Generic CPU discussion
19 Feb 2021 21:08:02 UTC
Topic 224870
(moderation:
Hi,
It currently looks like my gr#5 cpu tasks are taking a third again more time than the gr cpu tasks on the same box.
I am retesting that.
If gravity wave cpu tasks run faster and award more credits then I will be switching back to gravity wave cpu tasks.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Complete with GR typo when I
)
Complete with GR typo when I meant GW.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Who would have thought?I
)
Who would have thought?
I started running out of memory when I switched to nearly all E@H CPU tasks. I had to up the memory on the AMD 3950x box to 32GB again.
In the next day or two, it will finish off all the GR CPU tasks and start processing the faster running/better "paying" GW CPU tasks.
The RAC is tracking down as expected. My current ballpark estimates will place its RAC score down below 1,000,000
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
I was looking at the "ATLAS
)
I was looking at the "ATLAS AEI HANNOVER" user because they seem to be running pretty much all CPU systems.
On one system the wallclock time was significantly higher than the cpu time for GR cpu tasks.
I wonder what else that machine is processing?
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
May not be processing
)
May not be processing ANYTHING else other than normal host housekeeping work.
All it takes to overcommit a cpu is run 100% of the cpu on cpu tasks and not leave any other resource for housekeeping and the housekeeping is forced to steal timeslices away from running tasks. Hence you get more wall clock time than runtime.
I'm also seeing the s3
)
I'm also seeing the s3 spotlight GW CPU tasks running at 1/3 the times of the s1 spotlight CPU tasks.
I was originally running 5 CPU s1 tasks and then noticed the times reduced by half after the s3 tasks started. But it turns out 5 s3 tasks were running into memory limits on my machine with 16 GB of RAM.
I had to go down to 4 s3 spotlight CPU tasks to avoid memory limits. Watching the tasks running, they first use 2.1 GB of RAM per task, then about 15% of the way through they drop down to average around 1.2 GB of RAM, although there are fluctuations. So now running 4 s3 spotlight CPU tasks I see the tasks running at 1/3 the speed of s1 tasks.
I have been going around and
)
I have been going around and around with my AMD 3950x box right after I upgraded one of the Nvidia cards.
After swapping ram modules, reinstalling Boinc, running SFC /scannow etc.
I kept getting errors and the CPU GW tasks kept crashing.
Last night after backing up the BOINC folder I kicked off a Windows OS reset saving "nothing".
This morning after I set up the OS again, I downloaded/installed a slightly older version of Boinc and got the system crunching again on E@H.
Up to this point, I have been experiencing very high ram usage and not been able to run 26 CPU threads using GW tasks.
I had to throttle the CPU tasks to 18 or so to get the CPU load to below 100%.
I am trying again to run 26 CPU threads this morning and see if my CPU utilization stays below 100%.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Surprised that your 3950X is
)
Surprised that your 3950X is at 100% utilization while only running 26 tasks.
I didn't realize that Windows had THAT much overhead compared to Linux.
I am running 30 of 32 cpu cores and am only at 88-92% utilization under Ubuntu 20.04.2 LTS.
Keith Myers wrote:Surprised
)
I think the problem was/is the OS was damaged.
Once I got the OS reset and a slightly older version of the BOINC system installed I was able to run 15 CPU threads of Gravity Wave processing with very little HD activity. I have 32 GB of ram installed and have had to allow BOINC Mgr to use 90% of the available memory.
I have crowded that up to 18 but someplace in there I am going to start getting a lot of paging activity which makes the whole system laggy at the keyboard and probably slows the CPU task processing down.
So far the CPU utilization has not pegged 100% and stayed there. As of this moment it is running about 75%.
And apps/Boinc mgr have stopped crashing.
I think I am hitting the same kind of issues we had when we migrated to Rosetti. The GW cpu tasks are using a lot more memory than Seti@Home did. I think the GR cpu tasks use less memory but process slower and pay less credits.
Just for giggles I priced some Amd-specific ram to jump to 64 GB. And ran across some that total for that amount of memory for about $200. Everything else was a bunch higher. Cash flow insufficient. Sigh.
===edit===
Just bounced it up to 26 cpu tasks and the cpu is doing 100% then it slowed down a little, but memory pegged to 100% and I now have a busy hard disk and a laggy machine.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Yes, the memory
)
Yes, the memory overcommitment at Rosetta when I first joined is what prompted me to upgrade from 16GB to 32GB.
But that still didn't solve the issue of overflowing into paging and virtual memory usage which ground the system to a complete halt being unresponsive. Had to power reset the host to gain it back.
I stopped Rosetta on the PC's after that. Only running on the SBC now.
Keith Myers wrote:Yes, the
)
Looks like I can sustain 18 cpu tasks and 2 gpu tasks without getting the paging file too active.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)