Decoding my computer's utilization patterns

Message boards : Questions and problems : Decoding my computer's utilization patterns
Message board moderation

To post messages, you must log in.

AuthorMessage
AdsoYo

Send message
Joined: 24 Jan 21
Posts: 10
United States
Message 106786 - Posted: 16 Jan 2022, 8:16:31 UTC

I'm wondering if productivity is being left on the table here. I run Einstein and Milky Way on both my computers. One has a Ryzen 5800x, Radeon 6800XT, custom water cooled. The other is a Ryzen 3700x, GeForce 980ti, custom water cooled and overclocked 20%. They both show similar temperature and utilization patterns on MSI Afterburner. CPU is always pegged at 100%, usually running nice and hot (although the 3700x runs much cooler than the 5800x), GPU shows either the sawtooth's or isn't being used at all. I was under the impression that BOINC projects would use the GPU as much as possible because it is so much more powerful than the CPU, especially in my computers. Is that not actually the case? Is there a way to tell BOINC to use the GPU more?



ID: 106786 · Report as offensive
Profile Dave
Help desk expert

Send message
Joined: 28 Jun 10
Posts: 2515
United Kingdom
Message 106788 - Posted: 16 Jan 2022, 8:41:02 UTC

What does manager say about how many GPU tasks are running? Also, worth looking at any web based preferences for the projects. Some projects you have to enable GPU processing for specific sub projects on the respective websites. Another thing is, have you checked the project websites to confirm they currently have GPU tasks.

With respect to GPUs being more powerful than CPUs the answer is, "it depends." Tasks that can make use of computing a large number of threads simultaneously can be much faster on graphics processors whereas my main project CPDN doesn't have any GPU applications because of the sequential nature of the tasks where each calculation depends on the result of the previous one a GPU would be a lot slower than a CPU with a fast floating point unit.

Pretty sure some of those with actual experience of the projects you run will be chipping in soon and they will likely have answers to the questions I posed.
ID: 106788 · Report as offensive
Bryn Mawr
Help desk expert

Send message
Joined: 31 Dec 18
Posts: 284
United Kingdom
Message 106790 - Posted: 16 Jan 2022, 15:00:41 UTC

If the GPU utilisation is showing an on/off pattern it would be worth looking into whether you can run multiple tasks on the GPU as it can be indicative of high whilst a task is running then low whilst that task is completing then being swapped out and the next task starting. By having two tasks loaded the second task takes up the slack whilst that is happening.

I don’t use it myself but from memory of previous discussions it’s a case of changing the GPU/CPU usage via the app config file for that project.
ID: 106790 · Report as offensive
Profile Keith Myers
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 17 Nov 16
Posts: 863
United States
Message 106791 - Posted: 16 Jan 2022, 17:08:56 UTC - in response to Message 106790.  

Your gpu utilization depends on your gpu application efficiency and how the devs wrote the app. Your sawtooth pattern is indicative of tasks at Milkyway where utilization is typically very low with very low power usage. You can check/verify with nvidia-smi to look at utilization and wattage consumed on the gpus that are running Milkyway Separation tasks.

To get more utilization on Milkyway it is beneficial to run multiple tasks concurrently at Milkyway with <gpu_usage>0.5</gpu_usage> in an app_config.xml file. You can read the docs here:
https://boinc.berkeley.edu/wiki/Client_configuration#Application_configuration

The gpu tasks at Einstein use the gpu more efficiently with higher power usage up to the 90% or 98% completion point depending on the application being Gravity Wave or Gamma Ray Pulsar and then the computations switch to the cpu for completion and the gpus are not being used at all till the next task starts. This also shows a larger frequency gpu sawtooth.

All is normal with your cards and hosts.
ID: 106791 · Report as offensive
AdsoYo

Send message
Joined: 24 Jan 21
Posts: 10
United States
Message 106805 - Posted: 18 Jan 2022, 6:47:10 UTC

Good info thanks guys!

How do I check GPU tasks? I searched the manager high and low and couldn't figure that out.

I'll have to read up on running concurrent tasks. I'm sure my computers can handle it.
ID: 106805 · Report as offensive
Profile Keith Myers
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 17 Nov 16
Posts: 863
United States
Message 106815 - Posted: 18 Jan 2022, 19:57:30 UTC - in response to Message 106805.  
Last modified: 18 Jan 2022, 20:00:07 UTC

By your response I'm guessing you are running in Simple View? You need to switch to Advanced View in the Manager (SHFT-CTRL-A) and then your gpu tasks will be vieweable in the Manager when set to Show Active Tasks.
You would need to write a app_config.xml for Milkyway Separation tasks to switch to running multiples concurrently on each gpu. Typically for Nvidia cards, 2X or 3X is about all you can get away with. But that does bump the utilization much closer to 100% on the card.

This is a simple app_config.xml file for Milkyway Separation tasks at 2X concurrently for example.

<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
</app_config>
ID: 106815 · Report as offensive

Message boards : Questions and problems : Decoding my computer's utilization patterns

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.