About CUDA

[Archer0915]

Can CUDA be used to take some of the load off of the cpu in day to day tasks kinda like a CoPro? I know the GPU is excellent at parallel ops and the like and I just kina wondered if it would be worth it to buy an 8XXX or above card and tro writing some load sharing software or would I be wasting my time?

[Archer0915]

Any takers?

[Zap]

NVIDIA CUDA ZONE (http://www.nvidia.com/object/cuda_home.html)

Basically if whatever you are trying to do can take advantage of that kind of massively parallel processing and someone takes the time to make a CUDA app for it, then you will benefit. It is like MMX/SSE/SSE2/SSE3/etc. Some applications can and do take advantage of the extra instructions, while others do not. For CUDA more specifically, looks as if there is a lot of interest in using it for major number crunching as any time savings on tasks that may take days to compute will be tremendous, while making Windows boot a few seconds faster or shaving a few seconds off converting that Word document into a PDF isn't going to seriously imrove lives.

CUDA is an emerging technology, so don't expect it to sing and dance overnight. Still, some interesting uses for consumers are becoming available. Beyond the Baddaboom encoder that NVIDIA touts, TMPGEnc and Nero now support CUDA. I'm sure you'll see more and better support in the future.

[don'tknow]

I'm having some very good results with TMPGEnc 4.0 XPress and CUDA. Very good quality with the right settings, using 1 pass Constant Quantization to encode HD in h.264 uses 55% CPU + 45% CUDA and finishes much faster than any other software I've used. To get the same quality with other programs I had to use 2 passes with lots of process-intensive quality settings and it took approx. 10 times longer, while using both CPU cores 100% the whole time. In either case the CPU is definitely not 'worthless', but using it along with GPU is the most beneficial.

[Archer0915]

I'm having some very good results with TMPGEnc 4.0 XPress and CUDA. Very good quality with the right settings, using 1 pass Constant Quantization to encode HD in h.264 uses 55% CPU + 45% CUDA and finishes much faster than any other software I've used. To get the same quality with other programs I had to use 2 passes with lots of process-intensive quality settings and it took approx. 10 times longer, while using both CPU cores 100% the whole time. In either case the CPU is definitely not 'worthless', but using it along with GPU is the most beneficial.
That is what I was looking for I was kinda looking at the encoding aspect as well as other offloading that could be done with a little emulation:cool: