Hopeless over-allocated; but why and how ?

[ChristianVirtual]

Somehow I'm not aware that I configured anything for this client. Just plain CentOS and installed from boinc-source code

Epyc 24/48 but how the hell it can run a 48T yafu, a VM for LHC and drink some sake at the same time ?

 

[Deleted member 88227]

Well that's a 32 thread YAFU WU so it's most likely only using 32 threads. Then again, knowing YAFU it's probably only using 1 thread by now.

 

[Gilthanis]

It may not be running properly either. I would close the client and start up again or just do a full pc restart and see what happens.

 

[RFGuy_KCCO]

Somehow I'm not aware that I configured anything for this client. Just plain CentOS and installed from boinc-source code

Epyc 24/48 but how the hell it can run a 48T yafu, a VM for LHC and drink some sake at the same time ?

View attachment 108267

You do not appear to be running the optimized app for Rake Search. Do yourself a favor and install the optimized app and cut those run times down by 1/4 or more.

For your EPYC, you'll want to run the "rakesearch_windows_64_avx2nopext" app.

Info: http://rake.boincfast.ru/rakesearch/forum_thread.php?id=39#172

Direct download: https://github.com/sirzooro/RakeSearch/releases/tag/v1.0

 

[ChristianVirtual]

You do not appear to be running the optimized app for Rake Search. Do yourself a favor and install the optimized app and cut those run times down by 1/4 or more.

For your EPYC, you'll want to run the "rakesearch_windows_64_avx2nopext" app.

Info: http://rake.boincfast.ru/rakesearch/forum_thread.php?id=39#172

Direct download: https://github.com/sirzooro/RakeSearch/releases/tag/v1.0

Thanks for the pointer; didn't followed rake too much in the past and it sure will help to get the million faster ! Maybe then even the 10M

 

[ChristianVirtual]

Second question: conscious overallocation ... a good idea ?

Some projects not fully utilizing the CPU at any time. With some "manual" workaround I',m able to squeeze more then 100% constant load on the system. But wonder to what degree that is helpful or harmful in terms of performance. Understand that switching tasks / processes on OS level can be expensive too.
On the positive side I make sure I have no idle CPU cycles; on the negative side I slow down the overall delivery of WU due to overhead.

I will try to get it to 130% - 150% max ; those 200% seems a bit high. Any experience with that ?