affinity settings for 2xE2650

[g33py]

Hi,

I have tried two options:
a) running 2 workers with 10 threads on one socket - the average time per iteration is about 20ms
b) running 2 workers with 10 threads one on each socket - the average time per iteration is about 8 ms

It seems that prime95 is not really independent of the cpu cores used...

I will continue with the b) setting as it seems to get the most iterations per seco´nd out of the power I can spare.

[Madpoo]

Originally Posted by g33py

Hi,

I have tried two options:
a) running 2 workers with 10 threads on one socket - the average time per iteration is about 20ms
b) running 2 workers with 10 threads one on each socket - the average time per iteration is about 8 ms

It seems that prime95 is not really independent of the cpu cores used...

I will continue with the b) setting as it seems to get the most iterations per seco´nd out of the power I can spare.

Your a) option of 2 workers of 10 threads each wouldn't work well because then it's using the hyperthreads which really don't help with the FFT code (may actually slow things down).

Essentially each worker is really only using 5 physical cores, and in theory it would be twice as slow as a single worker with 10 physical cores. The fact that it's even slower than that (slower than 16ms per iteration) implies the hyperthread usage is slowing it down by introducing delays into the overall arrangement.

Option b) is a good way to go though.

[LaurV]

Originally Posted by g33py

It seems that prime95 is not really independent of the cpu cores used...

I have two bundles of rags, each of 100kg each and I have tried two options:
a) I put them on my ass and he almost fainted.
b) But when I put my bundles on two different asses, each bundle on one ass, they both run with the bundles into the forest...
It seems that my bundles of rags are not really independent of the ass I put them on...