Folding Forum

A number of units are highly optimized for the SSE or SSE2 instructions.
Well, having in mind that there is a hardware difference between those two I feel like an optimization possibility exists - is it possible to run a SSE unit along with SSE2 unit at the same time on a single CPU not having the one interfering with the other?
In other words, if you have a SSE WU that would finish in let me say 100 hours and another SSE2 WU that would finish in 70 hours on a given cpu and those two run one after other it'd take 170 hours to finish, but if those two are running together I think that the execution time would be downgraded to less than 120 hours or even less and as a bonus a third WU would be done at approximately 30 to 50%.
Well, I don't feel like I have made a major discovery here - probably there is someone else having the same thoughts earlier. Is there?

Regardless of the optimization method used, running two 100 PPD WUs at the same time on a single core will give a total output of about 100 PPD. The WUs would fight each other for cpu time and each run at half their normal speed, taking twice as long to finish. There would not be a "bonus" 3rd WU.

UF is correct. SSE and SSE2 are not the gating factor for processing speed. The CPU is, and sharing is not recommended.

pafka wrote:Well, having in mind that there is a hardware difference between those two I feel like an optimization possibility exists - is it possible to run a SSE unit along with SSE2 unit at the same time on a single CPU not having the one interfering with the other?

Since most of the execution state is in common between SSE and SSE2, you'd soak up a lot of time doing context switching.