CUDA slow all of a sudden

Questions and Answers : GPU applications : CUDA slow all of a sudden
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Codeman05

Send message
Joined: 16 Dec 01
Posts: 33
Credit: 15,457,430
RAC: 0
United States
Message 881516 - Posted: 2 Apr 2009, 3:10:19 UTC
Last modified: 2 Apr 2009, 3:11:25 UTC

Hey guys, I have the following setup:
i7 920 @ 4.0 GHz
3x 2GB DDR3 1600MHz
3x GTX 285 GPUs

earlier in the week, I was completed a CUDA WU in 2.5 minutes on each GPU (probably completed 1200 at this rate)...
Now all of a sudden it seems to be taking anywhere from 5 to 45+ minutes per WU.

All of the hardware appears to be running fine, so I'm not sure if my client is jacked up, or these new WUs just take longer to crunch.

Anyone else seeing anything similar?

thanks!
ID: 881516 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 881518 - Posted: 2 Apr 2009, 3:13:15 UTC - in response to Message 881516.  

There's a good chance that you weren't completing a WU in 2.5 minutes since the time displayed is the CPU time and not GPU time.

There's also the probability that you are receiving different Angle Ranges which have varying times to completion.
ID: 881518 · Report as offensive
Profile Codeman05

Send message
Joined: 16 Dec 01
Posts: 33
Credit: 15,457,430
RAC: 0
United States
Message 881520 - Posted: 2 Apr 2009, 3:16:05 UTC

correct, I am referring to CPU time. actual time was probably closer to 5-10 mins.

Your probably right about the range angles...atleast I hope that's the case :)
ID: 881520 · Report as offensive
Profile Rick Chabot

Send message
Joined: 10 Apr 02
Posts: 6
Credit: 8,791,333
RAC: 0
United States
Message 881533 - Posted: 2 Apr 2009, 3:50:12 UTC - in response to Message 881520.  

What determines which type of WU is downloaded?
I have two machines with CUDA boards in them.
1 is a Q9300 with an 8800GT
2 is a Q9650 with a GTX285

The Machine with the 8800GT always runs CUDA tasks
The other machine never does..

Any idea why? The 8800GT is kicking the Q9650 in the rear on performance.

Would like to see the GTX285 do some work..

Thanks

neither machine is overclocked
both have 8GB of RAM and WinVista 64-bit SP1
ID: 881533 · Report as offensive
Profile BigDaddyDave
Avatar

Send message
Joined: 8 Oct 00
Posts: 67
Credit: 20,337,995
RAC: 15
United States
Message 881535 - Posted: 2 Apr 2009, 3:50:27 UTC - in response to Message 881516.  



earlier in the week, I was completed a CUDA WU in 2.5 minutes on each GPU
Now all of a sudden it seems to be taking anywhere from 5 to 45+ minutes per WU.


Anyone else seeing anything similar?

thanks!


Hi Codeman05,

Yes it depends on the work unit size. Most of mine are around 5 mins or so but sometimes I receive a CUDA WU that takes 25 mins or so.

Just a thought though, check your GPU temps and also if you are running anything simultaneously, e.g. video games, DVD movies.

Try SpeedFan, it's free and accesses the sensors available on the hardware monitor chips. SpeedFan shows temperatures, voltages and fan speeds.

http://www.almico.com/sfscreenshots.php

Dave
ID: 881535 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 881778 - Posted: 2 Apr 2009, 20:14:37 UTC - in response to Message 881533.  

What determines which type of WU is downloaded?


Other than your preferences, nothing. Work is handed out pseudo-randomly.


The Machine with the 8800GT always runs CUDA tasks
The other machine never does..


Does BOINC "see" the other machine's CUDA card? It should state in the Messages tab after BOINC is first start up if BOINC detects the presence of the CUDA card.

If so, it may simply be that work has been hard to come by for the last two weeks due to the server problems SETI's been having.
ID: 881778 · Report as offensive
Profile Rick Chabot

Send message
Joined: 10 Apr 02
Posts: 6
Credit: 8,791,333
RAC: 0
United States
Message 881869 - Posted: 3 Apr 2009, 2:57:54 UTC - in response to Message 881778.  

4/1/2009 7:55:19 PM|SETI@home|Found app_info.xml; using anonymous platform
4/1/2009 7:55:19 PM||Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU Q9650 @ 3.00GHz [Intel64 Family 6 Model 23 Stepping 10]
4/1/2009 7:55:19 PM||Processor features: fpu tsc pae nx sse sse2 pni
4/1/2009 7:55:19 PM||OS: Microsoft Windows Vista: Ultimate x64 Editon, Service Pack 1, (06.00.6001.00)
4/1/2009 7:55:19 PM||Memory: 8.00 GB physical, 16.04 GB virtual
4/1/2009 7:55:19 PM||Disk: 279.46 GB total, 206.25 GB free
4/1/2009 7:55:19 PM||Local time is UTC -5 hours
4/1/2009 7:55:19 PM||Not using a proxy
4/1/2009 7:55:19 PM||CUDA devices found

Well it says Cuda device found so I am not sure why this one us not running them, but the other one exclusively runs them.. and Astropulse... It runs those too.

But this machine.. no.. I do have some of these messages though:
4/2/2009 2:34:39 PM|SETI@home|Message from server: Your app_info.xml file doesn't have a version of Astropulse v5.

I do not know if this may have anything to do with it.

Thanks for replying.

Rick
ID: 881869 · Report as offensive
Profile Rick Chabot

Send message
Joined: 10 Apr 02
Posts: 6
Credit: 8,791,333
RAC: 0
United States
Message 881880 - Posted: 3 Apr 2009, 3:26:30 UTC - in response to Message 881869.  

Figured out the problem. Had been using an older installation with a different computation engine... V8 AK... something... sorry for ignorance.. I cleaned all that out and removed and reinstalled the Boinc Client and it assumed defaults and downloaded an AP WU and about 20 Cuda units immediately.

Thanks.
ID: 881880 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 881897 - Posted: 3 Apr 2009, 5:19:10 UTC - in response to Message 881880.  

Ahh yes. You were using an optimized application. The AKv8 app is still the latest optimized app, but it does not process CUDA apps, and by having an app_info.xml file (used to run the optimized apps), you are telling BOINC not to download anything other than what is listed in the app_info.xml file (which would only be AKv8-compatible apps, i.e. MultiBeam; non-CUDA).
ID: 881897 · Report as offensive

Questions and Answers : GPU applications : CUDA slow all of a sudden


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.