NVIDIA Announces CUDA x86 But Omits Launch Date
At the GPU Technology Conference today, the CEO of NVIDIA, Jen-Hsun Huang, unveiled a new CUDA initiative, dubbed CUDA-x86. As the name implies, the new framework will allow developers to write CUDA code natively for x86. Don't confuse this announcement with the PhysX issues we discussed last month—when we spoke to NVIDIA back then we were told that certain legacy performance issues would be addressed in the next major version of the PhysX SDK.
Porting CUDA to x86 is a smart move for NVIDIA given Intel's own intentions towards the high performance computing (HPC) market. One of the core advantages of Intel's hardware will be the fact that it's based on the ubiquitous x86 standard—something NVIDIA can't claim with regard to CUDA. As huge as Intel is, NVIDIA is not without advantages of its own. CUDA is only 3.5 years old, but the GPU designer has been talking about flexible, programmable GPUs ever since the G80 launched five years ago. NVIDIA's advantage isn't just that it was first to market, but that it's spent years (and a great deal of money) developing CUDA and working with developers.
Coming soon, to a processor undoubtedly newer than this one.
Porting CUDA to x86 allows anyone whose curious about the framework to run and examine code whether they own an NVIDIA GPU or not. It also improves Team Green's ability to face off with Intel or even AMD. Both of these companies are touting OpenCL, but again, NVIDIA's OpenCL implementation is considered one of the best. Any high-performance part that Intel launches looks as though it'll have a tough upward fight on its hands. We've no word on when the tools will be available; the heavy lifting is being handled by the Portland Group, which is owned by STMicro.
Porting CUDA to x86 is a smart move for NVIDIA given Intel's own intentions towards the high performance computing (HPC) market. One of the core advantages of Intel's hardware will be the fact that it's based on the ubiquitous x86 standard—something NVIDIA can't claim with regard to CUDA. As huge as Intel is, NVIDIA is not without advantages of its own. CUDA is only 3.5 years old, but the GPU designer has been talking about flexible, programmable GPUs ever since the G80 launched five years ago. NVIDIA's advantage isn't just that it was first to market, but that it's spent years (and a great deal of money) developing CUDA and working with developers.
Coming soon, to a processor undoubtedly newer than this one.
Porting CUDA to x86 allows anyone whose curious about the framework to run and examine code whether they own an NVIDIA GPU or not. It also improves Team Green's ability to face off with Intel or even AMD. Both of these companies are touting OpenCL, but again, NVIDIA's OpenCL implementation is considered one of the best. Any high-performance part that Intel launches looks as though it'll have a tough upward fight on its hands. We've no word on when the tools will be available; the heavy lifting is being handled by the Portland Group, which is owned by STMicro.
