New to cuda

Message boards : Number crunching : New to cuda
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Profile John Gillett
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 53
Credit: 11,401,398
RAC: 0
United States
Message 1266209 - Posted: 31 Jul 2012, 14:57:06 UTC
Last modified: 31 Jul 2012, 14:57:53 UTC

Running Win7 Pro x64, quad-core processor, 8GB, BOINC 7.0.28, Lunatics 0.40 Wizard accepting all install defaults. app_info.xml has not been edited in any way.

Recently acquired a GeForce 9500 GT (I know - old) with the following specs...

Operating System: Windows 7 Professional, 64-bit (Service Pack 1)
DirectX version: 11.0
GPU processor: GeForce 9500 GT
Driver version: 301.42
DirectX support: 10
CUDA Cores: 32
Core clock: 550 MHz
Shader clock: 1375 MHz
Memory clock: 800 MHz (1600 MHz data rate)
Memory interface: 128-bit
Total available graphics memory: 4095 MB
Dedicated video memory: 512 MB GDDR3
System video memory: 0 MB
Shared system memory: 3583 MB
Video BIOS version: 62.94.1E.00.06
IRQ: 16
Bus: PCI Express x8

Currently the card is happy running only one work unit of cuda, cuda23, or cuda_fermi at a time. Can it handle more than just one work unit, and, if so, exactly what do I need to edit?

Been in computers as a profession for over 47 years, and running Seti since 26 May 1999. However, cuda is new stuff to me.

Thanks.
John
ID: 1266209 · Report as offensive
Mike Davis
Volunteer tester

Send message
Joined: 17 May 99
Posts: 240
Credit: 5,402,361
RAC: 0
Isle of Man
Message 1266230 - Posted: 31 Jul 2012, 20:00:50 UTC - in response to Message 1266209.  

Bit too on the old side for more than one to give any increase in performance, im afraid. Its only advisable on fermi cards or above (gtx4xx series+)
ID: 1266230 · Report as offensive
Profile John Gillett
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 53
Credit: 11,401,398
RAC: 0
United States
Message 1266274 - Posted: 31 Jul 2012, 22:40:08 UTC - in response to Message 1266230.  

Thanks, Mike. Kinda what I thought.
John
ID: 1266274 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 1266295 - Posted: 31 Jul 2012, 23:45:57 UTC - in response to Message 1266209.  

... Lunatics 0.40 Wizard accepting all install defaults.
...

The Core 2 Quad CPU would almost certainly do somewhat better with the SSSE3x application rather than the AMD/Intel SSE3 it's using. With a modest GPU the CPU apps will be responsible for a large fraction of the host's productivity, it looks like the 9500 GT is about like adding two additional cores to the CPU.
                                                                   Joe
ID: 1266295 · Report as offensive
Profile John Gillett
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 53
Credit: 11,401,398
RAC: 0
United States
Message 1266326 - Posted: 1 Aug 2012, 3:09:44 UTC - in response to Message 1266295.  
Last modified: 1 Aug 2012, 3:14:23 UTC

Joe,

Just ran the Wizard again making your suggested change. Will see what happens.

FWIW, here's the startup log...

7/31/2012 20:05:58 | | No config file found - using defaults
7/31/2012 20:05:58 | | Starting BOINC client version 7.0.28 for windows_x86_64
7/31/2012 20:05:58 | | log flags: file_xfer, sched_ops, task
7/31/2012 20:05:58 | | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6
7/31/2012 20:05:58 | | Data directory: C:\ProgramData\BOINC
7/31/2012 20:05:58 | | Running under account John
7/31/2012 20:05:58 | | Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU @ 2.66GHz [Family 6 Model 15 Stepping 7]
7/31/2012 20:05:58 | | Processor: 4.00 MB cache
7/31/2012 20:05:58 | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 syscall nx lm vmx tm2 pbe
7/31/2012 20:05:58 | | OS: Microsoft Windows 7: Professional x64 Edition, Service Pack 1, (06.01.7601.00)
7/31/2012 20:05:58 | | Memory: 8.00 GB physical, 21.67 GB virtual
7/31/2012 20:05:58 | | Disk: 279.46 GB total, 115.57 GB free
7/31/2012 20:05:58 | | Local time is UTC -7 hours
7/31/2012 20:05:58 | | NVIDIA GPU 0: GeForce 9500 GT (driver version 301.42, CUDA version 4.20, compute capability 1.1, 512MB, 302MB available, 132 GFLOPS peak)
7/31/2012 20:05:58 | | OpenCL: NVIDIA GPU 0: GeForce 9500 GT (driver version 301.42, device version OpenCL 1.0 CUDA, 512MB, 302MB available)
7/31/2012 20:05:58 | SETI@home | Found app_info.xml; using anonymous platform
7/31/2012 20:05:58 | Quake-Catcher Network | URL http://qcn.stanford.edu/sensor/; Computer ID 13271; resource share 100
7/31/2012 20:05:58 | SETI@home | URL http://setiathome.berkeley.edu/; Computer ID 5283496; resource share 100
7/31/2012 20:05:58 | SETI@home | General prefs: from SETI@home (last modified 31-Jul-2012 07:51:03)
7/31/2012 20:05:58 | SETI@home | Computer location: home
7/31/2012 20:05:58 | SETI@home | General prefs: no separate prefs for home; using your defaults
7/31/2012 20:05:58 | | Reading preferences override file
7/31/2012 20:05:58 | | Preferences:
7/31/2012 20:05:58 | | max memory usage when active: 4094.85MB
7/31/2012 20:05:58 | | max memory usage when idle: 7370.73MB
7/31/2012 20:05:58 | | max disk usage: 100.00GB
7/31/2012 20:05:58 | | (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)
7/31/2012 20:05:58 | | Not using a proxy
7/31/2012 20:05:58 | Quake-Catcher Network | Restarting task qcnm_089418_0 using qcnsensor version 706 (nci) in slot 0
7/31/2012 20:05:58 | SETI@home | Restarting task 24fe12ab.27285.382587.12.10.12.vlar_0 using setiathome_enhanced version 603 in slot 1
7/31/2012 20:05:58 | SETI@home | Restarting task 24fe12ab.27285.382587.12.10.16.vlar_0 using setiathome_enhanced version 603 in slot 2
7/31/2012 20:05:58 | SETI@home | Restarting task 26fe12ah.31343.11283.5.10.132_0 using setiathome_enhanced version 610 (cuda_fermi) in slot 3
7/31/2012 20:05:58 | SETI@home | Restarting task 24fe12ab.27285.382587.12.10.52.vlar_0 using setiathome_enhanced version 603 in slot 5
7/31/2012 20:05:58 | SETI@home | Restarting task 24fe12ab.27285.382587.12.10.22.vlar_0 using setiathome_enhanced version 603 in slot 4

Thanks.
John
ID: 1266326 · Report as offensive
Profile Fred J. Verster
Volunteer tester
Avatar

Send message
Joined: 21 Apr 04
Posts: 3252
Credit: 31,903,643
RAC: 0
Netherlands
Message 1266539 - Posted: 1 Aug 2012, 22:41:41 UTC - in response to Message 1266326.  
Last modified: 1 Aug 2012, 22:53:56 UTC

Two years ago I used a 9500GT and a 9800GTX+* ran very hot and within
a year they started to make errors, very often found triplets twice... as
error message.
I changed to a GTX470 and a GTX480* (FERMI), > 5x faster* and less CPU time/usage. Very quiet fan @ 45% temp 77C GPU load ~96% average over 1 MB WU.
One WU per GPU.

Today I'd go for a GTX570, 670/680 or 690(2x GPU). €990,00 for the latter
is a bit too much $$ or €€ .

Host .


Your mileage may vary.............
ID: 1266539 · Report as offensive
Profile John Gillett
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 53
Credit: 11,401,398
RAC: 0
United States
Message 1266544 - Posted: 1 Aug 2012, 23:37:43 UTC - in response to Message 1266539.  

Fred,

Thanks for the info. Unfortunately, my problem is pretty simple- out of work for over 3 years! Finances? Non-existent.

I got this 9500 used for 40 bucks. So far, a pretty good deal. It's really punching out data. I can put my hand on the card, heat sink, etc., and it feels 'normal' to the touch.

Current numbers are going up rapidly, and, even though I've been running Seti since '99, my user average is already way above what it has ever been, even running multiple computers. {Thanks, Joe - your suggestion seems to be working!}

Down the road - who knows? Maybe by then I'll have some $$ to be able to upgrade.

Thanks again.
John
ID: 1266544 · Report as offensive
Profile Bill G Special Project $75 donor
Avatar

Send message
Joined: 1 Jun 01
Posts: 1282
Credit: 187,688,550
RAC: 182
United States
Message 1266590 - Posted: 2 Aug 2012, 2:57:45 UTC - in response to Message 1266544.  

Fred,

Thanks for the info. Unfortunately, my problem is pretty simple- out of work for over 3 years! Finances? Non-existent.

I got this 9500 used for 40 bucks. So far, a pretty good deal. It's really punching out data. I can put my hand on the card, heat sink, etc., and it feels 'normal' to the touch.

Current numbers are going up rapidly, and, even though I've been running Seti since '99, my user average is already way above what it has ever been, even running multiple computers. {Thanks, Joe - your suggestion seems to be working!}

Down the road - who knows? Maybe by then I'll have some $$ to be able to upgrade.

Thanks again.

I run a 9500 on one of my systems and I run 2 WUs at a time on it and it just does just fine. It was fun to get it doing that. I have believe it does better than when it ran a single WU at a time.

SETI@home classic workunits 4,019
SETI@home classic CPU time 34,348 hours
ID: 1266590 · Report as offensive
Profile John Gillett
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 53
Credit: 11,401,398
RAC: 0
United States
Message 1266611 - Posted: 2 Aug 2012, 5:08:50 UTC - in response to Message 1266590.  

Bill,

Exactly how do I set that up (keep in mind the title of this topic!)? Would like to give it a try.

Thanks.
John
ID: 1266611 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1266624 - Posted: 2 Aug 2012, 6:33:41 UTC - in response to Message 1266611.  


If you used the default locations for your installation, there should be
C:\ProgramData\BOINC\projects\setiathome.berkely

In the setiathome.berkely folder should be an app_info.xml file. Open it with a text editor (Notepad is good enough- Wordpad is evil).

Change every instance of <count>1</count> to <count>0.5</count>
Being a 64 bit version of Windows there should be 6 of them (from memory).

Save the file, exit & restart BOINC.
You should now have 2 WUs being run on the video card at a time.
Grant
Darwin NT
ID: 1266624 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 1266719 - Posted: 2 Aug 2012, 15:18:59 UTC

A couple of additional notes:

The directory is of course "setiathome.berkeley.edu".

Bill G's 9500 GT has 1024 MB of VRAM which leaves plenty for Windows display use even when running two tasks. With 512 MB, it may be necessary to turn off aero or otherwise reduce Windows' display usage to run two tasks, or it may not be possible at all.
                                                                   Joe
ID: 1266719 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1266724 - Posted: 2 Aug 2012, 15:36:16 UTC - in response to Message 1266719.  

nobody said that running 2 WU's couldn't be done. It shouldn't be done because it is slower than running 1 at a time. Looking at Bill G's Wu's shows that his GPU is running WU's slower than what a CPU can do.

I'd suggest Bill G do a test. Check the angle ranges and time to completion on 10--20 WU's when running 2 at a time then run 1 at a time and check the angle ranges and time to completion again. You should have enough similar AR results to compare your 2X work vs your 1X work. I'm betting you'll much better (>2X faster) on your 1X WU's.


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1266724 · Report as offensive
Profile John Gillett
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 53
Credit: 11,401,398
RAC: 0
United States
Message 1266727 - Posted: 2 Aug 2012, 15:46:19 UTC - in response to Message 1266724.  

Well - it failed.

The restart of BOINC was good, and everything looked OK. There were 2 cudas running. However, after about 10 seconds I started getting these...


8/2/2012 08:31:09 | SETI@home | Task 07ap10ab.30197.16452.11.10.128_0 exited with zero status but no 'finished' file
8/2/2012 08:31:09 | SETI@home | If this happens repeatedly you may need to reset the project.
8/2/2012 08:31:09 | SETI@home | Starting task 07ap10ab.30197.16452.11.10.164_0 using setiathome_enhanced version 610 (cuda_fermi) in slot 7
8/2/2012 08:31:29 | SETI@home | Task 07ap10ab.30197.16452.11.10.164_0 exited with zero status but no 'finished' file
8/2/2012 08:31:29 | SETI@home | If this happens repeatedly you may need to reset the project.
8/2/2012 08:31:29 | SETI@home | Starting task 07ap10ab.30197.16452.11.10.94_0 using setiathome_enhanced version 610 (cuda_fermi) in slot 8
8/2/2012 08:31:46 | SETI@home | Task 07ap10ab.30197.16452.11.10.94_0 exited with zero status but no 'finished' file
8/2/2012 08:31:46 | SETI@home | If this happens repeatedly you may need to reset the project.

etc., etc.

So - as suggested - it appears that card doesn't have the brain space to handle two. Reset the counts (6 was correct - good memory) to 1 and all is well again.

My thanks to all for your help and suggestions!
John
ID: 1266727 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1266728 - Posted: 2 Aug 2012, 15:48:55 UTC

I´m not a cuda guy but so far i´m concerned running multiple instances is not possible on pre fermi cards.



With each crime and every kindness we birth our future.
ID: 1266728 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1266734 - Posted: 2 Aug 2012, 16:15:51 UTC - in response to Message 1266728.  

I´m not a cuda guy but so far i´m concerned running multiple instances is not possible on pre fermi cards.

It's perfectly possible (provided you have enough memory on the GPU card), but there's much less point. Hardware acceleration for the necessary context switching was introduced with Fermi cards - older cards can do it, but slower.
ID: 1266734 · Report as offensive
Profile Fred J. Verster
Volunteer tester
Avatar

Send message
Joined: 21 Apr 04
Posts: 3252
Credit: 31,903,643
RAC: 0
Netherlands
Message 1266736 - Posted: 2 Aug 2012, 16:31:01 UTC - in response to Message 1266734.  
Last modified: 2 Aug 2012, 16:31:42 UTC

I´m not a cuda guy but so far i´m concerned running multiple instances is not possible on pre fermi cards.

It's perfectly possible (provided you have enough memory on the GPU card), but there's much less point. Hardware acceleration for the necessary context switching was introduced with Fermi cards - older cards can do it, but slower.


Tried it on a 9800GTX+ and run-times just about doubled also got way too hot!
Even on my GTX470 (@ 900MHz, not 1400MHz) I run 1 MB WU also on the 480.

Just benchmarking my i7-2600 (@ 3560MHz. and 2 HD5870 GPUs, stock speed 850MHz.
Upped it to 896MHz stress tested it with a few MW WUs and got too hot.

Restarted SETI but after 5 hours GPU-driver crashed, so back to stock speed.
ID: 1266736 · Report as offensive
Profile Bill G Special Project $75 donor
Avatar

Send message
Joined: 1 Jun 01
Posts: 1282
Credit: 187,688,550
RAC: 182
United States
Message 1266788 - Posted: 2 Aug 2012, 19:07:23 UTC - in response to Message 1266727.  

Well - it failed.

The restart of BOINC was good, and everything looked OK. There were 2 cudas running. However, after about 10 seconds I started getting these...


8/2/2012 08:31:09 | SETI@home | Task 07ap10ab.30197.16452.11.10.128_0 exited with zero status but no 'finished' file
8/2/2012 08:31:09 | SETI@home | If this happens repeatedly you may need to reset the project.
8/2/2012 08:31:09 | SETI@home | Starting task 07ap10ab.30197.16452.11.10.164_0 using setiathome_enhanced version 610 (cuda_fermi) in slot 7
8/2/2012 08:31:29 | SETI@home | Task 07ap10ab.30197.16452.11.10.164_0 exited with zero status but no 'finished' file
8/2/2012 08:31:29 | SETI@home | If this happens repeatedly you may need to reset the project.
8/2/2012 08:31:29 | SETI@home | Starting task 07ap10ab.30197.16452.11.10.94_0 using setiathome_enhanced version 610 (cuda_fermi) in slot 8
8/2/2012 08:31:46 | SETI@home | Task 07ap10ab.30197.16452.11.10.94_0 exited with zero status but no 'finished' file
8/2/2012 08:31:46 | SETI@home | If this happens repeatedly you may need to reset the project.

etc., etc.

So - as suggested - it appears that card doesn't have the brain space to handle two. Reset the counts (6 was correct - good memory) to 1 and all is well again.

My thanks to all for your help and suggestions!


John, you did everything correct but I see now that you have a 9500 with only 512K memory, that is not enough. My 9500 has 1024K memory. Sorry to have raised your hopes on this but what happened should be considered normal considering you did not have enough GPU memory. At least now you know how to do it when you perhaps get a different video card.

SETI@home classic workunits 4,019
SETI@home classic CPU time 34,348 hours
ID: 1266788 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1266796 - Posted: 2 Aug 2012, 19:27:43 UTC - in response to Message 1266788.  

another very simple test is running GPUz while processing a WU. If the GPU is maxxed out on 1 then it certainly won't run 2 efficiently


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1266796 · Report as offensive
Profile John Gillett
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 53
Credit: 11,401,398
RAC: 0
United States
Message 1266808 - Posted: 2 Aug 2012, 19:50:49 UTC - in response to Message 1266788.  

John, you did everything correct but I see now that you have a 9500 with only 512K memory, that is not enough. My 9500 has 1024K memory. Sorry to have raised your hopes on this but what happened should be considered normal considering you did not have enough GPU memory. At least now you know how to do it when you perhaps get a different video card.


Bill, not a problem at all. It was a good exercise, and, as you say, at least now I know for when I win the sweepstakes and can get a better card... ;-)

Thanks.
John
ID: 1266808 · Report as offensive
Profile John Gillett
Volunteer tester
Avatar

Send message
Joined: 26 May 99
Posts: 53
Credit: 11,401,398
RAC: 0
United States
Message 1266814 - Posted: 2 Aug 2012, 20:14:49 UTC - in response to Message 1266796.  

another very simple test is running GPUz while processing a WU. If the GPU is maxxed out on 1 then it certainly won't run 2 efficiently


As suggested...



John
ID: 1266814 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : New to cuda


 
©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.