depending on price it could lead to a VERY cheap good media center PC, which is exciting.
 
A Phenom dual cored CPU and a RV7xx GPU.. Wha??

Im confused, how will a dual cored CPU on the same PCB of a GPU help performence, what workload will it take and how will it communicate with the GPU..

meh..

No, it's a dual core Phenom on a low-end RV800 part.

It won't help performance, but it will lower cost. Performance isn't the issue for this particular part.
 
It's meant to move to non-homogeous cored chips in the future, and this is what AMD-ATI and Intel's Larrabee are really trying to do - integrate GPU and CPUs for teh future
 
It's meant to move to non-homogeous cored chips in the future, and this is what AMD-ATI and Intel's Larrabee are really trying to do - integrate GPU and CPUs for teh future
did you see today's article on anandtech about Larrabee? They're really totally different parts. Larrabee is a massively parallel x86 CPU capable of running DX and OpenGL code paths with a software wrapper that converts the DX/OpenGL code into x86 code. Meanwhile, Fusion appears more targetted at budget applications and is more or less a CPU, northbridge, and GPU slapped together on the same piece of silicon. Larrabee is a complete fundamental redesign of the way graphics are processed, while Fusion sticks to a tried and true tradition approach. When Larrabee gets launched, you'll still need a tradition CPU (Nehalem) to run the OS, while Fusion will be a standalone part. From a technical/architectural stand point, the details released today on Larrabee made me jizz my pants but it remains to be seen if it's a practical approach from a performance standpoint.
 
It's meant to move to non-homogeous cored chips in the future, and this is what AMD-ATI and Intel's Larrabee are really trying to do - integrate GPU and CPUs for teh future

Nope, that is not at all what they are trying to do. A GPU cannot replace a CPU. It doesn't work. The two are just far too different. A GPU can augment a CPU in general purpose work, though, which is more the goal of CUDA/Larrabee. To take tasks that the CPU is bad at (massively parallel tasks) and move them to hardware designed for such tasks (the GPU). Eventually in the far distant future we *may* see a merger of the two, but not any time soon. CPUs focus largely on single threaded integer performance, whereas GPUs focus on massively parallel floating point calculations.

The goal of Fusion is to replace IGPs. A low power, low performance part that is perfect for those super-budget builds (such as those $300-400 Dells)
 
No I know exactly what Larrabee does, my point is that they are trying to merge the two together in the long run. It's not too hard to imagine Larrabee cores one day being integrated around say... Nehalem cores on a 32nm process sometime in 2010-2011.

My point is that both sides see the fusion of GPU and CPU - Intel's approach is that Larrabee is built on pre-existing CPU cores and re-engineered to perform in-order vector operations while AMD's approach is to bring GPUs as a core. The ultimate goal for both sides is to enable everything on one package - GPU functionality along with CPU functionality all in one die.

Both projects are just stepping stones to their ultimate goal. And of course the CPU can't be easily replaced, and Intel and AMD make the bulk of their money off CPUs and ensuring they are functional. But I do not doubt that they both see the possibility in the future of combining featuers even more so you can do CPU + GPU tasks on the same die with non-homogeneous cores.
 
If you have two parts fighting for space, separating them will always yield a better overall package. It is actually impossible to imagine larrabee cores being integrated around Nehalem in 2010, as that is when Intel hopes to have them out using 45nm and old pentium cores. The space that all those cores and vector units take up could be used for a larger cache, or other improvements to improve single threaded performance. Likewise, GPUs don't need a massive cache, so the huge cache on a CPU would be wasted space in a GPU. There will always be this tradeoff. As such, I don't think we will see a merger of the two in a high performance environment within the next 5-7 years.

I also don't think that either side wants to merge the two in the long run, as they both know that 1) They are very different parts with very distinct and separate goals and designs, and 2) They make more money selling them separate.
 
I also don't think that either side wants to merge the two in the long run, as they both know that 1) They are very different parts with very distinct and separate goals and designs, and 2) They make more money selling them separate.

However, in the low-end you could make MORE money by combining the two. You get to sell a CPU and a GPU without the cost of packaging the GPU. In a mobile part you could even save quite a bit of power.
 
However, in the low-end you could make MORE money by combining the two. You get to sell a CPU and a GPU without the cost of packaging the GPU. In a mobile part you could even save quite a bit of power.

Exactly, hence, Fusion :)
 
Back
Top