Questions and Answers :
GPU applications :
Manually configuring to run only one type of GPU app
Message board moderation
Author | Message |
---|---|
Penguin Send message Joined: 5 Sep 14 Posts: 13 Credit: 16,544,489 RAC: 360 |
I was wondering if it is possible to tell SETI to use only one type of GPU app... It found the NVIDIA SOG app to be fastest on all my machines which is great. But, now I have set it to run more than one instance of SOG on my cards, 2 - 4 depending on the cards. This is getting my better throughput and full use of everything but the side effect is that the tasks take longer and it is going back and rebenching the other application types such as CUDA42 and CUDA50 which are not configured to also run with 2-4 instances. The problem is that it may find these faster and switch to them when they are actually slower throughput. I checked SETI@home preferences and computing preferences and searched the forums and looked at the FAQ but didn't find any answer. Thanks in advance. |
Penguin Send message Joined: 5 Sep 14 Posts: 13 Credit: 16,544,489 RAC: 360 |
What I've done in the meantime is just set all the apps to run the same.. so it should still find SOG fastest... cuda23, cuda32, cuda42, cuda50, SOG |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
Manually configuring? Yes, use the Lunatics app. Do a search for Lunatics on these forums. I think Keith Myers linked to it not too long ago. |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
What I've done in the meantime is just set all the apps to run the same.. so it should still find SOG fastest... cuda23, cuda32, cuda42, cuda50, SOG The only way to lock in what YOU desire is to run the anonymous platform. You define exactly which applications you want to run in an app_info.xml file which the client and scheduler follow. The easiest way to create one is to use the Lunatics optimized app installer. You can download it here: http://mikesworld.eu/download.html Just select the SoG gpu app on the Nvidia card selection page. The default on that page is CUDA50 which you don't want. You still need to use an app_config.xml file to run multiple concurrent tasks on each card. That may or may not produce more output than running singles. Verify. The SoG app likes to have a full cpu thread to support each gpu tasks for best crunching times. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Penguin Send message Joined: 5 Sep 14 Posts: 13 Credit: 16,544,489 RAC: 360 |
What I've done in the meantime is just set all the apps to run the same.. so it should still find SOG fastest... cuda23, cuda32, cuda42, cuda50, SOG OK, thank you both for the speedy response. I won't be messing with an app_info setup right now. I can live with running one task or at least let it finish up benching the other apps now.. I think it will work ok if it benches all the apps with 0.25 gpu or 0.5 GPUs. That's a nice app though and thanks a bunch for that link! And yes, I devote an entire CPU core to each instance of the SoG app on whatever box I'm running things on. |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Maybe I wasn't clear. You don't have to worry about writing the app_info file. The Lunatics installer does it automatically for you based on your choices in the configurator. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13876 Credit: 208,696,464 RAC: 304 |
I think it will work ok if it benches all the apps with 0.25 gpu or 0.5 GPUs.Why? Whith the hardware you have, anything more than 1 WU at a time will result in less work per hour, not more. Making use of some command line values will improve the GPUs output. Grant Darwin NT |
©2025 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.