IPhVu::iParallel

NVIDIA’s CEO Jen-Hsun Huang has given a keynote speech at GTC 2010 last week. There are some interesting announcements in the speech that I note here.

The first and foremost announcement is about CUDA-x86 compiler which is developed by PGI. The new compiler is expected to show the first time at SuperComputing conference this November. It is fancy to know that we can write, compile, debug and deploy CUDA programs on every workstation, server or cluster even when there is not any CUDA-accelerated GPU. When running on systems which do not have CUDA devices, CUDA-x86 will help CUDA programs run as usual thanks to the streaming SIMD instructions set in Intel’s and AMD’s CPUs.

The question is how well do they perform on CPUs-only system? It would be great if they can provide approximate speedup as on GPU, but I do not expect this much. Let’s wait until SC2010 and see what can we do with CUDA-x86.

The remaining part of Jen-Hsun Huang’s speech is about some fancy demos and some guests from nVidia’s partners. This includes  CUDA-accelerated real-time rendering tool in Autodesk 3ds Max, a GPU-derived engineering simulation solution from ANSYS, GE Intelligent Platforms’ new development environment, CUDA officially supported right in MATLAB (one of the best news!), OpenCV is now supported by CUDA (another good news, but I hope that CUDA-OpenCV, or whatever they call it, will not add any bugs into OpenCV), Wolfram Research’s Mathematica added built-in CUDA GPU programming.

The most important part of the conference surely is not the speech from the CEO, but the research posters. There are many interesting researches there. Hope that I will have enough time to read and summarize some of them someday.