Tweet
From my experience, old 6-core can't be faster than 22-core, except maybe in single threaded ops. So, I was kinda surprised - 6-core 5820 is faster when solving watery stuff, than 22-core. Sometimes, twice faster, sometimes 50%. But is always faster, until me starting to limit the threads phx allowed to use.
After spending 12-hours to simulate with different threads, I found this:
max threads = 0 (default and it is 44-threads) 1.05mvox to 1.67mvox after wetmap starts to dry (so I can see wetmap particles amount starts to get lower)
max threads = 4 0.72mvox to 0.76mvox
max threads = 8 1.10mvox to 1.30mvox
max threads = 12 1.27mvox to 1.69mvox
max threads = 16 1.36mvox to 1.82mvox
max threads = 20 1.36mvox to 1.96mvox
max threads = 24 1.24mvox to 1.84mvox
max threads = 28 1.21mvox to 1.85mvox
So, the fastest is 20-threads, the slowest is 44 and 4.
And just 5820 results:
max threads = 0 1.49mvox to 2.10mvox
The only thing came to my stupid brains - smth is very single threaded, because 5820@3.4Ghz (uncore 3.6), while my xeon 2699v4@2.4Ghz (uncore 2.6).
Also, smoke simulations are fine - the more threads you gave the faster it will be. In houdini, my xeon is 3.5 times faster, than 5820, when simulating lava.
P.S. Phx scene uses: wetmap/air effects, SPF=4, grid is about 15mill, water is 7mill, wetmap 150K at highest. Hope, you'll gonna find bottleneck in your watery stuff. I send svetlin logs from 6-core and my xeon. Hope, they help.
After spending 12-hours to simulate with different threads, I found this:
max threads = 0 (default and it is 44-threads) 1.05mvox to 1.67mvox after wetmap starts to dry (so I can see wetmap particles amount starts to get lower)
max threads = 4 0.72mvox to 0.76mvox
max threads = 8 1.10mvox to 1.30mvox
max threads = 12 1.27mvox to 1.69mvox
max threads = 16 1.36mvox to 1.82mvox
max threads = 20 1.36mvox to 1.96mvox
max threads = 24 1.24mvox to 1.84mvox
max threads = 28 1.21mvox to 1.85mvox
So, the fastest is 20-threads, the slowest is 44 and 4.
And just 5820 results:
max threads = 0 1.49mvox to 2.10mvox
The only thing came to my stupid brains - smth is very single threaded, because 5820@3.4Ghz (uncore 3.6), while my xeon 2699v4@2.4Ghz (uncore 2.6).
Also, smoke simulations are fine - the more threads you gave the faster it will be. In houdini, my xeon is 3.5 times faster, than 5820, when simulating lava.
P.S. Phx scene uses: wetmap/air effects, SPF=4, grid is about 15mill, water is 7mill, wetmap 150K at highest. Hope, you'll gonna find bottleneck in your watery stuff. I send svetlin logs from 6-core and my xeon. Hope, they help.
Comment