SETI@home v8.22 Windows GPU applications support thread

Message boards : Number crunching : SETI@home v8.22 Windows GPU applications support thread
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 3 · 4 · 5 · 6 · 7 · Next

AuthorMessage
Profile xixou
Volunteer tester

Send message
Joined: 9 Mar 02
Posts: 10
Credit: 2,815,187
RAC: 0
Belgium
Message 1867668 - Posted: 17 May 2017, 7:42:25 UTC - in response to Message 1867587.  

Got no BSOD anymore since I removed seti@home.
ID: 1867668 · Report as offensive     Reply Quote
Terror Australis
Volunteer tester

Send message
Joined: 14 Feb 04
Posts: 1806
Credit: 236,611,763
RAC: 56,615
Australia
Message 1867669 - Posted: 17 May 2017, 7:49:27 UTC
Last modified: 17 May 2017, 7:50:46 UTC

Thanks to everybody who gave me help and advice.
After the usual cycle of dramas you get when bringing an old machine that's been off for a while up from cold, the Q9550 + 2 x GTX750's is now quite stable, and while running 2 units per card, is getting through a normal unit in ~20 mins and, 30 to 40 mins for a guppi. I'm happy with that.
All crunching is being done on the GPU's

Mike wrote
Check for a file called mb_cmdline_win_x86_SSE3_OpenCL_NV_SoG.txt in the projects folder.
Add the following.

-sbs 1024 -period_iterations_num 40 -spike_fft_thresh 2048 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 32 -oclfft_tune_cw 32
Save as text.

This worked wonders for the GTX750's, chopped nearly 15% off the single unit crunching time. Could you please give me some settings for the 1070Ti ?

Raistmer wrote
he issue you encountered on 750Ti is:
ERROR: OpenCL kernel/call 'clGetEventProfilingInfo' call failed (-7) in file ..\..\..\src\GPU_lock.cpp near line 546.

-7 is CL_PROFILING_INFO_NOT_AVAILABLE

It's NV OpenCL Runtime issue. Maybe can be cured by driver change.
As workaround try to update to r3584 (your build is r3557). Newer builds account for such Runtime failures and treat this error as non-important just omitting tuning.

I upgraded to r3584 and the 2nd last WinXP driver, I now get tasks "wating to run" but not "postponed". Which I think is more a BOINC scheduling issue than anything else. It was the "Postponed" message that threw me..

Once again, thanks to all those who helped out
T.A.
ID: 1867669 · Report as offensive     Reply Quote
Stephen "Heretic"Project Donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 2638
Credit: 48,796,103
RAC: 137,660
Australia
Message 1867691 - Posted: 17 May 2017, 12:05:32 UTC - in response to Message 1867668.  

Got no BSOD anymore since I removed seti@home.


. . May I ask what BSOD are?

Stephen

??
ID: 1867691 · Report as offensive     Reply Quote
Stephen "Heretic"Project Donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 2638
Credit: 48,796,103
RAC: 137,660
Australia
Message 1867692 - Posted: 17 May 2017, 12:12:26 UTC - in response to Message 1867669.  


I upgraded to r3584 and the 2nd last WinXP driver, I now get tasks "wating to run" but not "postponed". Which I think is more a BOINC scheduling issue than anything else. It was the "Postponed" message that threw me..


. . If you are getting that "waiting to run" status a lot then you might try reducing the value of -sbs. Maybe 768 or even 512 would work better for you.

Stephen

:)
ID: 1867692 · Report as offensive     Reply Quote
Profile Brent Norman
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 1824
Credit: 108,261,062
RAC: 457,605
Canada
Message 1867693 - Posted: 17 May 2017, 12:14:26 UTC - in response to Message 1867691.  

. . May I ask what BSOD are?

Windows Famous ... Blue Screen Of Death
ID: 1867693 · Report as offensive     Reply Quote
rob smithProject Donor
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 15229
Credit: 253,025,643
RAC: 323,155
United Kingdom
Message 1867696 - Posted: 17 May 2017, 12:57:07 UTC

In the past they were strongly linked with the PC's stress detection system - the higher your personal stress level the more probable one would experience a BSOD :-(
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1867696 · Report as offensive     Reply Quote
Stephen "Heretic"Project Donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 2638
Credit: 48,796,103
RAC: 137,660
Australia
Message 1867749 - Posted: 17 May 2017, 17:12:09 UTC - in response to Message 1867693.  

. . May I ask what BSOD are?

Windows Famous ... Blue Screen Of Death


. . D'oh! I was trying to think of something in terms of Seti. Funny how you can miss the obvious ..... Thanks for that.

Stephen

:)
ID: 1867749 · Report as offensive     Reply Quote
Terror Australis
Volunteer tester

Send message
Joined: 14 Feb 04
Posts: 1806
Credit: 236,611,763
RAC: 56,615
Australia
Message 1867921 - Posted: 18 May 2017, 12:09:52 UTC - in response to Message 1867692.  
Last modified: 18 May 2017, 12:11:41 UTC


I upgraded to r3584 and the 2nd last WinXP driver, I now get tasks "wating to run" but not "postponed". Which I think is more a BOINC scheduling issue than anything else. It was the "Postponed" message that threw me..


. . If you are getting that "waiting to run" status a lot then you might try reducing the value of -sbs. Maybe 768 or even 512 would work better for you.

Stephen

:)

I tried that, and whatever part of the system that was just managing to keep itself together..... broke completely :(

All I get now when I try to run SoG, is half a dozen or so units going round in a continuous, Run -> Waiting to run -> Postponed -> Run -> n loop.

It's only "normal" units that do this, guppi's crunch quite happily

I tried going back to 41zi, but some of the "normal" units coming through must be unlabelled vlars because they were taking nearly 2 hours to crunch.

Is there any way to modify the app_info so that files with the suffix .vlar get crunched by SoG and those units without a suffix use x41z ?

Is there any way to make this thing behave ? Or do I just put this box back to sleep ?

T.A.
ID: 1867921 · Report as offensive     Reply Quote
Profile Brent Norman
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 1824
Credit: 108,261,062
RAC: 457,605
Canada
Message 1867926 - Posted: 18 May 2017, 12:44:11 UTC - in response to Message 1867921.  

All I get now when I try to run SoG, is half a dozen or so units going round in a continuous, Run -> Waiting to run -> Postponed -> Run -> n loop.
Is that your Quad 2 doing that? From what I see it is.

My AMD 4200 (w/750Ti) use to do it too with SoG tasks. I cured it by changing to Linux.

The best I could figure with it was the slow CPU casing it. In the sense that IF the GPU tasks were assigned to the CPU, it would be over the 10 day limit, causing the Arecibo tasks to only run for a very short time before switching tasks, the BLC tasks ran fine. I set it to 10s check points to get through them. I don't believe I ever found a cure for it. Thinking maybe a smaller cache might work, don't think I tried that.

I know the it sounds strange that the CPU time would be considered in the GPU equation, but it was the best thing I could think of. I know it started with a changing Lunatics and BOINC versions (upgraded both at the same time), and believe it was because it ran benchmarks to come up with something new. I reverted back to previous versions and that did not cure it. I thought at the time maybe manually entering a flop value may help, but didn't try it.

It's been over a year now, that's the best I can recall from that ordeal ...
ID: 1867926 · Report as offensive     Reply Quote
Stephen "Heretic"Project Donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 2638
Credit: 48,796,103
RAC: 137,660
Australia
Message 1867966 - Posted: 18 May 2017, 16:24:51 UTC - in response to Message 1867921.  
Last modified: 18 May 2017, 16:30:59 UTC


I tried that, and whatever part of the system that was just managing to keep itself together..... broke completely :(

All I get now when I try to run SoG, is half a dozen or so units going round in a continuous, Run -> Waiting to run -> Postponed -> Run -> n loop.

It's only "normal" units that do this, guppi's crunch quite happily

I tried going back to 41zi, but some of the "normal" units coming through must be unlabelled vlars because they were taking nearly 2 hours to crunch.

Is there any way to modify the app_info so that files with the suffix .vlar get crunched by SoG and those units without a suffix use x41z ?

Is there any way to make this thing behave ? Or do I just put this box back to sleep ?

T.A.


. . OK I am confused ... I thought that x41z was a CUDA variant. Maybe I am missing something. But I would try to get one or the other working properly rather than mix things up. SoG works pretty well if you get it setup right. Maybe something went wrong when you tried to setup r3584. Not seeing what is setup and seeing r3557 on your result STDERR section I am guessing maybe reinstall Lunatics Beta6 and see if you can get things working properly again. Just run SSE3 for the CPU in the Core2 and AVX for the CPU in the i7. The use SoG for the GPUs and concentrate on getting that back to working AOK. Oh, and make sure you have at least one CPU core free for each GPU, so set the Core 2 Quad Boinc manager to only use 50 or 25 % of CPU cores. I had my Core2 Duo with a GTX1050ti putting up a 13K RAC and I was crunching on one CPU core on that machine (but it does have 6GB Ram). So I know that that series of CPUs can work AOK with SoG on a GPU.

. . Some Normal WUs are what I call Quite Low AR units (QLARs) which have AR values close to but not low enough to be classified as VLAR. (Which is below 0.12 I believe). These QLARs are often below 0.2 and they will run very slowly compared to other normal tasks but SoG can process even Arecibo tasks classified as .VLAR if you wanted. SoG is probably the best thing for crunching GBT work as you have noted but should be AOK for all Arecibo tasks as well. When in doubt take it back to the setup where you know it was working AOK then try to move forward from there.

. . Also I notice that even though you are running doubles your still have -sbs set to 1024. If you look at your STDERR you sill see that it is setting the buffer size to 511MB anyway. So I would suggest you stick to -sbs 512 if you really want to run doubles. But that's up to you.

Stephen

??
ID: 1867966 · Report as offensive     Reply Quote
Profile xixou
Volunteer tester

Send message
Joined: 9 Mar 02
Posts: 10
Credit: 2,815,187
RAC: 0
Belgium
Message 1869302 - Posted: 25 May 2017, 8:24:36 UTC

I also had BSOD when suspending my miner crushing.
I recently moved to latest Nvidia drivers 382.33 (was 382.19 before) and got no bsod so far.
This may solve the problem for seti as well.
ID: 1869302 · Report as offensive     Reply Quote
Profile xixou
Volunteer tester

Send message
Joined: 9 Mar 02
Posts: 10
Credit: 2,815,187
RAC: 0
Belgium
Message 1869307 - Posted: 25 May 2017, 11:00:01 UTC - in response to Message 1869302.  

I also had BSOD when suspending my miner crushing.
I recently moved to latest Nvidia drivers 382.33 (was 382.19 before) and got no bsod so far.
This may solve the problem for seti as well.


mm still got a bsod after supsend on mining, nvidia issue i guess
ID: 1869307 · Report as offensive     Reply Quote
Miklos M.

Send message
Joined: 5 May 99
Posts: 840
Credit: 18,058,806
RAC: 4,956
United States
Message 1895711 - Posted: 16 Oct 2017, 19:13:30 UTC

I have tried various problem solvings, but my computer with 780ti's can only get 8.00 units and I wanted 8.22.
Another issue on my other computer with 2 1080 ti's keeps running only one gpu for SETI, in Primegrid it uses both. What can I do?
Any help would be welcome.
ID: 1895711 · Report as offensive     Reply Quote
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 8894
Credit: 115,355,780
RAC: 70,403
Australia
Message 1895808 - Posted: 17 Oct 2017, 4:34:20 UTC - in response to Message 1895711.  

I have tried various problem solvings, but my computer with 780ti's can only get 8.00 units and I wanted 8.22.

The first one (8.00) you have used the Lunatics Installer, the second one (8.22) is running stock.

The one running stock will perform the best, as you have selected the wrong application to run using the Lunatics Installer. Running the CUDA32 application will result in the poorest possible performance from your hardware.
If you were to install the SoG application it would be running what is essentially (other than some very minor differences), the stock application.
Grant
Darwin NT
ID: 1895808 · Report as offensive     Reply Quote
Miklos M.

Send message
Joined: 5 May 99
Posts: 840
Credit: 18,058,806
RAC: 4,956
United States
Message 1895847 - Posted: 17 Oct 2017, 12:30:24 UTC - in response to Message 1895808.  

Thank you Grant very much. If you do not mind would you check if I did it right? There were so many choices. I see significant improvement in the speed already.
On my other computer: Could you tell me why am I getting only the one 1080ti card running in SETI( I have two installed) I tried a copy and paste directions, but must have done it wrong. This computer is getting the 8.22 ocl units fine.
ID: 1895847 · Report as offensive     Reply Quote
Profile Jimbocous
Volunteer tester
Avatar

Send message
Joined: 1 Apr 13
Posts: 1003
Credit: 91,194,797
RAC: 89,259
United States
Message 1895864 - Posted: 18 Oct 2017, 0:08:10 UTC - in response to Message 1895847.  
Last modified: 18 Oct 2017, 0:11:36 UTC

...
On my other computer: Could you tell me why am I getting only the one 1080ti card running in SETI( I have two installed) I tried a copy and paste directions, but must have done it wrong. This computer is getting the 8.22 ocl units fine.

In Program Data/BOINC/cc_config.xml, verify under
</proxy_info>
.
.
          <use_all_gpus>1</use_all_gpus>
.

Otherwise only the most capable will be used.
Jim ...
ID: 1895864 · Report as offensive     Reply Quote
Miklos M.

Send message
Joined: 5 May 99
Posts: 840
Credit: 18,058,806
RAC: 4,956
United States
Message 1895868 - Posted: 18 Oct 2017, 0:43:48 UTC - in response to Message 1895864.  

Thank you. I will do my best to insert this. And hope for the best.
ID: 1895868 · Report as offensive     Reply Quote
Un4givenProject Donor

Send message
Joined: 1 May 04
Posts: 17
Credit: 4,442,271
RAC: 6,052
United States
Message 1900064 - Posted: 9 Nov 2017, 19:47:22 UTC

Hey folks. Just replaced an AMD 280X with an RX Vega 56. I pretty well had my config files optimized for the 280X, but I'm not so sure about any changes needed for the newer Vega cards. Am I good to go with the old card's config file for Vega, or are there any suggestions as to modifications I should make?

Any and all suggestions welcome!
ID: 1900064 · Report as offensive     Reply Quote
Profile MikeProject Donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 30617
Credit: 57,715,808
RAC: 30,059
Germany
Message 1900174 - Posted: 10 Nov 2017, 7:49:54 UTC

You can use the old config files but maybe it can be further optimized.
With each crime and every kindness we birth our future.
ID: 1900174 · Report as offensive     Reply Quote
Un4givenProject Donor

Send message
Joined: 1 May 04
Posts: 17
Credit: 4,442,271
RAC: 6,052
United States
Message 1900768 - Posted: 12 Nov 2017, 20:37:17 UTC - in response to Message 1900174.  

OK, thanks! I'll just keep going with what I've got until someone that knows this stuff better than I has something tangible.
ID: 1900768 · Report as offensive     Reply Quote
Previous · 1 . . . 3 · 4 · 5 · 6 · 7 · Next

Message boards : Number crunching : SETI@home v8.22 Windows GPU applications support thread


 
©2017 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.