AMD has hinted about CPU-GPU chips for servers, but hasn't committed to any specific products or roadmap.
For this to make sense economically, the semiconductor process will have to be small enough to get a high-end CPU and GPU on the same die. That probably won't be practical until chips can be manufactured below the 32nm node. Also, software that can take advantage of heterogeneous designs will have to be in place to support a broad market for these chips in the enterprise -- i.e., not just for high performance computing. Because of these constraints, I think the earliest we'll see CPU-GPU server chips will be 2012,and more likely 2013.
So where does this leave CPU-less NVIDIA? Right now, the company sits atop the GPGPU computing market, but has no public plans to integrate its high-end GPUs with a CPU. For the time being, at least, NVIDIA seems content to pursue the GPU computing market with discrete devices, like its Tesla products, connected remotely to x86 processors.
Ironically, though, the greater success NVIDIA has in building a GPGPU business and bringing more applications into the fold, the greater the demand will be for CPU integration. And if both AMD and Intel start offering high-end CPU-GPU products, NVIDIA's discrete GPU business will suffer.
It's worthwhile noting that NVIDIA actually does have a CPU-GPU platform in its current Tegra line of processors for mobile devices. The CPU in this case is the ARM processor, a compact little chip that is quite popular for low-power platforms like cell phones. It's not too far a stretch to think NVIDIA may be designing a chip that marries its CUDA-class GPUs with ARM CPUs. This week, startup Smooth-Stone revealed it will build servers based on ARM processors. If these are able to gain a foothold in datacenters, an NVIDIA Tesla-ARM server chip would look very interesting indeed
If you liked this article, please give it a quick review on Reddit, or StumbleUpon. Thanks
How to Make Money