Originally Posted by jeffmd
yea, sounds like nvidia want's to allow apps like the gpu version of folding@ to run on cpu's as well (bad example, but the work units for the CUDA client are different and can't be run one the cpu client atm. this would help allow this).
I am not sure how helpfull this will be performance wise, it would be faster to design a piece of code to run on a cpu and under 2-4 threads then for cuda and then through and interpreter that then stuffs the 100-200 threads under the cpu.
The additional performance is not the main point, but the ability to have same software for clients who have CPU only systems and might be interested for to change to GPUS if need in future.