GPU Folding: Dell XPS 1340 (Hybrid SLI)

[muhon_zen]

Hi there dear folders,

I just got back into the folding hype yay! I'm currently using a Dell Studio XPS 13 (Codename XPS1340) with a nVidia GeForce 9500MG (9200MGS + 9400MG) in Hybrid SLI. I was wondering if there is a way to fold using both the GPUs, since I tried the method described in the Official Team 32 Install Guide and the 2nd GPU (9400MG) throws an error message indicating that the GPU is not supported.

As for now, I'm only folding using the 9200MGS. BTW, the computer becomes really slow when folding with GPU, and I don't get why this is happening. It isn't CPU folding, and I'm only using Firefox + MSN (switching between windows, and even writing this post... characters are appearing with some lag). Is it normal, or maybe the nVidia version of the GPU Fold client is faulty?

Finally, just a question regarding multi-fold. Is it possible to fold using both CPU and GPU? I tried doing that but the CPU client stuck in "Attempting to get work packet" and didn't start.

Thanks a lot guys!

[chuckerants]

#1. Yes, you can Fold with both GPU and CPU. That's exactly what I'm doing.

As for Folding with the GPUs that you have, first are you sure your GPUs can Fold at all? What I mean is that I bought a 9500GT thinking it was a cheap way to start GPU Folding. But, it turned out the 9500GT (at least the one I have) was not very stable at all and would only complete around 1 in 5 WUs.

So, first figure out if your GPU can Fold at all before trying to Fold in SLI.

[muhon_zen]

Thanks for answering so fast! Well, that is what I'm planning to do... However, everywhere I read it seems that VMWare with Linux is better than Windows... should I get into that then? (Is there a difference using VMWare and VirtualBox?).

Regarding the GPUs... if I disable the 9200MGS, the 9400 does fold. It seems each of them is folding independently, the problem is joining them together.

[chuckerants]

Since they both Fold independently, someone more knowledgeable than me will have to confirm if different GPUs can Fold together as SMP.

Worst case, you can Fold both together as different Machine IDs, or two instances of FAH.

What's the estimated PPD on the two GPUs you have?

To give you a frame of reference, my single Asus 9600GSO is getting around 3500~4000 PPD and my VMWare/Ubuntu FAH SMP is around 1200 PPD.

[ChasR]

I don't think either the 9200m (8 sp) or 9400m (16 sp) will make the deadlines on 1888 point WUs. You have to be under 43:12/frame(1 %) to make the preferred deadline on those WUs. If you can't make the prefered deadlines, you shouldn't fold on the gpus.

THe cpu will produce a lot more ppd than your gpus anyway. Which processor do you have, p8600?

[muhon_zen]

Thanks. Oh, but still, wouldn't they get lower point WUs? I have a P8700 btw.

[ChasR]

Right now the mix of WUs is predominantly 353 and 787 point WUs that you will probably be able to finish if the 9200 and 9400 will fold stably. Stanford will send out whatever it needs folded and you will get 1888 point WUs at the same rate as every one else. There is no avoiding them.
On the cpu, vmplayer with notfreds SMP virtual appliance is the easiest to install. You should get over 1500 ppd. Ubuntu is more useful as a guest if you are inclined to use Linux.

[Sydney]

Did you setup the second client with the -gpu 1 flag and set machineid to 3?

You might have to enable a second dummy monitor somehow if you can, windows doesn't like using 2 GPUs when you only have one monitor if they aren't configured for physx or SLi.

[muhon_zen]

Thanks for the replies! Well I did install successfully VirtualBox with Ubuntu x64 and 2x folding clients. Regarding the GPU issue, I did setup with -gpu 1, but machine to 2. Is there a difference? My nVidia video cards are configured with PhysX and hybridSLI mode.

[Sydney]

Each GPU client needs to have a unique machineid number, so set one GPU to machineid 2 and one GPU to machineid 3.

[ChasR]

Production with virtual box is lower than with VMware and far more prone to instability.

[muhon_zen]

Thanks, I just did that. It worked with the dummy Display (otherwise it wouldn't start.) However, both displays show the same Hardware (GeForce 9200M GS) and neither show the GeForce 9400M. Here I attach a screenshot, I hope it can be seen.

http://i38.photobucket.com/albums/e133/muhon_zen/Overclocking/GPUFolding.png

Also, it seems that somehow GPU1 (9400M) is screweing up or mixing with the other one... The log stablishes the following:


[17:27:59] *------------------------------*
[17:27:59] Folding@Home GPU Core - Beta
[17:27:59] Version 1.19 (Mon Nov 3 09:34:13 PST 2008)
[17:27:59]
[17:27:59] Compiler : Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 14.00.50727.762 for 80x86
[17:27:59] Build host: amoeba
[17:27:59] Board Type: Nvidia
[17:27:59] Core :
[17:27:59] Preparing to commence simulation
[17:27:59] - Looking at optimizations...
[17:27:59] - Created dyn
[17:27:59] - Files status OK
[17:27:59] - Expanded 46734 -> 252912 (decompressed 541.1 percent)
[17:27:59] Called DecompressByteArray: compressed_data_size=46734 data_size=252912, decompressed_data_size=252912 diff=0
[17:27:59] - Digital signature verified
[17:27:59]
[17:27:59] Project: 5766 (Run 0, Clone 163, Gen 1035)
[17:27:59]
[17:27:59] Assembly optimizations on if available.
[17:27:59] Entering M.D.
[17:28:06] mdrun_gpu returned
[17:28:06] Going to send back what have done -- stepsTotalG=0
[17:28:06] Work fraction=0.0000 steps=0.
[17:28:10] logfile size=0 infoLength=0 edr=0 trr=25
[17:28:10] - Writing 637 bytes of core data to disk...
[17:28:10] Done: 125 -> 124 (compressed to 99.2 percent)
[17:28:10] ... Done.
[17:28:12]
[17:28:12] Folding@home Core Shutdown: UNSTABLE_MACHINE
[17:28:15] CoreStatus = 7A (122)
[17:28:15] Sending work to server
[17:28:15] Project: 5766 (Run 0, Clone 163, Gen 1035)


[17:28:15] + Attempting to send results [October 10 17:28:15 UTC]
[17:28:16] + Results successfully sent
[17:28:16] Thank you for your contribution to Folding@Home.
[17:28:20] EUE limit exceeded. Pausing 24 hours


While GPU0 (9200) gives the following log:


[17:25:35] *------------------------------*
[17:25:35] Folding@Home GPU Core - Beta
[17:25:35] Version 1.19 (Mon Nov 3 09:34:13 PST 2008)
[17:25:35]
[17:25:35] Compiler : Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 14.00.50727.762 for 80x86
[17:25:35] Build host: amoeba
[17:25:35] Board Type: Nvidia
[17:25:35] Core :
[17:25:35] Preparing to commence simulation
[17:25:35] - Looking at optimizations...
[17:25:35] - Created dyn
[17:25:35] - Files status OK
[17:25:35] - Expanded 46633 -> 252912 (decompressed 542.3 percent)
[17:25:35] Called DecompressByteArray: compressed_data_size=46633 data_size=252912, decompressed_data_size=252912 diff=0
[17:25:35] - Digital signature verified
[17:25:35]
[17:25:35] Project: 5768 (Run 10, Clone 248, Gen 480)
[17:25:35]
[17:25:35] Assembly optimizations on if available.
[17:25:35] Entering M.D.
[17:25:41] Working on Protein
[17:25:42] Client config found, loading data.
[17:25:42] Starting GUI Server


What can this be? Am I configuring something wrong? (Do I need the -local flag also?)

[Sydney]

Is SLi disabled? If you haven't disabled it that could be the cause for this problem.

Also, in the display / nvidia control panel, does it say the 9400 is governing the secondary display? Because if the 9200 is running both displays, you wouldn't be able to fold with the 9400.

[ChasR]

You need to disable SLI and extend the desktop on the the second card. Don't be surprised if neither card will fold stably. Some 32 sp cards won't fold.

[muhon_zen]

Interesting... I thought the SLI would help the uber folding. It seems it's better without folding then. How do I disable SLI then?

[ChasR]

nvidia control panel, 3D settings.

[muhon_zen]

Okay, I disabled the SLI (MultiDisplay/mixed-GPU acceleration ->setting Single Display Performance Mode). PhysX is enabled, but it still doesn't work. How do I make each card handle each display separately? I think the 9200MGS is managing both displays at the moment.

[Sydney]

Check the nvidia control panel for displays, the video adapter handling the display should be listed behind the name of the display. If you can only have the 9200 as the primary display, and you can only connect a dummy display to the 9200, I guess you are out of luck :(

Actually, you don't need dummy displays under Linux, you should be able to fold with both cards there. If you want to give the penguin a try, I've written a detailed howto, link is in my sig.

[muhon_zen]

I think I'mma try linux in native form. Both Windows GPU and VirtualBox SMP Linux are lagging me like hell with other processes. I guess VMWare should help, but I'm kinda staying in the non-pirating side at the moment, and i don't know if vmplayer is free or not. However, it seems that folding in Linux Native would give me the best performance right?

The thing is, Linux is still having problems with Hybrid SLI and it only detects my 9400MG (and not hte 9200MGS). Do you know anyonw who has successfully confiured Hybrid SLI in Linux?

[ChasR]

Vmware Player and Server are both freeware.

Higher end cards will fold with SLI enabled on drivers 190.38 and up. However, lower end cards tend to barf on FAH GPU on those drivers with or without SLI.

Sydney and I probably don't agree, but it's my opinion that Windows/VMware/Linux SMP + GPU is a bit faster than native Linux SMP + GPU via wine. Whichever of us is correct, the difference isn't very large. In your case, I don't believe you're ever going to be able to fold on your current hardware. Native Linux SMP is definately faster (5%) than VM/Linux SMP

[muhon_zen]

Oh thanks for the tip on VMWare!

Well, I kinda figured that out as well, my GPU isn't made for heavy processing.. it's a laptop GPU (good one for laptop though). However, folding with SMP in Linux will work for my current hardware right? (Only CPUs).

[ChasR]

Yes it'll work fine. How much do you run your laptop? THe SMP client is a high performance client intended to be run 24/7. Preferred deadlines are 3 days on most WUs. You'll have to run at least 10.5 hours/day to make the deadlines.

[muhon_zen]

I run my laptop around 15-17 hours a day. So there's no way to fold below those 10 hours a day (No SMP, but perhaps 2 independent cores?)

[ChasR]

??
You have to fold more than 10.5 hours per day. Preferably 24/7.