NVIDIA: ARM will rule PC CUDA not open

David Kirk, a former NVIDIA academician who served as chief scientist, recently said in an interview with the media that an SoC chip based on an ARM processor architecture and a GPU graphics core will eventually dominate the entire PC.

The fact that x86 cannot be obtained compels NVIDIA to strive to improve the status of the GPU, while relying on the ARM architecture. It is natural to make such an argument.

David Kirk said: "Obviously, the PC experience will be dominated by SoCs that integrate ARM cores with GPUs. All of this is becoming a reality and will be consolidated under Windows Next (that is, Windows 8 operating system) support for ARM. But as I said, CPU+GPU will coexist for a long time."

David Kirk also mentioned CUDA technology, claiming that it will not be open. He said: "There are no plans to program the CUDA open standard. Currently, x86 CPUs and NVIDIA GPUs that are widely used in the server space have all supported the CUDA development kit. NVIDIA offers developers a rich choice, CUDA C, CUDA C++, CUDA Fortran, OpenCL, DirectCompute, etc. can all be used to program CPU+GPU systems."

With the rise of CPU+GPU heterogeneous computing system, can NVIDIA CUDA adapt to the new situation? David Kirk looks full of confidence. He said: “CUDA is our innovation platform... The beauty of the CUDA programming model is that it is designed for a heterogeneous system based on CPU+GPU. CPU and GPU integration will not change the programming model. Integration is simply For cost considerations, the time of our research on ARM+GPU SoC Tegra processors is similar to that of CUDA, and other driver-level APIs such as OpenCL treat the GPU as a device different from the CPU, which means that the defined OpenCL must be expanded in the future. CPU+GPU devices are supported, and applications written using CUDA run smoothly on our CPU+GPU devices."