Folding Forum

I have a couple of Intel Core i7 980x CPUs; one is liquid cooled and running at 4.4 GHz, folding on 10 threads with the last two serving GPU slots, and the other is running at 4.0 GHz on air, folding on all 12 threads with no GPU slots. The 4 GHz CPU which has 12 threads assigned to the client is only producing around half of the PPD that the 10 thread client is. Can anyone think of a reason behind this? I only noticed it this afternoon but looking back at the logs and stats it seems to have been underperforming for several weeks. TPF on Project 11635 is 9 minutes 38 seconds, which seems extremely slow for a 4 GHz 12 thread CPU.

The i7 980x has 12 threads but only 6 cores. Running 12 threads means each thread has to compete with another thread for the floating point hardware. Running only 10 threads means that two threads have accesss to full floating point power and the other 8 have to compete with another thread. By the time the various threads have to synchronize, it won't be as simple as 8 of one type and 2 of another as they'll sort of average out, but there's no doubt that adding two more threads won't contribute very much. (While the two CPUs that support the GPUs use the full thread, the floating point arithmetic is done on the GPU so they don't compete for the CPU's FP hardware.)

You didn't state that you're comparing two WUs from the same project. (There are some variations between projects.

I don't have enough information to suggest that this is the only difference that you're seeing.