CUDA Now Available For Multiple x86 Processors


I know GPUGrid is focusing on GPUs and that previous attempts of running ACEMD on x86 weren't satisfactory, but can any of the devs say if this may change things?

100 views and no other posts?

Well I guess I know the answer then... :(

You asked a dev's opinion, so I guess none of the 100 viewers were developers. Well, neither am I, but I can give you my opinion, which has not changed since the last time the question was asked. This is essentially a GPU project and I can't see that changing. For now at least it is NVidia only, and again I can't see that changing any time soon. The problem with CPU's is that they can't do as much work. Even an i7-2600 is about 1/14th as powerful as a GTX470. A GTX590 would get through about 30 times the work volume of an i7-2600.

CUDA for x86 might still present an opportunity to research, when CPU's become a bit faster (perhaps within a year the top performance will almost double), but probably not here.

I can only see this being used when a small amount of CUDA needs to be run and a lot of more complex CPU calculations. Perhaps a project that could use a CUDA GPU, if it was there, and if not then use the CPU for CUDA. When I read about this a while back I got the impression this strategy might be intended to allow people to start designing CUDA apps now on their CPU only systems; apps that in the future will be moved to more powerful GPU based systems.

Dear Alexh,
I have read with great interest the article you linked "CUDA Now Available For Multiple x86 Processors". Is there a way that you know to get it to work already for boinc projects?
Ty and take care!
Francesco