Can't get GPU2 working

[Gizmo]

I've tried everything I can find. I just CANNOT get GPU Folding working on this machine.

The 195.36.24 drivers with CUDA 2.3 were telling me that they couldn't find a CUDA device. The 256.25 drivers start to process for about 1 second and then throw an exception ("exception thrown during GuardedRun").

[ThunderRd]

Two things come to mind, and one of them you won't like.

guarded run/unstable machine errors sometimes are from a faulty WU, so you could try to get another RCG by dumping the work folder along with the queue.dat and the client.cfg files. The client will run the configuration routine after this, when you start it up again. This should get you another [different] WU to try.

Since the 256.x drivers seem to start up and fail, the other thing is to check the memory on that card. See my post here:
http://www.aoaforums.com/forum/nvidi...dia-cards.html

People don't believe that it can be a memory error, but it's well-documented. I know for a fact it can be true; I had to exchange one of my 9600GTs after I'd had it for two days for a similar problem. Memtest failed the memory after about an hour.

BTW, is this a Windows box or are you running the client in WINE? What version of 'nix/Windows? Are you running an SMP client simultaneously?

[Gizmo]

Wine on Fedora 13, and yes, I have the SMP client running also.

I ran the memtestg80, but only for about 5 iterations. Based on your comment, I'll try running it overnight and see what I get.

edit: huh, now I can't run memtestg80; it tells me it can't allocate 128 MiB on a 512 MiB card. I wonder if it has anything to do with resuming from suspend?

[ThunderRd]

Have you tried it without the SMP client running? OTOH, I guess it shouldn't matter if it's in a VM.

[Gizmo]

Ok, I think I've got it running.....at least if this means what I think it means:

[03:05:18] Working on Protein
[03:05:21] Client config found, loading data.
[03:05:21] Starting GUI Server

[Gizmo]

Wow

Video response sure goes to hell......

[Gizmo]

Getting about 7:10 per frame, so about 12 hours on Project 6602 (Run 9, Clone 244, Gen 104), should work out to about 900 ppd.

Got my CPU running 81C, and my GPU hitting 78C.

Basically, I had to install CUDA 2.3 along with a newer cudart.dll.so and the nVidia 256.25 drivers. Apparently the 195.36.24 drivers simply don't work properly with CUDA 2.3 (although they might work with CUDA 2.0, 2.1, or 2.2, didn't try those).

There was a bit more to it than that, but that's the gist of it.

[ThunderRd]

What are the specs/cooling of that machine. Sounds hot. Also that 900 ppd sounds suspect on anything =>8800. I've never heard of only 900 ppd from any card, I would think it would be in the thousands of ppd, depending on which one. My 9600GTs do nearly 4000 ppd each.

I'm thinking there's a problem with the installation and/or hardware. Video grinding to a halt is not normal either.

[Gizmo]

It's a Quadro FX 570M (G84M core) running on my lappy. The drivers report 32 CUDA cores.

Temps are about right, actually, given that it's a laptop. However, I think the strain of everything running is just a bit too much; it errored out again at 23%.

[ThunderRd]

In the configuration dialog of the GPU3 client, did you:

a. set the priority to "slightly higher" [this is important if you are running a CPU client simultaneously]

b. check the box "do not lock cores to a specific CPU"

I'd be curious if the GPU client would run successfully without the CPU client running.

[Gizmo]

What configuration dialog?

I'm running the console client on Fedora 13 using Wine. That's the only way to run it on Linux AFAIK.

I changed the priority setting as you suggested, and that seems to have made it a lot nicer. I also set the desired CPU usage to 5% instead of 100% (even though it was using less than 5% already).

[ThunderRd]

In the systray client that I run, there is a config page. You enter your mane and team info, and your passkey. On another page you set proxy information and on the last page there are some config options, like priority and CPU usage, as well as the lock cores option. I would assume that in the client you run there is a similar place to set options. I haven't run the console client so I don't know where, but it sounds as if you are in the right place.

I'm guessing that you would get there with the -configonly argument on the CL.

[Gizmo]

dsio seems to think that I should switch to the 256.29 drivers, as the '25 drivers apparently have a bug causing them to try to run in power conserve mode all the time. Apparently it affects folding as well as video playback, so I'm going to give those a try.

[Gizmo]

Returned my first full GPU2 WU today. I'm working on my second unit now.....

[ThunderRd]

Good to see it's running, but I can't wait to see what your 8-core box will do. It must be faster than i7 with 8 actual cores rather than hyperthreading.