NVIDIA has just announced their Tegra 2 architecture at CES. Any ideas whether we will be able to develop CUDA applications for this?
I would be very surprised if it were possible. It (and the previous version) do have some sort of programmable shaders, so some old school shader language GPGPU might be possible, but I suspect CUDA or OpenCL would be completely out of the question.
Avidday, could you please tell me the biggest reason why Tegra would never support CUDA? By the way CUDA on Tegra would be amazing !!
The Tegra gpu doesn’t have all of the features necessary to run CUDA. In fact it probably doesn’t have any. It is intended to be able to run OpenGL ES code, and that is about all. There is apparently some degree of programmability in the shaders, but nothing like what would be necessary for CUDA or OpenCL. The entire Tegra 2 SOC die is 49mm^2, and that includes a total of 8 individual cores (of which the GPU is only a small portion of the total die area) - see here for a die micrograph. There is no way even a single G80/90/200 class MP and support logic can fit on something of that size and power envelope.
Thanks for your reply Avidday!!