For the 2700, I have been experimenting with only using 14-15 cores for BOINC (two of those cores are supporting GPU tasks).
Something seems off, but I don't know what. CPU is keeping very cool, and I've got a steady clock of 3.7 GHz.
Just a thought -- "CPU is cool and jobs take a long time" sounds like L3 cache is being hammered so the apps effectively become memory-bound.
I don't run Einstein CPU jobs so I've no idea as to how much L3 cache might be needed per task to avoid this; I'd suggest limiting the number of GW tasks (and any other large-footprint tasks if you run other projects!) to see if things improve...
(I had a similar experience on WCG with their MIP1 project (based on Rosetta code) - running more than one of those per 4MB of L3 cache was a sure-fire way of slowing the machine down even though it stlll reported reasonable clock rates!)
This may not be exactly related, but I'm having similar concerns about the GW CPU tasks.
I just upgraded to a 2700 (non-X), and the GW 2.09 CPU tasks are taking about a day to complete on Windows 10. I'm running E@H on a really old laptop, and those tasks are taking about 7 hours...but they are using the GW 2.08 application.
I have discovered that GW CPU tasks appear to be taking 3 CPU threads. On an AMD 3950x (16c/32t) if I run more than 10 GW CPU tasks they slow down considerably. Right now they run 4-6 hours. If I run 26 tasks they slow to more than 12 hours.
I had to upgrade my ram to 64 GB to discover this.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Bill wrote: For the 2700, I
)
Just a thought -- "CPU is cool and jobs take a long time" sounds like L3 cache is being hammered so the apps effectively become memory-bound.
I don't run Einstein CPU jobs so I've no idea as to how much L3 cache might be needed per task to avoid this; I'd suggest limiting the number of GW tasks (and any other large-footprint tasks if you run other projects!) to see if things improve...
(I had a similar experience on WCG with their MIP1 project (based on Rosetta code) - running more than one of those per 4MB of L3 cache was a sure-fire way of slowing the machine down even though it stlll reported reasonable clock rates!)
Cheers - Al.
Gary Roberts wrote: The
)
Rosetta being a CPU-only project, has nothing to do with PCIe bandwidth.
If you watch PCIe bus use while crunching Einstein, it's very low anyway, 0-1%. PCIe bandwidth is not a factor here. It's a CPU/memory bottleneck.
_________________________________________________________________________
Bill wrote: This may not be
)
I have discovered that GW CPU tasks appear to be taking 3 CPU threads. On an AMD 3950x (16c/32t) if I run more than 10 GW CPU tasks they slow down considerably. Right now they run 4-6 hours. If I run 26 tasks they slow to more than 12 hours.
I had to upgrade my ram to 64 GB to discover this.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)