Running multiple tasks concurrently per GPU - How to revert to just one task.
Language
Copyright © 2023 Einstein@Home. All rights reserved.
Copyright © 2023 Einstein@Home. All rights reserved.
Is it possible to run more
)
Is it possible to run more than one of O2AS20-500 GPU wu's simultaneously? It looks like my RTX2070 can do alot more than it's doing now. It's currently crunching with 0.9 cpu's + 1 Nvidia GPU, but it seems to be very slow compared to FGRPopenclTV-nvidia wu's which it did two at the same time, very effortlessly.
Eskomorko wrote:Is it
)
Sure there is (but posting in a thread about how to not run more than one seems a bit odd
)
Go to your Account -> Preferences -> Project and scroll down to "GPU utilization factor of GW apps:" change that to something less than 1.0 (e.g. 0.5 = 2 task at a time, 0.33 = 3 task at a time) save and then make Boinc download at least one task for O2AS20-500 for the GPU. A normal contact (click on update in Boinc Manager) won't get the setting downloaded, you need to download new work for it to apply.
Or use an app_config.xml to fine tune the settings, but this options has some drawbacks if you want to revert to not using it.
Well, i put it down here
)
Well, i put it down here because my problem was GPU related. Thanks for the tip anyway!
Hi everyone I´d know that
)
Hi everyone
I´d know that maybe I´d sound like some sort of "Rookie style" but I would like to have a clear and simple method to run only 2 tasks at the same time per pc.
Does anyone know how to do this? It seems to tech for my little knowledge and I would like to have a clear "step by step" path.
If so, could you write me a private message so that I can fix it up on a clear way?
Thanks for your help.
Need more information. How
)
Need more information. How many gpus are in the host?
You can control the number of tasks per gpu directly in your Account Project preferences.
https://einsteinathome.org/account/prefs/project
At the bottom of the page are toggles for the mulitplicity of how many tasks per gpu are run concurrently for each gpu application.
The default is 1.00 but if you have changed that, simply revert to 1.00 and Save the Preference.
If you want to run two tasks per gpu, change the value to 0.50. To run 3 tasks per gpu, change to 0.33.
The setting is explained right there in the page for the configuration options.
If you have run a app_config.xml file for Einstein work, you will have to change that on your host locally.
If you only have one low end gpu you should be only running one gpu task at a time. If you have a more high-end gpu, you can run 2 concurrently. The AMD cards do better with higher counts of multiplicity since they use very little cpu to support a task. Due to the nature of the Nvidia OpenCL implementation, they need much more cpu support for each gpu task. You need to experiment to see what is the optimum configuration.
If you are getting too much work, reduce your days of cache in Preferences.
Miguel Esteban Pérez
)
You need to mention your hardware details and the types of tasks you wish to run. Your computers are 'hidden' so we can't see any of these details.
Since your RAC (recent average credit) is fairly small, I suspect you might just be running CPU tasks. If so, your easiest solution is to change the percentage of cores that BOINC is allowed to use. For example, if your PC has 8 cores, you would set that value to 25% and then BOINC would only allow two CPU tasks to run concurrently.
You can change these settings on the website (account -> preferences -> computing -> advanced settings) and scroll down to find the setting for "Use at most: X % of the processors". (Don't forget to save any changes.)
Alternatively, you could make the same change locally using BOINC Manager, Advanced view. Local changes override website settings.
If what you need is more complicated, please give more details. You could also consider making your computers 'visible' so that others could more easily offer advice without you having to mention everything. (Account -> preferences -> privacy).
Cheers,
Gary.
Greetings from the first of
)
Greetings from the first of many refugees from Milky Way, which has just concluded whatever study was leveraging CUDA. I imagine many of us will be looking for a new GPU project.
On that note, I take it this is the thread where GPU concurrency would be discussed despite its title.... Has anyone figured out the optimal GPU task concurrency for Einstein running on an NVidia Titan V? Thanks, BGU
A fellow teammate, who is
)
A fellow teammate, who is also a refugee from Milkway brought his Titans here.
Sticking with the usual 2X concurrency on FGRPB1G Gamma-Ray tasks.
Host 13109747
Gary, I too am running an
)
Gary,
I too am running an older NVIDIA card a GTX 1060 6GB.
Right now running singles a Gamma-ray pulsar binary search #1 on GPUs v1.28 () windows_x86_64 shows a Run Time of 768 with a CPU time of 734.
Would my GPU be capable of 2 at a time or would that not be recommended ?
Thanks
Bill
In October of 1969 I took an oath to support and defend the Constitution of the United States against all enemies, foreign and domestic;
There was no expiration date.
You can do 2X on your 1060
)
You can do 2X on your 1060 6GB just fine. Running 2X GR#1 tasks only uses around 3725MB of card memory.
You will have to look at your 2X times per task, divide by two and see whether that is smaller than your 768 second average 1X time to determine whether 2X is beneficial for production.
Will certainly increase the utilization factor of the card at 2X compared to 1X.