Returning to Seti, but CUDA is really slow?

Questions and Answers : GPU applications : Returning to Seti, but CUDA is really slow?
Message board moderation

To post messages, you must log in.

AuthorMessage
GammelMan Project Donor

Send message
Joined: 24 Jun 07
Posts: 3
Credit: 811,161
RAC: 0
Sweden
Message 928864 - Posted: 26 Aug 2009, 19:19:45 UTC

Hello all,

I'm sort of returning to Seti after running Foldign@Home for a while on GPU.

When I run on my hardware which is
26/08/2009 19:38:50 CUDA device: GeForce 8800 GT (driver version 19038, compute capability 1.1, 512MB, est. 60GFLOPS)

it seems to be really slow. After two hours I have completed about 25% of a workunit. Is this about right or is there something wrong with my setup perhaps?

I have GPU enabled when computer is in use, and turned off CPU work. I've tried looking in the forums, but not found any clues.

Thanks in advance for any pointers.
ID: 928864 · Report as offensive
Fred W
Volunteer tester

Send message
Joined: 13 Jun 99
Posts: 2524
Credit: 11,954,210
RAC: 0
United Kingdom
Message 928869 - Posted: 26 Aug 2009, 20:08:06 UTC - in response to Message 928864.  

Hello all,

I'm sort of returning to Seti after running Foldign@Home for a while on GPU.

When I run on my hardware which is
26/08/2009 19:38:50 CUDA device: GeForce 8800 GT (driver version 19038, compute capability 1.1, 512MB, est. 60GFLOPS)


Could you just double-check the figures, please? 60GFlops seems remarkably high for an 8800GT.

F.
ID: 928869 · Report as offensive
GammelMan Project Donor

Send message
Joined: 24 Jun 07
Posts: 3
Credit: 811,161
RAC: 0
Sweden
Message 928879 - Posted: 26 Aug 2009, 20:35:06 UTC - in response to Message 928869.  

Hello all,

I'm sort of returning to Seti after running Foldign@Home for a while on GPU.

When I run on my hardware which is
26/08/2009 19:38:50 CUDA device: GeForce 8800 GT (driver version 19038, compute capability 1.1, 512MB, est. 60GFLOPS)


Could you just double-check the figures, please? 60GFlops seems remarkably high for an 8800GT.

F.


Well, this is a more complete snippet from the message log that describes my HW setup:
-----
26/08/2009 22:28:39 Processor: 2 AuthenticAMD AMD Athlon(tm) 64 X2 Dual Core Processor 4200+ [x86 Family 15 Model 43 Stepping 1]
26/08/2009 22:28:39 Processor features: fpu tsc pae nx sse sse2 3dnow mmx
26/08/2009 22:28:39 OS: Microsoft Windows XP: Home x86 Edition, Service Pack 2, (05.01.2600.00)
26/08/2009 22:28:39 Memory: 2.00 GB physical, 1.85 GB virtual
26/08/2009 22:28:39 Disk: 232.88 GB total, 14.42 GB free
26/08/2009 22:28:39 Local time is UTC +2 hours
26/08/2009 22:28:39 CUDA device: GeForce 8800 GT (driver version 19038, compute capability 1.1, 512MB, est. 60GFLOPS)
-----

When running a WU, the status line says "0.14 CPUs, 1 CUDA" if that's any help.
ID: 928879 · Report as offensive
Profile Gundolf Jahn

Send message
Joined: 19 Sep 00
Posts: 3184
Credit: 446,358
RAC: 0
Germany
Message 928893 - Posted: 26 Aug 2009, 21:30:54 UTC - in response to Message 928864.  

...it seems to be really slow...

Check the stderr.txt file in the slot directory for "true angle range."

Very Low Angle Range (VLAR) tasks tend to run very slow on CUDA. I'm not quite sure, but I think values around 0.01 are very low.

Gruß,
Gundolf

PS: just found this message on Number crunching:
Yes, vlar (Very Low Angle Range <= 0.05) WUs can end with Exit status -177 (0xffffffffffffff4f), noted later with the message "Maximum elapsed time exceeded". Each WU has a computation limit, and for the speed of your GTX 295 GPUs that will work out to be 4h11'.
So, my estimate was a little low.
Computer sind nicht alles im Leben. (Kleiner Scherz)

SETI@home classic workunits 3,758
SETI@home classic CPU time 66,520 hours
ID: 928893 · Report as offensive
Fred W
Volunteer tester

Send message
Joined: 13 Jun 99
Posts: 2524
Credit: 11,954,210
RAC: 0
United Kingdom
Message 928898 - Posted: 26 Aug 2009, 21:50:31 UTC - in response to Message 928879.  

Well, this is a more complete snippet from the message log that describes my HW setup:
-----
26/08/2009 22:28:39 Processor: 2 AuthenticAMD AMD Athlon(tm) 64 X2 Dual Core Processor 4200+ [x86 Family 15 Model 43 Stepping 1]
26/08/2009 22:28:39 Processor features: fpu tsc pae nx sse sse2 3dnow mmx
26/08/2009 22:28:39 OS: Microsoft Windows XP: Home x86 Edition, Service Pack 2, (05.01.2600.00)
26/08/2009 22:28:39 Memory: 2.00 GB physical, 1.85 GB virtual
26/08/2009 22:28:39 Disk: 232.88 GB total, 14.42 GB free
26/08/2009 22:28:39 Local time is UTC +2 hours
26/08/2009 22:28:39 CUDA device: GeForce 8800 GT (driver version 19038, compute capability 1.1, 512MB, est. 60GFLOPS)
-----

When running a WU, the status line says "0.14 CPUs, 1 CUDA" if that's any help.

Well, that's me told :) I apologise for doubting your figures; I don't run an 8800GT myself and was extrapolating from other 88xx / 9xxx figures that I had read here on the boards. Hopefully you will get a more enlightened response from someone who also runs the same type of card.

The "0.14 CPUs, 1 CUDA" is quite normal; that just means that just 14% of a CPU has been specified to feed the GPU so the other 86% can be used for other purposes (other activies or crunching a CPU WU as you prefer).

F.
ID: 928898 · Report as offensive
GammelMan Project Donor

Send message
Joined: 24 Jun 07
Posts: 3
Credit: 811,161
RAC: 0
Sweden
Message 928957 - Posted: 27 Aug 2009, 4:16:06 UTC - in response to Message 928893.  

...it seems to be really slow...

Check the stderr.txt file in the slot directory for "true angle range."

Very Low Angle Range (VLAR) tasks tend to run very slow on CUDA. I'm not quite sure, but I think values around 0.01 are very low.

Gruß,
Gundolf

PS: just found this message on Number crunching:
Yes, vlar (Very Low Angle Range <= 0.05) WUs can end with Exit status -177 (0xffffffffffffff4f), noted later with the message "Maximum elapsed time exceeded". Each WU has a computation limit, and for the speed of your GTX 295 GPUs that will work out to be 4h11'.
So, my estimate was a little low.



Looking in the stderr file, I see "WU true angle range is : 0.008326"
In that case, maybe the 3+ hours per WU I get is to be expected?
ID: 928957 · Report as offensive
Fred W
Volunteer tester

Send message
Joined: 13 Jun 99
Posts: 2524
Credit: 11,954,210
RAC: 0
United Kingdom
Message 928966 - Posted: 27 Aug 2009, 5:51:34 UTC - in response to Message 928957.  

Looking in the stderr file, I see "WU true angle range is : 0.008326"
In that case, maybe the 3+ hours per WU I get is to be expected?

That's your answer. One up to Gundolf :). That VLAR would take about 70 mins on my GTX295 where a "normal" WU takes between 5 and 50 mins (most are around 10 mins) according to Angle Range.

F.
ID: 928966 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 929177 - Posted: 28 Aug 2009, 2:14:26 UTC
Last modified: 28 Aug 2009, 2:21:14 UTC

If you don't want to compute this kind of long-time VLAR WUs you can use Optimized Apps - Lunatics' Unified Installer:

http://setiathome.berkeley.edu/forum_thread.php?id=54288&nowrap=true#927649

(They will be automatically rejected (killed))
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 929177 · Report as offensive
Profile Gundolf Jahn

Send message
Joined: 19 Sep 00
Posts: 3184
Credit: 446,358
RAC: 0
Germany
Message 929197 - Posted: 28 Aug 2009, 4:07:26 UTC - in response to Message 929177.  

Or, better yet, try the rescheduling tool. See this thread.
ID: 929197 · Report as offensive
Chuck Gorish

Send message
Joined: 19 Jun 00
Posts: 156
Credit: 29,589,106
RAC: 0
United States
Message 929238 - Posted: 28 Aug 2009, 12:30:46 UTC - in response to Message 929197.  

Or, better yet, try the rescheduling tool. See this thread.


yes. i'm using the linux perl script version of this and it is really good. my vlar killer application no longer gets to see any vlar or vhar units so everything gets processed optimally.
ID: 929238 · Report as offensive

Questions and Answers : GPU applications : Returning to Seti, but CUDA is really slow?


 
©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.