Tesla C1060 + Boinc = Insta-error?

rflcptr

Supreme [H]ardness
Joined
Mar 27, 2008
Messages
6,900
First off, thanks for anyone's time with this. :)

I've had the chance at school to temporarily try some Boinc workloads on this card, but haven't any success, yet. All of the GPU WUs immediately fail. I've tried Collatz, MilkyWay, and AQUA. SETI's been down all day, so that wasn't an option.

Specs:
- Xeon E5540
- 4 gigs memory
- Tesla C1060

This is under Boinc ver 6.10.58, running on Win XP 32-bit. Also running the latest Tesla driver (256.98, with CUDA 3.1 support):
http://www.nvidia.com/object/tesla-winxp-x32-258.96-whql-driver.html

So far, resetting or re-attaching the projects haven't helped. I'm running the WUs as protected apps, which should only affect older, incompatible ones' ability to render graphics.

CPU WUs aren't affected under any project, so I'm stumped. :confused:
 
Last edited:
Dont know muc about your boinc problem, but if you can we are in a tight battle with EVGA for the #1 spot on F@H. If can help it would be greatly appreciated by all around here.
 
rflcptr - Check your BOINC logs (Messages tab) - Does it recognize a CUDA device? It would be at the very beginning of the log.

Code:
9/23/2010 9:50:18 PM        NVIDIA GPU 0: GeForce 8800 GS (driver version 19745, CUDA version 3000, compute capability 1.1, 368MB, 264 GFLOPS peak)
 
Well i have read about a couple different people using them for SETI so I would think they should work fine.
http://setiathome.berkeley.edu/forum_thread.php?id=52632

I wonder if a slightly older driver might work?

I'll assume that the Tesla card is the only graphics card in the system? Or are their other CUDA enabled cards?
 
Last edited:
I'll throw on an earlier driver in the morning, thanks. There's another card (some small Quadro for the display), but it's not CUDA enabled afaik. I'll also check to be sure. :)
 
I'll throw on an earlier driver in the morning, thanks. There's another card (some small Quadro for the display), but it's not CUDA enabled afaik. I'll also check to be sure. :)

Does the CUDA card drive a monitor? If not it might be worth hooking one up. I have seen some weird things with that before too. ;)
 
Tesla cards have no video outputs as it's strictly compute, compute, compute. :p

Chances are, the display card is in that list. I just can't remember which one. :(


if your using the 258.xx drivers it should list both cards or grab gpu-z..
 
if your using the 258.xx drivers it should list both cards or grab gpu-z..
or even better CUDA-Z - with recent drivers, I've noticed that GPU-Z doesn't always list the correct order of cards.
 
I'm running the WUs as protected apps, which should only affect older, incompatible ones' ability to render graphics.

Can you try it not in protected mode? I've never used that option when installing. Everything else looks good, so I can't spot any obvious problems.

Try running GPUGRID. It's generally a solid NVIDIA project.
 
or even better CUDA-Z - with recent drivers, I've noticed that GPU-Z doesn't always list the correct order of cards.


nviida's drivers dont list them correctly either.. gpu2 is primary and gpu1 is secondary for me..
 
Found out the display card is an NVS 295. I don't want to use it for CUDA, which Boinc already has set.
NVIDIA GPU 0: Tesla C1060 (driver version 25896, CUDA version 3010, compute capability 1.3, 4096MB, 622 GFLOPS peak)
NVIDIA GPU 1 (not used): Quadro NVS 295 (driver version 25896, CUDA version 3010, compute capability 1.1, 256MB, 21 GFLOPS peak)
 
Fixed it, it seems. :)

The nVidia control panel let me set which GPU actually receives CUDA work (one, both, or none), so the WUs are actually making progress now, with the Tesla selected. If Boinc "targets" a particular GPU, it won't work correctly unless the nVidia defaults are changed.
 
Last edited:
I guess the same WU was being issued to both GPUs. Yup, weird. :confused:
 
Last edited:
Well, Collatz all failed today, but MilkyWay are working fine.

Hmmm ...
 
Can you try it not in protected mode? I've never used that option when installing. Everything else looks good, so I can't spot any obvious problems.

Try running GPUGRID. It's generally a solid NVIDIA project.
Yeah, GPUGRID does support the Tesla C1060 and is does pretty good on it.
From GPUGRID - "
Recommended: Geforce GT 220,240 and GTX 260-216, 275, 280, 285, 295, 460, 465, 470, 480 and Tesla10, Tesla20
Supported: Read this thread"
 
I think I figured out the problem.

If I use Remote Desktop on a system, the GPU is temporarily disabled, which I think is killing any GPU WU being worked on. Boinc doesn't realize this, so it keeps sending all the queued work to error out, as long as that remote session is taking place.

Basically, that leaves a lot of dead units.
 
Last edited:
Have you looked at this Optimized Applications page over at Collatz?
http://boinc.thesonntags.com/collatz/power_apps.php

Good to hear its working on some apps! How long is it taking for a MW WU?


the so called optimized apps at collatz really arent optimized.. they just havent changed them.. the current stable(auto download) cuda app collatz is using is the same app thats in the optimized app page..
 
Back
Top