GPU2 Nvidia Linux Folding

[dsio]

Just curious as to who's running this at the moment, and who is running Linux (presumably for SMP) and would like a how-to type thing on running GPU2 clients under Linux.

The GPU2 client for me is running ~4500ppd on an 8800GT, using 5% of 1 cpu core to do so, and you can run Linux SMP, and multiple GPU2 clients at the same (if you have multiple GPUs).

It took me about 5 hours to set up due to it being a Windows app wrapped by wine for Linux, I'm just wondering if anyone would benefit from it.

[Gizmo]

Is the guide that CCPerf did ( http://www.aoaforums.com/forum/nvidi...guide-gpu.html ) no longer useful?

[dsio]

Fedora is a bit different, and since then newer versions of X Server, such as those present in Fedora 12 require much newer versions of the Nvidia drivers to run, which in turn require newer cuda drivers.

Its a similar process, just with some small changes for 2010 and for fedora.

[ThunderRd]

Could we edit Cliff's post, or do you think it would be better to write a new guide?

I'm sure that there are some who could use it. I haven't been able to find the time to get GPU going on my Linux rigs as yet, so I'm in the dark.

[dsio]

Might be worth updating Cliff's post just with the new versions of the drivers and cuda toolkit, but leaving it intact otherwise, because its a very well written guide for Ubuntu, and then doing a separate one for Fedora.

I'll have a go at documenting it in a similar fashion to Cliffs later tonight.

It'd be well worth doing, those 2x 8800gts would be +10k ppd between the pair. Some of the teams that are overtaking us in coming months literally have 1 guy propping up the entire team, with say 50k-70k ppd.

Linux SMP + GPU2 is where its at for competitiveness, if we had some more nvidia gpus on board we'd get a few of those overtakes off our list.

[dsio]

btw, since WZ came back and pushed a few people back into it, we've gained about 30-40k ppd. I had no idea how good these Nv cards were until WZ started cranking 14k.

[ThunderRd]

Originally Posted by dsio

btw, since WZ came back and pushed a few people back into it, we've gained about 30-40k ppd. I had no idea how good these Nv cards were until WZ started cranking 14k.

Some of that boost in ppd is from the newly hatched points system as well. It rewards WUs finishing by the earlier date, so if you run 24/7, you see a gain in ppd...

[gedon]

Hi all,

I'm still folding for AoA and still visiting this forum.

Well, I miss something, something about the new SMP2 Client, about getting a passkey to get bonus points, about the ppd production of the new a3-cores.

I can remember the times when only a few folders of AoA run the nvgpu client and the rest of our team didn't even know it was wearth spending gpu cycles.

dsio, you are talking about 14k a day with several gpus. What about a single SMP2 client with a Phenom 2 X6 and 12K+. I have a P2 X3 @ 3.5GHz doing 5k a day. The faster you have finished a unit, the more points you'll get. That Q9450 in your sig should be a lot faster than my X3.
Oh, and oc'ed I7 quads can do 20k+, no joke!
Use HFM to monitor a3-cores, it can calculate bonuspoints or go here choose wu and insert time per frame

dsio's to-do-list

1. Get a passkey from Stanford click
2. Get the SMP2 Client (ver.6.29)
3. Use -configonly flag to enter passkey, username, team,...
4. No bonus points for the first 10 workunits folded under this passkey
5. wu #11+
6. Feel free to insert passkey into non smp-clients, too.

[ThunderRd]

Make sure you enter the passkey into every client you run. If you run an SMP client and 2 GPU clients on one machine this means enter the key three times, once in each config dialogue.

[dsio]

Yea I've been using the passkey, only just noticed the bonus points though, think my SMP is good for 6k.

I just scored a free 9800GT too, trying to get that up and running for SMP + 2x GPU2.

[ThunderRd]

Sounds about right. My Q6600 runs at 3.6G and gets between 6000 and 6500 ppd, depending on the WU.

[weldzilla]

I have a 285GTX inbound but, I did not use the passkey till I read this thread!!
Lets get the AOA folding hard again.....

[dsio]

Passkey gets you bonus points on your WU's when completed, mainly 6xxx series WUs. Lets say the WU gives 800 points normal, if you have the passkey, they record when the WU was issued, and then record the time it was submitted, based on how fast it was returned, they work out your bonus points.

For me, FAHMon shows 1250ppd off my quad @ 3.6 running 6xxx WUs, but in reality, bonus points are making it closer to 8k ppd. That's how big a difference they make.

AFAIK, its SMP only, not GPU2, and I'm pretty sure that the A3 cores do their best work under Linux SMP, so much so that Windows guys are running F@H SMP inside VMWare Linux environments to get the most of their PPD.

To check what bonus points you will get, try here:

Folding@home SMP2 Bonus Point Calculator

As an example, project 6014 with 5 minutes, 8 seonds per frame, was netting me 8072ppd according to that, and it seams about accurate.

Definitely do it if you're running SMP.

You have to sign up for a passkey and they email it to you.

[weldzilla]

My 285GTX showed up and is on the job as I type!!! I got to tell you though I ran a couple benches with it and it is really no faster than my 260GTX...... On folding it is just a little faster if at all. The good news is I replaced the 250 with my 260 and now that machine is ripping.... and my 250 is now a back up card. I am also using a little more cpu power to help with the folding. Matt says I can use his 8800 ultra a little more also.
So we'll jusat have to see tommorrow afternoon how this all pans out.


PS. My MSI 260GTX for whatever reason is one of the fastest single cards in the world. Just got a golden one.....
Hits right at 300K on AQMK3, 95,000 on 2001se, broke 30,000 on 05!!! 21,000 on 06. Easily clocks up to 810mhz on the gpu. It has just been an amzing card.

OK so now I installed latest driver and overclocked this 285GTX and it is now a monster!!!! About 33% faster than my 260GTX ...... It is cooking!!
It runs a little warmer too... about 57C while folding... I will have to do a little researching to see just where this card can be stabily clocked too.

Come on AOA and FOLD!!!

[dsio]

240 shaders on the 285 vs 216 shaders on the 260 isn't a big difference, the main diff will be the clocks and the ram, but of that, the ram means nothing to F@H GPU2, and the core clock doesn't mean that much either.

From what I've read, the shader clock is the only thing that matters, so if you crank that and leave everything else stock you should be golden.

[weldzilla]

My 260GTX is the older 192 shader model. But it has never met a 216 that it didn't absolutely flatten in the benches. We will see tommorrow if running this 285 helps much. adding the 285 and taking aay the 250. The difference will be right there. I think it was a worth while deal.
The next overtake I have is going to feel good!!

[dsio]

I'm going to have to see about some other SMP machines or that overtake could be me.

[ThunderRd]

Originally Posted by dsio

From what I've read, the shader clock is the only thing that matters

That's correct. In fact, overclocking the core and/or memory can lead to folding instability.

[dsio]

BTW, for shader overclocking, I'm actually dumping the vbios in windows, adjusting the shader clock, flashing my card, and then booting back into linux to get my shader OC how I like it (this is a shortcoming in CoolBits), anyone else using the GPU2 under wine method may want to try that to max their ppd.

[dsio]

Originally Posted by weldzilla

My 260GTX is the older 192 shader model. But it has never met a 216 that it didn't absolutely flatten in the benches. We will see tommorrow if running this 285 helps much. adding the 285 and taking aay the 250. The difference will be right there. I think it was a worth while deal.
The next overtake I have is going to feel good!!

I've got the same problem at the moment, I've got a couple of other GPUs like a 9600GT laying around, but I just don't have the computers to put them in. My mother's Sempron 3100+ is running the 8800GT atm (~5k ppd) because it was the only machine with a spare PCI-E slot.