Generic CPU discussion

mikey
mikey
Joined: 22 Jan 05
Posts: 12639
Credit: 1839025349
RAC: 5518

Tom M wrote: I haven't

Tom M wrote:

I haven't succeeded in getting Oracle VM to run reliably under Linux.

Tom M  

One thing, just like in Windows, is you must install the VirtualBox Extension Pack as well and upgrading to the latest version of VirtualBox is not always the best thing in Linux. Some of my Linux pc's run LHC and LHC/dev tasks fine but some just won't. I have also found that they do a lot better on Windows pc's but even then some just won't run them successfully.

hadron
hadron
Joined: 27 Jan 23
Posts: 62
Credit: 89512474
RAC: 589525

mikey wrote: Tom M wrote: I

mikey wrote:

Tom M wrote:

I haven't succeeded in getting Oracle VM to run reliably under Linux.

Tom M  

One thing, just like in Windows, is you must install the VirtualBox Extension Pack as well and upgrading to the latest version of VirtualBox is not always the best thing in Linux. Some of my Linux pc's run LHC and LHC/dev tasks fine but some just won't. I have also found that they do a lot better on Windows pc's but even then some just won't run them successfully.

Installing the extension pack is not necessary; the only essential thing is that you need to add User boinc to Group vboxusers.

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6375
Credit: 9411277495
RAC: 17533516

The All-Sky GW v1.07/v1.15 is

The All-Sky GW v1.07/v1.15 is very sensitive to how fast the CPU is (assuming you hold the GPU make/model constant).  I am running Titan V's under Nvidia's MPS mode/server.

I currently am running an Epyc 7282 (16c/32t) which is showing 3191 MHz peak processing speed for the 20 (4 gpus) CPU threads it is running.

I ran across a used 7F52 (16c/32t) for $295. The seller specs are claiming a 3.9Mhz peak (Turbo) processing speed.

So it Looks like I could get another boost in All-Sky GW task performance.

And that SKU also comes in a 24 core version.  Which means a 8 GPU system running 5 tasks / GPU could also take advantage of this.  This core count appears to have a slightly lower peak performance.

This assumes gpu-only processing.

Anyone want to jump in a try it out?

:)

 

 

 

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4955
Credit: 18611164314
RAC: 5691533

Ignore the peak or turbo

Ignore the peak or turbo clock speed of any cpu.  Does not apply to our processing.


You need to use the max all-core clock speed which is usually defined as base clock speed.


That is how fast your cpu will be running with all cores loaded by Boinc work.

If you are only actually using 20 threads, then the clocks will be closer to the peak or turbo speed.  Maybe halfway between the base loaded clock and the turbo clock.

 

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3927
Credit: 45669042642
RAC: 64014447

Keith Myers wrote: Ignore

Keith Myers wrote:

Ignore the peak or turbo clock speed of any cpu.  Does not apply to our processing.


You need to use the max all-core clock speed which is usually defined as base clock speed.


That is how fast your cpu will be running with all cores loaded by Boinc work.

If you are only actually using 20 threads, then the clocks will be closer to the peak or turbo speed.  Maybe halfway between the base loaded clock and the turbo clock.

 

its not quite so simple as just the base clock speed. It depends on several factors. Like what TDP the CPU has, what application is being run and what optimizations they might run, how many cores the CPU has, etc. 

my 7402P would run basically max single core speed even when all cores were loaded because it was only 24-c with 200W TDP and the max boost clock was capped at 3.35GHz anyway. while the 240W 64-c parts do see much more reduction and closer to base clocks under heavy loads. 

AMD didn’t really define all core load clocks for the 7000 series EPYCs, but it looks like they do for the 9000 series (Genoa/Bergamo), with caveats/constraints 

_________________________________________________________________________

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6375
Credit: 9411277495
RAC: 17533516

Thank you Keith and

Thank you Keith and Ian&SteveC.

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6375
Credit: 9411277495
RAC: 17533516

https://www.xda-developers.co

https://www.xda-developers.com/is-arm-efficient-x86/

Say Duh!

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

Sabrina Tarson
Sabrina Tarson
Joined: 14 Jul 12
Posts: 2
Credit: 162522588
RAC: 795649

Curious to know if anyone

Curious to know if anyone else knows the answer to this, I've tried scouring the internet for a definitive answer but really only found threads from 2005/2006.

On PrimeGrid, it was advised to limit PC's to 50% CPU usage as the applications were so optimized that you're harming performance by running the primaility tests on hyperthreads.

For Einstein@Home, is it more time efficient to use hyperthreads or not? I don't really care about credits, just that work is being done in the most optimal way.

 

taketwicedailey
taketwicedailey
Joined: 30 Nov 17
Posts: 23
Credit: 661546283
RAC: 2703961

In my experience, for my

In my experience, for my hardware, it is the same as you quote for PrimeGrid with the Gamma Ray Pulsar search. Limiting to 50% CPU and then 100% CPU gives roughly the same credit/hour for me.

mikey
mikey
Joined: 22 Jan 05
Posts: 12639
Credit: 1839025349
RAC: 5518

Sabrina Tarson

Sabrina Tarson wrote:

Curious to know if anyone else knows the answer to this, I've tried scouring the internet for a definitive answer but really only found threads from 2005/2006.

On PrimeGrid, it was advised to limit PC's to 50% CPU usage as the applications were so optimized that you're harming performance by running the primaility tests on hyperthreads.

For Einstein@Home, is it more time efficient to use hyperthreads or not? I don't really care about credits, just that work is being done in the most optimal way. 

Alot of the reason PrimeGrid recommends reducing cpu usage to 50% is because the Hyper Threads can be detrimental to faster crunching, so reducing your cpu settings to 50% means you will crunch faster than if you use those 'extra' cpu corers. Now todays cpu's require some testing to see if they do in fact slow down the crunching as they are not like the older Intel cpu cores with HT. The AMD cpu's also have a form of HT and todays AMD cpu's are, like the Intel cpu's, much better at using those 'extra' cpu cores so again testing is required to see if reducing the crunching 50% is really faster.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.