app_config settings for multiple GPU apps

hadron
hadron
Joined: 27 Jan 23
Posts: 62
Credit: 100919024
RAC: 576326
Topic 231178

Hopefully in the near future I will be adding a second GPU to run two apps, O3AS for certain (being a gravitation theorist) and likely also BRP7.

I need to confirm two things. First, the number of tasks that can run concurrently on a 16GB card; and second, the correct

<gpu_usage> settings to get it all running properly.

I know the O3AS tasks will require 2.5 GB each, so running 4 of these will use 10 GB, leaving 6 GB for BRP7 tasks. All I know about the latter is from a post by Oliver Behnke last Sept: "The BRP7 app is also highly optimized and makes very efficient use of your GPUs. If you had planned to keep your home warm in winter with E@H, the BRP7 app won't let you down" So I need to know the actual vram usage of each task to determine how many will run in that 6 GB.

Next, I need to confirm how the<gpu_usage> setting in app_config.xml works. From the manual description, I believe it is determined by the number of tasks for each app, that is, if there are 4 tasks running for O3AS, then the line for that app should be <gpu_usage>0.25</gpu_usage>, and not some number determined by the total number of tasks of both apps. (I hope I've expressed this clearly.)

 

PS. Not that it really matters, but for the curious, this is the card I'll probably be getting: https://www.newegg.ca/asus-radeon-rx-7600-xt-dual-rx7600xt-o16g/p/N82E16814126700

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6453
Credit: 9580079840
RAC: 7388514

My current experience is

###edit###

My current experience is running all sky GW and brp7 on the same GPU  at the same time slows both down.

Your gpu_usage example is correct. And is task specific not project specific.

This task slowdown experience is specific to Linux and to some versions of the applications.  It might not hold for the standard versions on any OS like windows.

A Radeon VII will probably /actually be more productive for all Sky GW than your proposed GPU. Unfortunately that will not be true of every project and app.

A Nvidia rtx 3080 ti is still the most bang for the buck on most systems. 

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

hadron
hadron
Joined: 27 Jan 23
Posts: 62
Credit: 100919024
RAC: 576326

Tom M wrote: ###edit### My

Tom M wrote:

###edit###

My current experience is running all sky GW and brp7 on the same GPU  at the same time slows both down.

Your gpu_usage example is correct. And is task specific not project specific.

This task slowdown experience is specific to Linux and to some versions of the applications.  It might not hold for the standard versions on any OS like windows.

A Radeon VII will probably /actually be more productive for all Sky GW than your proposed GPU. Unfortunately that will not be true of every project and app.

A Nvidia rtx 3080 ti is still the most bang for the buck on most systems. 

Thanks for your response. However, the Radeon VII was discontinued in Sep. 2019, barely 6 months after it was introduced, while the RTX 3080 TI costs twice as much as I am prepared to spend.

Maybe someone else knows how much VRAM the BRP7 tasks require.

While we're at it, could someone please post the app names for the app_config file?

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4964
Credit: 18742487427
RAC: 7001873

The app_name for BRP7 is

The app_name for BRP7 is just einsteinbinary_BRP7

You can always find the correct app_name in the client_state file for a project

Or just put in some random name for the app in the project app_config and Boinc will correct you with the valid 'known' names for all the apps in the Event Log.

 

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6453
Credit: 9580079840
RAC: 7388514

Have you priced a Radeon VII

Have you priced a Radeon VII on eBay?

And examined the top performing Radeon VIi on the top 50 list? I believe he is in 16th place.

He is currently beating my rtx 3080 ti. And probably will continue to.

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

hadron
hadron
Joined: 27 Jan 23
Posts: 62
Credit: 100919024
RAC: 576326

Keith Myers wrote: The

Keith Myers wrote:

The app_name for BRP7 is just einsteinbinary_BRP7

You can always find the correct app_name in the client_state file for a project

Or just put in some random name for the app in the project app_config and Boinc will correct you with the valid 'known' names for all the apps in the Event Log.

Thanks. Not unexpected, given the name for BRP4G.

And for O3AS?

There is only one client_state.xml file here, and it contains only the apps that are currently running, or in the Boinc job queue.

hadron
hadron
Joined: 27 Jan 23
Posts: 62
Credit: 100919024
RAC: 576326

Tom M wrote: Have you priced

Tom M wrote:

Have you priced a Radeon VII on eBay?

And examined the top performing Radeon VIi on the top 50 list? I believe he is in 16th place.

He is currently beating my rtx 3080 ti. And probably will continue to.

Thank you for your contributions, Tom, but please stop. I am asking questions about getting two sub-projects working together on a single GPU. I am not asking for recommendations about which GPU I should get.

I never mentioned the price of a Radeon VII, because I am not interested. That processor has been out of production for nearly 5 years. I am not, and have never been, prepared to buy used or refurbished hardware of any kind.

San-Fernando-Valley
San-Fernando-Valley
Joined: 16 Mar 16
Posts: 409
Credit: 10204213455
RAC: 23419280

HADRON: maybe this will

HADRON:

maybe this will help to answer your question.

I'm running (for example) 8 BRP7 tasks on one 32GB GPU.

The used VRAM is around 6,200 MB for all eight.

The gpu-usage is have set is "0.121".

 

The name for O3AS is "einstein_O3AS".

 

Note on the side: I always run everything with used GPUs and had never any problems.

They are running (some of them) for at least 4 years and will hopefully run much longer.

 

Happy crunching ...

S-F-V

hadron
hadron
Joined: 27 Jan 23
Posts: 62
Credit: 100919024
RAC: 576326

San-Fernando-Valley

San-Fernando-Valley wrote:

HADRON:

maybe this will help to answer your question.

I'm running (for example) 8 BRP7 tasks on one 32GB GPU.

The used VRAM is around 6,200 MB for all eight.

The gpu-usage is have set is "0.121".

OK, then 4 O3AS plus 4 to 6 BRP7 should be no problem. Thanks.

BTW, where does one find the VRAM usage?

Quote:
The name for O3AS is "einstein_O3AS".

Thanks again. Why they didn't just name them O3AS, BRP7, etc is beyond me.

San-Fernando-Valley
San-Fernando-Valley
Joined: 16 Mar 16
Posts: 409
Credit: 10204213455
RAC: 23419280

The naming convention has a

The naming convention has a long evolution ...

The VRAM info is, for example, shown in the nice tool from TechPowerUp named "GPU-Z".

Welcome.

hadron
hadron
Joined: 27 Jan 23
Posts: 62
Credit: 100919024
RAC: 576326

San-Fernando-Valley

San-Fernando-Valley wrote:

The naming convention has a long evolution ...

The VRAM info is, for example, shown in the nice tool from TechPowerUp named "GPU-Z".

Welcome.

Argh, Windows only. Would you know of a similar Linux tool?  All I could find were some rather uninspiring command line tools that didn't really give any useful information.

Comment viewing options

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