Thinkpad nVidia Quadro FX 3700M (UNSTABLE_MACHINE) Error

Installed the Tracker on my Thinkpad W700 with a nVidia Quadro FX 3700M, running Vista x64.

The GPU client always fails with an UNSTABLE_MACHINE error

Code:


[08:50:37] Folding@Home GPU Core
[08:50:37] Version 1.31 (Tue Sep 15 10:57:42 PDT 2009)
[08:50:37]
[08:50:37] Compiler  : Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 14.00.50727.762 for 80x86
[08:50:37] Build host: amoeba
[08:50:37] Board Type: Nvidia
[08:50:37] Core      :
[08:50:37] Preparing to commence simulation
[08:50:37] - Assembly optimizations manually forced on.
[08:50:37] - Not checking prior termination.
[08:50:37] - Expanded 65332 -> 344335 (decompressed 527.0 percent)
[08:50:37] Called DecompressByteArray: compressed_data_size=65332 data_size=344335, decompressed_data_size=344335 diff=0
[08:50:37] - Digital signature verified
[08:50:37]
[08:50:37] Project: 5782 (Run 0, Clone 56, Gen 178)
[08:50:37]
[08:50:37] Assembly optimizations on if available.
[08:50:37] Entering M.D.
[08:50:43] Tpr hash work/wudata_01.tpr:  1774752380 867941947 1108098810 1317646640 1161617719
[08:50:43]
[08:50:43] Calling fah_main args: 14 usage=100
[08:50:43]
[08:50:43] mdrun_gpu returned
[08:50:43] Going to send back what have done -- stepsTotalG=0
[08:50:43] Work fraction=0.0000 steps=0.
[08:50:47] logfile size=0 infoLength=0 edr=0 trr=25
[08:50:47] + Opened results file
[08:50:47] - Writing 637 bytes of core data to disk...
[08:50:47] Done: 125 -> 124 (compressed to 99.2 percent)
[08:50:47]  ... Done.
[08:50:47] DeleteFrameFiles: successfully deleted file=work/wudata_01.ckp
[08:50:47]
[08:50:47] Folding@home Core Shutdown: UNSTABLE_MACHINE
[08:50:52] CoreStatus = 7A (122)
[08:50:52] Sending work to server
[08:50:52] Project: 5782 (Run 0, Clone 56, Gen 178)
[08:50:52] - Read packet limit of 540015616... Set to 524286976.


[08:50:52] + Attempting to send results [June 15 08:50:52 UTC]
[08:50:52] + Results successfully sent
[08:50:52] Thank you for your contribution to Folding@Home.
[08:50:56] + -oneunit flag given and have now finished a unit. Exiting.
Folding@Home Client Shutdown.


Tried with and without Force GPU Flag, same problem.

Also searched different forums and blogs without success.

Any idea ?

Have you installed the latest drivers? Tried both GPU2 and or GPU3 clients?
Which version of the tracker are you running?

[EDIT] Can you Download the latest GPU-Z 0.4.3 and post a screen shot of it. I don't think that card supports CUDA and Physx but I could be wrong though.

Here's the GPU-Z screenshot...



I have installed the latest video drivers, they are from December 2009, newer are not available. Only Lenovo drivers can be used on Thinkpads.

I'm using now Tracker 2.89 and tried all combinations (GPU2/GPU3/Force/noForce), always the same error.

Replace the cudart.dll file in the GPU0 folder with the one from this archive:
cudart.zip

This may fix your problem since the included cudart.dll file is for CUDA 2.3 (190.xx and newer drivers) This was done to allow folding on PhysX GPUs without extending the monitor.

Hey great, it's working now with this DLL

Thanks for your help and your great tool !

BTW, I saw you have a PPD of 16-17K for SMP and 15K for each GPU. OK, you have overclocked your CPU and a faster GFX card.

But anyway, your numbers are far above what I get according to FahMon. I have 4K for each GPU (ATI 5870 and nNvidia 3700M), 1K on SMP (i7-960 5 of 8 cores) and 0.5K on SMP (X9100 2 of 2 cores).

I'm using a Passkey and no other processes are really using the CPU.

What did I wrong or did you right ?

yoshi41 wrote:Hey great, it's working now with this DLL

Thanks for your help and your great tool !

BTW, I saw you have a PPD of 16-17K for SMP and 15K for each GPU. OK, you have overclocked your CPU and a faster GFX card.

But anyway, your numbers are far above what I get according to FahMon. I have 4K for each GPU (ATI 5870 and nNvidia 3700M), 1K on SMP (i7-960 5 of 8 cores) and 0.5K on SMP (X9100 2 of 2 cores).

I'm using a Passkey and no other processes are really using the CPU.

What did I wrong or did you right ?

Ok, FahMon does not calculate bonus points for the SMP clients. You are likely getting about 3-4 times the PPD that FahMon is estimating. I recommend HFM.NET for monitoring clients because it calculates the bonus points and PPD in real time for SMP A3 WUs.

My CPU is overclocked to 4.2GHz and the SMP client is set to use 8 cores. My GTX 480s are overclocked to 1550MHz shader, and even at stock they are the fastest single-GPU folding cards by far. ATI cards get way less PPD than comparable Nvidia cards. 4,000 PPD sounds about right for a 5870.

I am so jealous of everyone getting new CPUs, GPUs and such..lol!

About the only thing I have spent money on, this year, is a new 2TB drive and my Technet Subcription (don't build test systems without it..lol).

I haven't even attempted to do an egg wish list, because it will depress me. My next purchase is probably another 2TB drive and SATA addon card.

Also, did you put in a passkey for the SMP client?

Yes, I've setup the passkey. As Jedi said, it's seems to be the FahMon client, which is not able to calculate SMP points. I will switch to HFM.NET.

Thanks guys for your support

Yes HFM, is awesome. I am so glad someone suggested it to me a couple months back. Just make sure you check the options to enable the bonus points view. It is also nice, it will import your FAHMon config, if you want it to.