Monday, September 28th 2009
NVIDIA Rolls Out its First OpenCL GPU Drivers
In the tussle between its own CUDA GPGPU standard and the OpenCL, NVIDIA is making the right moves by offering its first drivers for OpenCL GPGPU acceleration. The drivers are available for Windows, Linux, and Mac. Any CUDA-capable NVIDIA GPU will be able to use these. The drivers can be downloaded from here, which provide compliance with OpenCL 1.0.
NVIDIA has also released the OpenCL Visual Profiler software that helps developers improve their code by recognizing possible bottlenecks and room for improvements. At a higher level, it profiles actual hardware signals, kernel efficiency, and instruction issue rate; memory latencies; auto-analysis to point of serialization problems, among other things. More information on this can be found here.
Source:
TechConnect Magazine
NVIDIA has also released the OpenCL Visual Profiler software that helps developers improve their code by recognizing possible bottlenecks and room for improvements. At a higher level, it profiles actual hardware signals, kernel efficiency, and instruction issue rate; memory latencies; auto-analysis to point of serialization problems, among other things. More information on this can be found here.
21 Comments on NVIDIA Rolls Out its First OpenCL GPU Drivers
They already have beta CPU drivers available, and numerous articles and blog postings available on AMD's website.
It's looking like AMD/ATI might be getting their pawns lined up for the official 9.10 CAT release, to corrolate with the first official CAT drivers for the HD5000 series. ATI is pretty good about dropping the hammer hard, we saw how they played out the HD4000 series releases . . . I guess only time will tell, eh?
Anyhow, I'd much rather see an open standard like this become more prevalent - good to see nVidia pushing it forward as well, instead of hiding behind PhsyX . . . now the question is . . . what will Intel do with Larrabee?
Of course, any of that is unlikely to happen, with Nvidia's latest move of excluding PhysX on any machine that detects non-Nvidia hardware (what a dick thing to do). I mean, if someone asks for help in an ATI/Nvidia situation, Nvidia is fully entitled to tell that person they do not support such a configuration and leave it at that. But blocking it all together says something about Nvidia's attitude.
Direct Computer is MS's alternative to OpenCL, Stream & CUDA.
On-topic, I thought they had released them 2 months ago or so. It could be they were beta and these are official?