GPU2 client for nVidia cards

[gedon]

the new gpu2 client is crazy!

AMD X2 3600+/1.9GHz
Zotac 8800gt Amp
p5002: 46min -> 3000ppd

tomorrow I'll put this card into my 24/7 rig with a X2 clocked at 2.8GHz

[gedon]

ok, as I said I put this card into my folding rig and played a bit with the gpu2 nv client

the system:

5000+ X2 Black Edition
Abit AN-M2HD
2GB DDR2-800 Cl5
Zotac 8800GT AMP/driver: 177.35 with modded .inf

The GPU2 client loads a whole CPU core, but it seems to be independent from CPU speed. I clocked the CPU up to 3.2GHz and down to 1GHz, both shows nearly the same time per %

3.2GHz 17sec per %
1GHz 17.5sec per %

I decided to downclock and undervolt the processor. A SMP client at the same time cuts the ppd coming from the GPU2 client too much and needs more power

I've got p5001 and p5002 till now, each of them is only worth 98 points, but nearly 50 units a day leads to 4700 ppd and more
Lets see how this theory works in practice. You can follow my stats at eoc, the one GPU2 client is the only one that folds for me.

@ccperf721p: What time per % do you get? Is it a old 8800GTS with 320/640MB or a new 512MB?

[ccperf721p]

It's an older gts 320, 45 seconds per %, it's worth about 2000ppd at stock clocks. I will try clocking it up this evening and see what it does.

[gedon]

Do you use 174.55 or 177.35 driver?

[ccperf721p]

it's a fresh xp install on the 175 drivers

[gedon]

try 177.35, this should improve your output by ~50%

[ccperf721p]

Thanks, I will load them up tonight. I did a little testing with the GPU2 and 5 series smp client running and it actually surprised me that the GPU ran the same and the smp only took a 30 sec per % hit. Might be able to throw a 6K per day average down until I have to unplug my rig to move.

[Samuknow]

Am I missing something here? GPU---folding -----Nvidia????

[ccperf721p]

I thought you knew Sam, Troad posted it up in the smp thread..

second article for download links, it's just sending out testing packets that are good for 98 points but it will do a bunch. Not sure about SLI compatibility or if you can run two clients..

Folding@home

[ThunderRd]

They don't recommend using the GPU client and the CPU SMP client concurrently, not sure of the exact rationale but it isn't stable that way.

Instead, they say it's better to use the GPU client and a standard console CPU client(not SMP).

@cc, gedon; should we create a thread only for the GPU client, like the one we have for SMP? I think it's gonna be important to some, and we can keep the info and experiences in one place. Whaddya think?

@Sam; you've got mail in yer PM box. Didja see it?

[ccperf721p]

I will give running the two together a shot for a day just to see what's what.

I think it would be a good idea to have a GPU thread opened up, probably going to see a lot of people move in to it.

[Samuknow]

I got it to work first time. Had to hack the 177 drivers to install though. Had a good lock up first time with SMP running. Was ok after a restart. I guess we will see what happens. I will try it both in dual view and in SLI to see if it make a difference in performance or stability.

[ccperf721p]

@gedon: Thanks for the 177 tip, almost doubled the production going to those drivers.

[ThunderRd]

Originally Posted by Samuknow

Had a good lock up first time with SMP running.

Yup. From what I'm reading, you can't run the GPU and SMP clients together on a dual-core cpu. The GPU client will use pretty much all of one core, and the SMP has to have both to function.

However, there are workarounds being invented to help users run this combination on Quads. For now, the clients don't use all of the CPU as yet on their own, so it can be done. That being said, it won't last forever, so when they release another client that scales better you will have to adjust/abandon your scheme.

This is the thread you need to do your homework, y'all, so check it out:

Folding Forum; View topic - smp & gpu2 [with a quad core]

[ThunderRd]

OK, no one has done it yet so I will...
Here's a new thread to post all of your experiences with the new GPU2 beta client. I think that many folks with graphics cards supporting CUDA will be interested.

Originally Posted by from Pande Group

We're happy to roll out the beta nVidia client, it's a nice beta, and so here we go...

First and before you do anything else, LOOK OVER THE FAQ: Folding@home - FAQ-NVIDIA
Lots of important things in there that you'll need to know if you do anything beyond run the installer and run it.

Drivers
You MUST use the 174.55 version of the drivers: Download CUDA Code - complete and free toolkit for creating derivative works
(or better, from that URL) for now. You do not need the other files, just the drivers.

Known bugs:
It's a beta, and yes we meant that, not like Google but like software engineers - expect glitches and updates. The main thing is that the viewer will not work, as we're in the middle of changing how the live coordinates get to the viewer. Other then that, visit our nVidia beta Forum to see what's happening.

Download: http://www.stanford.edu/group/pandeg...tray-612b6.msi
6.12 beta 6

Here's a starting place to visit for current information from other users on the community forum:

Folding Forum; View forum - Beta testing the GPU2 Win client (nVidia GPUs)

EDIT: Newer thread here: http://foldingforum.org/viewforum.php?f=49

Here's a list of the minimum nVidia hardware you need:
CUDA-Enabled GPU Products - NVIDIA

@Daniel~, Gizmo, Aedan:
Do you have the awesome power to move the last dozen posts or so [relative to the GPU client]from the other thread ["AOA is losing ground"] into here? There's some good info there already, maybe we can keep it together.

[Aedan]

Of course we can - although, would you mind listing which posts you want combined into a thread? What would you like the thread to be called?

How long before the CUDA client gets out of beta?

[ThunderRd]

Thanks, Aedan. Everything from this post:

AOA is loosing ground

to the end of that thread(about 15 posts) can be moved to this thread that we are in right now, "GPU2 client for nVidia". That would be great. I think this will be a very active thread in the near future.

As for getting out of beta, Pande operates in one of two modes: 1) we're working on it; and 2) It will be ready when it's ready.

They rarely give a timeline for development stuff.

[ccperf721p]

Since moving to the 177 drivers it has really picked up a bunch. I have also been running the smp client along with it, so far no issues. Note Will not work on a dual core as the gpu client will use one core itself and causes strange results. It does however run on quad without much of performance penalty

[Samuknow]

I have noticed the same. I am at this time running SMP and GPU. I can't do it in SLI. It is bad....

But I am not in SLI most of the time.

I feel bad for missing this. I have been buried at work and been doing several side jobs after. I get time to post and am lucky if I actually have a clue what I am posting about. i will spend a little time on this and help Bob and Cliff get this sorted and add it to the front page in the FAH section.

Thanks guys for staying on top of things.

[dabaerman]

certainly looks like a job for the TRI-core cpu's, eh!

Ron