GPU client on Quadro FX580

Moderators: slegrand, Site Moderators, PandeGroup

GPU client on Quadro FX580

Postby Picky88 » Wed Mar 10, 2010 2:22 pm

Hi there,

I am trying to get folding@home working on my graphics card, a Nvidia Quadro FX580. I have installed the CUDA toolkit, and have the latest nvidia drivers. When I start the install for the cuda driver, it says that I am trying to install an older version than the one I currently have. Should I go ahead and install the older version? Currently When I start the gpu folding program I get a message saying "At present your gpu is not supported or you need a current driver."

I am running Windows XP Pro 32bit.

Can anyone help?

Thanks,

Picky
Picky88
 
Posts: 6
Joined: Wed Mar 10, 2010 2:18 pm

Re: GPU client on Quadro FX580

Postby bretth603 » Wed Mar 10, 2010 2:34 pm

Well, I'm no expert, but I'll tell you what I think I know.

You don't need the CUDA toolkit, only the latest driver. The driver includes CUDA support. You can use the Nvidia Control Panel to check the driver version you have installed (software version). I wouldn't install an older driver. There have been changes and bug fixes for the Quadro cards in the newer drivers.

You will need to use the "-forcegpu nvidia_g80" flag to fool the folding@home client. This basically forces the client to think you have a g80 chipset (which you don't). It will get rid of that message when you start f@h.
bretth603
 
Posts: 43
Joined: Wed Nov 18, 2009 4:32 pm

Re: GPU client on Quadro FX580

Postby Picky88 » Wed Mar 10, 2010 4:50 pm

Adding the extra forcegpu command has sorted it. Many thanks!
Picky88
 
Posts: 6
Joined: Wed Mar 10, 2010 2:18 pm

Re: GPU client on Quadro FX580

Postby Picky88 » Wed Mar 10, 2010 6:27 pm

Another question about using the SMP client and the GPU2 client. I am currently running both in windows xp, but the SMP client is significantly faster when run in a linux environment. But as I understand it the GPU2 code is only compiled for windows, so to run SMP and GPU2 in linux involves setting up WINE. Which setup would give overall better performance if I want to run both SMP and GPU2? Linux or Windows? How buggy is the GPU2 client running in WINE?
Picky88
 
Posts: 6
Joined: Wed Mar 10, 2010 2:18 pm

Re: GPU client on Quadro FX580

Postby bruce » Wed Mar 10, 2010 7:20 pm

SMP is essentially the same for Windows, for Linux, and for MacOS. Windows has been restricted to FahCore_a1 while Linux/MacOS have been able to use FahCore_a2 which has been producing points at a higher rate. Recently FahCore_a3 has become available and it runs on all three platforms. Individual projects are for a specific FahCore, so during the transition phase, it's best to be able to be able to process all types, but most important to enable your systems for FahCore_a3 by upgrading to the v6.29 client and getting a passkey. Currently the only exception is for those who have large systems (8 or more cpu-cores, least 0.5 Gb of RAM available per folding core, etc.) where Linux is still required, but that's going to change too.

The GPU2 client/core are only available for WIndows though there's 3rd party support for WINE.

If you're being assigned SMP2 projects (FahCore_a3), adding a GPU2 client is a questionable decision since it add very little to none. If you're still running SMP projects (FahCore_a1 / _a2) adding a GPU2 client is probably worth the effort. Whatever you decide, things are more likely to change than to stay the same, so you'll need to reevaluate periodically.
bruce
 
Posts: 22729
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: GPU client on Quadro FX580

Postby Picky88 » Wed Mar 10, 2010 11:17 pm

Thanks for your reply. I have created a passkey and added to the core using the -configonly flag, so I should now be eligible to receive FahCore_a1, 2 or 3 work units?

Forgive my lack of knowledge, but how can running fah on my gpu be almost not worth doing? does it use a significant amount of CPU time aswell?
Picky88
 
Posts: 6
Joined: Wed Mar 10, 2010 2:18 pm

Re: GPU client on Quadro FX580

Postby Hyperlife » Wed Mar 10, 2010 11:19 pm

Picky88 wrote:How buggy is the GPU2 client running in WINE?

It's stable, for the most part. Every once in a while I see a WU hang midway through folding, but restarting the client always fixes it. Check out this thread about the hanging problem.
Image
User avatar
Hyperlife
 
Posts: 438
Joined: Sun Dec 02, 2007 7:38 am

Re: GPU client on Quadro FX580

Postby bruce » Thu Mar 11, 2010 6:37 am

Picky88 wrote:Thanks for your reply. I have created a passkey and added to the core using the -configonly flag, so I should now be eligible to receive FahCore_a1, 2 or 3 work units?

Forgive my lack of knowledge, but how can running fah on my gpu be almost not worth doing? does it use a significant amount of CPU time aswell?


The answer depends on two factors, which brand of GPU you have and which type of SMP assignment you get.

The drivers required to drive a NVidia GPU have optimized to use a very small amount of CPU time. The ATI drivers generally consume quite a bit more CPU time although it can be reduced quite significantly with a trial-and-error setting of some Environment Variables.

The FahCores needed by the older SMP (specifically FahCore_a1 and _a2) work very well with the NVidia GPU client and moderately well with the ATI client.

The FahCore used by SMP2 (specifically FahCore_a3) does not like to be interrupted. When another task interrupts SMP2, the througput is reduced quite drastically even if the interruptions only use small quantities of time. Running a GPU client simultaneously will add some points for the work completed by the GPU, but if the interruptions reduce the throughput of SMP2 by as much as the GPU adds, there is no net benefit.

Nobody can predict how this will change. Perhaps the upcoming release of a new GPU client will reduce the demands on the CPU. Perhaps SMP2 will become less susceptible to interruptions. -- or perhaps neither of these can be accomplished. These software packages are either not yet available yet or very new and still being developed, so I cannot predict what may be possible in the future.
bruce
 
Posts: 22729
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.


Return to NVIDIA specific issues

Who is online

Users browsing this forum: No registered users and 2 guests

cron