CPU tasks.

adrianxw
adrianxw
Joined: 21 Feb 05
Posts: 203
Credit: 202,047,135
RAC: 80,274
Topic 220711

I noticed an Einstein on one of my machines that had a long "Elapsed", but it seemed to be running. It was a CPU task, not the regular GPU. Can I specify somewhere that on  a certain machine, I only want CPU tasks? The reason I ask is that I have one machine where the graphics card is in a bad way, so I removed all projects that ran in the GPU to prevent it getting worse before I could replace it.

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.

Richie
Richie
Joined: 7 Mar 14
Posts: 542
Credit: 1,675,600,012
RAC: 125,093

https://einsteinathome.org/ac

https://einsteinathome.org/account/prefs/project

Look for 'Preference set'. There are four different preference sets available (generic, home, school, work). You can use them to save different settings for four different computation scenarios. Any of those preference sets are then available for any of your hosts. Visit the details page of a host and select a corresponding 'Location'. Save changes and click update for Einstein on Boinc.

Jan Vaclavik
Jan Vaclavik
Joined: 1 Sep 05
Posts: 10
Credit: 1,018,256
RAC: 1,715

I am wondering if the CPU

I am wondering if the CPU apps are worth the time and electric power compared to the GPU apps.

Tom M
Tom M
Joined: 2 Feb 06
Posts: 188
Credit: 258,034,418
RAC: 1,792,479

Jan Vaclavik wrote: I am

Jan Vaclavik wrote:

I am wondering if the CPU apps are worth the time and electric power compared to the GPU apps.

 

Great Question :)

If you have unused cpu threads that you want to use and haven't choosen to use them on a cpu only project like Rosetti@Home or World Community Grid then it is a question of your budget for electricity.

If your goal is maximum production on a system then all the "available" threads should be fully engaged.

If your goal is opitmin (sp not working) production vs. power effeciency (sp not working) then you are probably going to have to get a "Kill-a-watt" meter and test for the differences.

Proud Member of the OFA (Old Farts Assoc.)

 

 

Jan Vaclavik
Jan Vaclavik
Joined: 1 Sep 05
Posts: 10
Credit: 1,018,256
RAC: 1,715

Tom M wrote:If you have

Tom M wrote:

If you have unused cpu threads that you want to use and haven't choosen to use them on a cpu only project like Rosetti@Home or World Community Grid then it is a question of your budget for electricity.

If your goal is maximum production on a system then all the "available" threads should be fully engaged.

If your goal is opitmin (sp not working) production vs. power effeciency (sp not working) then you are probably going to have to get a "Kill-a-watt" meter and test for the differences.

Running WUs on CPU for hours, if those very WUs could be completed within minutes on GPU does not seem like an efficient way to burn electricity. Its not like my CPU is idling either, it is running multiple CPU-only projects including those you mentioned.

I am simply wondering, if running, say, FGRPSSE adds anything to the project other than few extra flops, while the same results could be achieved more efficiently by running FGRPopencl.

mikey
mikey
Joined: 22 Jan 05
Posts: 5,922
Credit: 544,853,085
RAC: 114,755

Jan Vaclavik wrote: Tom M

Jan Vaclavik wrote:

Tom M wrote:

If you have unused cpu threads that you want to use and haven't choosen to use them on a cpu only project like Rosetti@Home or World Community Grid then it is a question of your budget for electricity.

If your goal is maximum production on a system then all the "available" threads should be fully engaged.

If your goal is opitmin (sp not working) production vs. power effeciency (sp not working) then you are probably going to have to get a "Kill-a-watt" meter and test for the differences.

Running WUs on CPU for hours, if those very WUs could be completed within minutes on GPU does not seem like an efficient way to burn electricity. Its not like my CPU is idling either, it is running multiple CPU-only projects including those you mentioned.

I am simply wondering, if running, say, FGRPSSE adds anything to the project other than few extra flops, while the same results could be achieved more efficiently by running FGRPopencl. 

Not everyone has a gpu that can crunch or wants to use their gpu's for crunching so cpu workunits make sense for those people. For the most part I also do not run cpu workunits for projects that also have gpu workunits. I split my cpu's to one project and gpu's to a different project, it helps with the work cache too.

Jan Vaclavik
Jan Vaclavik
Joined: 1 Sep 05
Posts: 10
Credit: 1,018,256
RAC: 1,715

mikey wrote: Not everyone

mikey wrote:

Not everyone has a gpu that can crunch or wants to use their gpu's for crunching so cpu workunits make sense for those people. For the most part I also do not run cpu workunits for projects that also have gpu workunits. I split my cpu's to one project and gpu's to a different project, it helps with the work cache too.

I have an old Intel GPU and it runs exclusively Einstein (since it can not run pretty much any other project), but even that old GPU seems to be more efficient than my CPU threads.

Since I am already contributing some computing power, I dont need to run the CPU apps at all cost, but if they do something the GPU apps do not, I may consider throwing few into the mix.

mikey
mikey
Joined: 22 Jan 05
Posts: 5,922
Credit: 544,853,085
RAC: 114,755

Jan Vaclavik wrote: mikey

Jan Vaclavik wrote:

mikey wrote:

Not everyone has a gpu that can crunch or wants to use their gpu's for crunching so cpu workunits make sense for those people. For the most part I also do not run cpu workunits for projects that also have gpu workunits. I split my cpu's to one project and gpu's to a different project, it helps with the work cache too.

I have an old Intel GPU and it runs exclusively Einstein (since it can not run pretty much any other project), but even that old GPU seems to be more efficient than my CPU threads.

Since I am already contributing some computing power, I dont need to run the CPU apps at all cost, but if they do something the GPU apps do not, I may consider throwing few into the mix. 

I put some of my cpu's on Rosetta because they are working on COVID-19 but there are more projects that can NOT use a gpu than can use a gpu. If you scroll thru the 'add project' list in the Boinc Manager it will give you a basic idea of what app each project has in the box of  the right.

Holmis
Joined: 4 Jan 05
Posts: 1,117
Credit: 772,563,282
RAC: 622,357

From experience and comparing

From experience and comparing the Applications page and the Server status page I conclude the following:
(Disclaimer: I could very well be wrong in this! ;) Please do correct any faults!)

  • Gravitational Wave search O2 Multi-Directional:
    • Have both CPU and GPU applications, but the server status pages lists 2 different runs:
      O2MDF = GPU tasks
      O2MD1 = CPU tasks
      They are not the same and analyzes different targets, so both do contribute!
       
  • Gamma-ray pulsar search #5:
    CPU only, FGRP5
     
  • Gamma-ray pulsar binary search #1 on GPUs:
    GPU only, FGRPB1G

Gamma-ray tasks get data from the same place but run different searches on CPUs and GPUs, so both do contribute!

For completeness the "Binary Radio Pulsar Search (Arecibo)" (BRP4) is sent to iGPUs, Android & ARM devices. For ARM you need at least v7 to get it working, v6 is broken.
iGPUs depend on the type of device and for Intel the driver version (at least for the older ones) if it's going to work or not.

Jan Vaclavik
Jan Vaclavik
Joined: 1 Sep 05
Posts: 10
Credit: 1,018,256
RAC: 1,715

Holmis wrote:From

Holmis wrote:

From experience and comparing the Applications page and the Server status page I conclude the following:
(Disclaimer: I could very well be wrong in this! ;) Please do correct any faults!)

  • Gravitational Wave search O2 Multi-Directional:
    • Have both CPU and GPU applications, but the server status pages lists 2 different runs:
      O2MDF = GPU tasks
      O2MD1 = CPU tasks
      They are not the same and analyzes different targets, so both do contribute!
       
  • Gamma-ray pulsar search #5:
    CPU only, FGRP5
     
  • Gamma-ray pulsar binary search #1 on GPUs:
    GPU only, FGRPB1G

Gamma-ray tasks get data from the same place but run different searches on CPUs and GPUs, so both do contribute!

For completeness the "Binary Radio Pulsar Search (Arecibo)" (BRP4) is sent to iGPUs, Android & ARM devices. For ARM you need at least v7 to get it working, v6 is broken.
iGPUs depend on the type of device and for Intel the driver version (at least for the older ones) if it's going to work or not.

The way I understand this thread is that GW on both CPUs and GPUs are basically the same and the two were only divided so that CPUs do not slow down GPU validation.

 

I am using the HD Graphics P4000, which is basically the same as the 4000 in 3770K and I had no problems with FGRPB1G and only few invalid results when running BRP4.

Tom M
Tom M
Joined: 2 Feb 06
Posts: 188
Credit: 258,034,418
RAC: 1,792,479

Apparently Gamma-Ray#5 cpu

Apparently Linux Gamma-Ray#5 cpu tasks have the following results on two different Amd model cpus.

Neither run with Turbo settings.

Amd 2700 ~8 hours

Amd 3950x ~3 hours

Proud Member of the OFA (Old Farts Assoc.)

 

 

Comment viewing options

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