GPUs efficiency with BOINC

Message boards : GPUs : GPUs efficiency with BOINC
Message board moderation

To post messages, you must log in.

AuthorMessage
Urk_Alter

Send message
Joined: 15 Jul 11
Posts: 5
Message 38957 - Posted: 15 Jul 2011, 9:33:13 UTC

Hi all,

short time, first post.

I was used to do some so-called "folding", with GPUs computing.
With reference to Stanford F@H project, there are lots of information about the various GPUs efficiency (with AMD ones trailing by far), which often help to properly seize systems and resources. Even several computer-oriented websites offer a crop of comparison, such as http://www.bit-tech.net/hardware/graphics/2011/04/26/best-graphics-card-for-folding/1 (it's just one of the more recent).

As far as I know (and so, nearly nothing), such detailed & quantitative information seem to lack more than a bit with reference to BOINC.

So I'm asking to you: AFAYK, which are the most efficient GPUs currently for BOINC projects? Is it possible to point out sort of "all-arounder" GPUs which are able to cope the vast majority of current BOINC projects?

Thanks in advance for any hint,

best regards,
Luca

P.S.: sorry for my english (I come from Italy)
ID: 38957 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15477
Netherlands
Message 38958 - Posted: 15 Jul 2011, 10:09:29 UTC - in response to Message 38957.  

Well, part of the problem here is that BOINC comprises lots of different projects, which all have their separate own needs. Including to what hardware and operating systems they like best.

Some like ATI best, others Nvidia.
Most want the newest possible GPUs, others don't really mind if the card is a couple of years old.
Some want huge amounts of onboard video, others do not.

I don't have the time (or interest) to go scourge the various projects that do use GPUs to go ask them what their best coprocessor is that can be used. I did start a FAQ a while ago, but it needs (constant) updating, again I don't have the time to do so. It also only shows generic data, not specifics like which brand/model is better than which other brand/model.
ID: 38958 · Report as offensive
Urk_Alter

Send message
Joined: 15 Jul 11
Posts: 5
Message 38960 - Posted: 15 Jul 2011, 15:26:56 UTC - in response to Message 38958.  

Thanks for your answer, ageless

I don't know about BOINC, but the F@H computing is a very energy intensive task (lots of electrity, huge waste of heat, lots of noise), so in these days computing efficiency matters a lot: with reference to that, I mean it is actually unfortunate that some basic but comprehensive comparison cannot be performed, with the Berkeley architecture (even on a project-by-project basis).

Best regards,
Luca
ID: 38960 · Report as offensive
Urk_Alter

Send message
Joined: 15 Jul 11
Posts: 5
Message 38979 - Posted: 16 Jul 2011, 8:48:09 UTC - in response to Message 38962.  

The kind of information you seek doesn't exist for BOINC and it never will, not even on a project-by-project basis. It's impossible to compare GPUs across all BOINC projects because each project measures GPU output differently so a credit (point) at one project isn't equivalent to a credit at another project.

If mankind had ever thought such way, maybe we would still have sort of squared wheels.
I'm sorry, but I couldn't resist. ;-)

As for a project-by-project comparison, you would need to bring together a power meter with each GPU model. Since the GPUs are in the hands of the crunchers rather than the project, you would have to convince a number of different crunchers to buy a power meter. Or you could ask the project to obtain all the different GPUs. I doubt either scenario will ever happen.

As you may have seen with reference to the F@H project, it isn't Stanford nor some volunteer crunchers to spread out quantitative somewhat reliable comparisons (about GPU computing efficiency), but usually web reviewers all over the world do so (at least recently).

Just as an idea for future achievements, I think that it wouldn't be totally unreasonable if someone inside BOINC project would look for a little collaborative effort with some of those very same web reviewers, in order to test even some of the main BOINC clients/projects, such as Seti@Home, along with F@H one: as well known, it's just what already happens in the computer games scene, where the web reviewers usually test GPUs against several games, to give to their readers a better whole picture.

If you want power efficiency then go with the latest chip designs. They run on lower voltage which means less transistor leakage which means greater power efficiency.

It's just not so simple, I mean, but thanks for your answer.

Best regards,
Luca
ID: 38979 · Report as offensive
Urk_Alter

Send message
Joined: 15 Jul 11
Posts: 5
Message 39007 - Posted: 17 Jul 2011, 8:18:29 UTC - in response to Message 38979.  

I think that it wouldn't be totally unreasonable if someone inside BOINC project would look for a little collaborative effort with some of those very same web reviewers, in order to test even some of the main BOINC clients/projects, such as Seti@Home, along with F@H one

Just as a reference: nowadays it isn't uncommon to see web reviewers benchmarking some OpenCL application for GPGPU computing, as there: http://goo.gl/Liqgv (this shortened link is to a google automated translation of a german site).

Best regards,
Luca
ID: 39007 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15477
Netherlands
Message 39013 - Posted: 17 Jul 2011, 13:25:52 UTC - in response to Message 39007.  

So why don't you ask at the (web-)magazines for them to do a corresponding test of BOINC with GPUs?
ID: 39013 · Report as offensive
Urk_Alter

Send message
Joined: 15 Jul 11
Posts: 5
Message 39067 - Posted: 18 Jul 2011, 13:33:53 UTC - in response to Message 39013.  

So why don't you ask at the (web-)magazines for them to do a corresponding test of BOINC with GPUs?

Mainly cause I've not any role inside BOINC.
But it's an idea: I'll send out a couple of mail, maybe I will be lucky, thanks.

Best regards,
Luca
ID: 39067 · Report as offensive

Message boards : GPUs : GPUs efficiency with BOINC

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.