Welcome to MilkyWay@home

How many tasks on a GPU?

Message boards : Number crunching : How many tasks on a GPU?
Message board moderation

To post messages, you must log in.

AuthorMessage
Scalextrix[Gridcoin]

Send message
Joined: 6 Mar 09
Posts: 1
Credit: 752,214
RAC: 0
Message 61722 - Posted: 17 May 2014, 10:32:12 UTC

Hello all, I have been trying to get the max usage out of my 2 x GPUs as I noticed they were not near to 100% utilisation according to MSI Afterburner. Have been experimenting with cc_config.xml and app_config.xml.

Am sharing what I have discovered, but if you are new to this please don't take my experience as a guide, I am doing this as a total fool and any minute now I expect to receive an ID10T error. Id be interested to hear if anyone else tried and got similar/different results.

Intel i7-3770k @ 4.2Ghz
Nvidia GTX 670 @ stock
Nvidia GTX 780Ti @ stock

I run GPUGRID, Asteriods@Home and of course MW. GPUGRID and A@H didn't seem to work well with multiple WUs on 1 GPU, even though they don't utilise 100% of GPU resource splitting the GPU seemed to cause run times to increase a lot. However MW seemed to divide GPU resources well.
I found my GTX 670 could take 2 x WU at 50% each, Afterburner reporting 96-99% utilisation
I found my GTX 780Ti could take 4 x WU at 25% each, Afterburner reporting a similar 96-99% utilisation

By setting milkyway separation to max 4 tasks and 25% share, and locking them to the GTX780Ti, and having milkyway set to max 4 tasks and 50% share and not locking, both my GPUs seem to be able to crunch at max.

The WUs do run a little slower than if only 1 task is on the GPU, but I 'think' my overall crunching is higher. I don't know how to objectively measure that though, if anyone can tell me Id appreciate it.
ID: 61722 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,950,829
RAC: 21,429
Message 61726 - Posted: 19 May 2014, 11:41:01 UTC - in response to Message 61722.  
Last modified: 19 May 2014, 11:44:44 UTC

Hello all, I have been trying to get the max usage out of my 2 x GPUs as I noticed they were not near to 100% utilisation according to MSI Afterburner. Have been experimenting with cc_config.xml and app_config.xml.

Am sharing what I have discovered, but if you are new to this please don't take my experience as a guide, I am doing this as a total fool and any minute now I expect to receive an ID10T error. Id be interested to hear if anyone else tried and got similar/different results.

Intel i7-3770k @ 4.2Ghz
Nvidia GTX 670 @ stock
Nvidia GTX 780Ti @ stock

I run GPUGRID, Asteriods@Home and of course MW. GPUGRID and A@H didn't seem to work well with multiple WUs on 1 GPU, even though they don't utilise 100% of GPU resource splitting the GPU seemed to cause run times to increase a lot. However MW seemed to divide GPU resources well.
I found my GTX 670 could take 2 x WU at 50% each, Afterburner reporting 96-99% utilisation
I found my GTX 780Ti could take 4 x WU at 25% each, Afterburner reporting a similar 96-99% utilisation

By setting milkyway separation to max 4 tasks and 25% share, and locking them to the GTX780Ti, and having milkyway set to max 4 tasks and 50% share and not locking, both my GPUs seem to be able to crunch at max.

The WUs do run a little slower than if only 1 task is on the GPU, but I 'think' my overall crunching is higher. I don't know how to objectively measure that though, if anyone can tell me Id appreciate it.


The easiest way is to look at your time to complete a unit, both running only 1 unit at a time and then multiple units at a time. At some point you will overload the gpu and it WILL take longer to run x units at the same time then running x units but one at a time. The number x is the number you have to figure our for your system, as it is different then mine or anyone elses. Make sure you ALSO leave at least one cpu core free to keep the gpu fed so you can get the full benefit of using the gpu at near full speed. If you have multiple gpu's then you need to leave one cpu core free for EACH physical gpu.

One thing you don't want to do is have your gpu running at 100% load 100% of the time. The mid to high 90 percentile is plenty as that way you know the gpu itself isn't overloaded. Also make sure you have VERY good airflow to keep the gpu and other fans spinning and cool. If you haven't yet gotten the small file gpu-z then get it as it is an easy way to monitor you gpu. You can get the latest from here, it is free!
http://www.guru3d.com/files_details/gpu_z_download_techpowerup.html

The site also has plenty of other good, FREE, utility type programs that could be worth your while to browse thru. Another file I have started using lately is openhardwaremonitor and you can get it from them too here:
http://www.guru3d.com/files_details/open_hardware_monitor_2_1_beta.html

It will open a box on the screen showing you both the current and max temps of your cpu's, gpu's and drives etc. It is NOT as graphical as gpu-z is though. As I said BOTH are free. Both will also show you the current load on the gpu, which is what you are trying to accomplish by running more units at once. Be careful NOT to let the gpu temps get too high, into the mid and upper 90'sC is a BAD thing for the long term health of your gpu. Even the low 90'sC means there is room for improving the air flow.
ID: 61726 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : How many tasks on a GPU?

©2024 Astroinformatics Group