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 · 2 · 3 · 4 · 5 · 6 · 7 . . . 23 · Next

AuthorMessage
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1861523 - Posted: 15 Apr 2017, 1:24:04 UTC - in response to Message 1861517.  

All the long running task under r3584 are either VLAR BLC or Arecibo normal 0.44 AR. The shorter runtime time tasks were Arecibo shorties. I never saw much difference in run times for the r3557 to the r3584 app. The benefit was a reduction in the inconclusives list. But considering he is running with a very low performance card, there might be a measurable difference in the apps that is masked by high performance cards and tuning like I have with my 1070's.

He really should read Raistmer's notes about the app configuration for the low performance path. He isn't running the app correctly yet. I saw someones comment already about knocking back the -sbs buffer size to more reasonable settings.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1861523 · Report as offensive     Reply Quote
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1861524 - Posted: 15 Apr 2017, 1:26:53 UTC - in response to Message 1861486.  

Correction, i meant to say i am using GPU-only:


I installed the Lunatics beta 0.45 beta 6 and the Nvidia r_3584, also i decided to crunch with GPU only as these appear to have the most opportunities for optimizations.
GPU utilization looks around 99% (it runs HOT, also when using "-use_sleep").



. . Hey there Kets,

. . Did you manually edit app_info.xml to install r3584? If so can you give me some tips on what it took?

Stephen

??
ID: 1861524 · Report as offensive     Reply Quote
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1861530 - Posted: 15 Apr 2017, 1:34:48 UTC - in response to Message 1861523.  
Last modified: 15 Apr 2017, 1:37:37 UTC

All the long running task under r3584 are either VLAR BLC or Arecibo normal 0.44 AR. The shorter runtime time tasks were Arecibo shorties. I never saw much difference in run times for the r3557 to the r3584 app. The benefit was a reduction in the inconclusives list. But considering he is running with a very low performance card, there might be a measurable difference in the apps that is masked by high performance cards and tuning like I have with my 1070's.

He really should read Raistmer's notes about the app configuration for the low performance path. He isn't running the app correctly yet. I saw someones comment already about knocking back the -sbs buffer size to more reasonable settings.


. . I had meant to ask you that very thing in light of this convo. You may be right, that any apparent improvement on low performance cards disappears on high end units. But since I am running a GTX950 here I wonder if it might gain some small increase in performance. :)

. . So tell me please, what exact edits did you have to make to replace r3557 with r3584?

Stephen

??
ID: 1861530 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1861534 - Posted: 15 Apr 2017, 1:43:18 UTC - in response to Message 1861523.  

there might be a measurable difference in the apps that is masked by high performance cards and tuning like I have with my 1070's.


I'm still rocking r3500 lol.. All the improvement from that app forward were to make the app more responsive to lower end GPUs.
ID: 1861534 · Report as offensive     Reply Quote
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1861540 - Posted: 15 Apr 2017, 2:04:45 UTC
Last modified: 15 Apr 2017, 2:14:58 UTC

I hope I get this right as it's been a long time since I bothered doing this type of thing, though I'm sure that someone will correct me if I'm wrong, but here goes.

Install Lunatics Beta 6 and then shut BOINC down, remove the MB8_win_SSE3_OpenCL_NV.aistub in your Seti data folder, you download the r3584 app, unzip and move all the contents to your seti data folder, you can then run the aimerge.cmd there that will use the new MB8_win_SSE3_OpenCL_NV.aistub from the r3584 files.

Cheers.
ID: 1861540 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1861554 - Posted: 15 Apr 2017, 3:01:56 UTC - in response to Message 1861530.  


. . So tell me please, what exact edits did you have to make to replace r3557 with r3584?

Stephen

??

Hi Stephen, pretty simple really. You just have to do a Replace All with Notepad of the app_info. You are changing all instances of the MB8_win_x86_SSE3_OpenCL_NV_SoG_r3557.exe entries in app_info with the MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe app name. Same goes for the .CL file. Just Replace All any instance of the MultiBeam_Kernels_r3557.cl filename with the MultiBeam_Kernels_r3584.cl filename. Of course you have to put the actual files into your C:\ProgramData\BOINC\projects\setiathome.berkeley.edu directory after you are done editing and before you restart BOINC. I also do a sanity check of the app_info.xml file after editing by opening it in Chrome as an .XML file. What that does is check for any syntax errors that you might have fat-fingered into the edit. It will find any errors in opening/closing argument headers if you forgot or added a ] for example.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1861554 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1861557 - Posted: 15 Apr 2017, 3:08:30 UTC - in response to Message 1861540.  

I hope I get this right as it's been a long time since I bothered doing this type of thing, though I'm sure that someone will correct me if I'm wrong, but here goes.

Install Lunatics Beta 6 and then shut BOINC down, remove the MB8_win_SSE3_OpenCL_NV.aistub in your Seti data folder, you download the r3584 app, unzip and move all the contents to your seti data folder, you can then run the aimerge.cmd there that will use the new MB8_win_SSE3_OpenCL_NV.aistub from the r3584 files.

Cheers.

Well, that is new then. I didn't realize Raistmer had bundled an .aistub file into the app zip file. He hadn't when I downloaded it from his file repository. That is another way of incorporating the new app without a manual hand edit. That works too and makes it automatic with no self created errors in the app_info. As long as the aistub doesn't touch any of your other personalized edits. I find it real simple to accept the aistub defaults and do any personalized configuration via the app_config.xml file. The app_config.xml survives BOINC platform or app_info upgrades or updates.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1861557 · Report as offensive     Reply Quote
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1861568 - Posted: 15 Apr 2017, 3:42:54 UTC - in response to Message 1861557.  


Well, that is new then. I didn't realize Raistmer had bundled an .aistub file into the app zip file. He hadn't when I downloaded it from his file repository. That is another way of incorporating the new app without a manual hand edit. That works too and makes it automatic with no self created errors in the app_info. As long as the aistub doesn't touch any of your other personalized edits. I find it real simple to accept the aistub defaults and do any personalized configuration via the app_config.xml file. The app_config.xml survives BOINC platform or app_info upgrades or updates.

You could leave the MB8_win_SSE3_OpenCL_NV.aistub as you'll probably be asked if you want to replace it with the newer file, but as we all know that sometimes doesn't work properly (or you wind up with the newer file being shown as a copy) so it's better in my books to delete it 1st, and you could also remove all the older r3557 files to cut down on junk in the folder as well. ;-)

Cheers.
ID: 1861568 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1861572 - Posted: 15 Apr 2017, 3:53:09 UTC - in response to Message 1861568.  

I prune the directory pretty often to get rid of all old wisdom and compile files from older apps and graphics drivers. You always have the oldApp_backup directory that the installer creates to fall back on if things go south.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1861572 · Report as offensive     Reply Quote
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1861590 - Posted: 15 Apr 2017, 5:38:15 UTC - in response to Message 1861534.  

there might be a measurable difference in the apps that is masked by high performance cards and tuning like I have with my 1070's.


I'm still rocking r3500 lol.. All the improvement from that app forward were to make the app more responsive to lower end GPUs.


. . That would explain the variance in his results.

Stephen

:)
ID: 1861590 · Report as offensive     Reply Quote
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1861591 - Posted: 15 Apr 2017, 5:40:14 UTC - in response to Message 1861540.  

I hope I get this right as it's been a long time since I bothered doing this type of thing, though I'm sure that someone will correct me if I'm wrong, but here goes.

Install Lunatics Beta 6 and then shut BOINC down, remove the MB8_win_SSE3_OpenCL_NV.aistub in your Seti data folder, you download the r3584 app, unzip and move all the contents to your seti data folder, you can then run the aimerge.cmd there that will use the new MB8_win_SSE3_OpenCL_NV.aistub from the r3584 files.

Cheers.


. . Now where is a pen and paper ........ :)

Stephen

:)
ID: 1861591 · Report as offensive     Reply Quote
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1861593 - Posted: 15 Apr 2017, 5:44:59 UTC - in response to Message 1861557.  

I hope I get this right as it's been a long time since I bothered doing this type of thing, though I'm sure that someone will correct me if I'm wrong, but here goes.

Install Lunatics Beta 6 and then shut BOINC down, remove the MB8_win_SSE3_OpenCL_NV.aistub in your Seti data folder, you download the r3584 app, unzip and move all the contents to your seti data folder, you can then run the aimerge.cmd there that will use the new MB8_win_SSE3_OpenCL_NV.aistub from the r3584 files.

Cheers.

Well, that is new then. I didn't realize Raistmer had bundled an .aistub file into the app zip file. He hadn't when I downloaded it from his file repository. That is another way of incorporating the new app without a manual hand edit. That works too and makes it automatic with no self created errors in the app_info. As long as the aistub doesn't touch any of your other personalized edits. I find it real simple to accept the aistub defaults and do any personalized configuration via the app_config.xml file. The app_config.xml survives BOINC platform or app_info upgrades or updates.


. . So I take it that if there is an aistub file in the distro this would be the preferred method. Like you I do any personal configuration in the app_config.xml file. Thanks guys I will give that a go tomorrow, I have to go out tonight :)

Stephen

thanks
ID: 1861593 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1861610 - Posted: 15 Apr 2017, 7:17:28 UTC - in response to Message 1861593.  

Yes, I downloaded the zip file from Mike's World and unpacked and looked into the aistub. All you need is there and nothing more. That part is important. I would just put the unpacked files into the directory and run the aimerge.cmd. Make sure you move the old MB SoG aistub out of the directory and make sure you still have the MB and AP CPU aistub files in the directory. The aimerge works on all the aistub files. If you don't have the all the original aistubs in place you will get a whittled down app_info with just the SoG app. I made that mistake once and never again. I find it just as fast to do the hand edit since I've done it before a couple of times. You have total control of the app_info then and don't have to make sure all the correct conditions are met to run the aimerge. If you are starting out with a stock Lunatics install, that is not a problem. Only if you have pruned files after the original Lunatics install will that bite you in the ass. If you want to be extra safe, make a copy of app_info and move it out of the directory to somewhere else temporarily. Then you peruse the new app_info file to make sure things look correct and then give it a go and make sure to look at the first startup messages in the logfile to make sure everything is AOK.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1861610 · Report as offensive     Reply Quote
Kets_One

Send message
Joined: 5 Jun 03
Posts: 9
Credit: 975,168
RAC: 0
Netherlands
Message 1861611 - Posted: 15 Apr 2017, 7:18:56 UTC

@Stephen, @Keith

Indeed i performed exactly the actions that Keith describes for merging in r_3584 app into the Lunatics 0.45 beta 6.
Changing the .txt file by hand (find and replace 3557 with 3584) wasnt too difficult.

Thanks for your suggestion on changing back the -sbs argument back to -sbs 1024, indeed i experimented with that setting.

Also, the 'mixed' stderr.out i can confirm. halfway through a few WUs i changed the app to r_3584.

Also, i have been monitoring the GPU utilization closely for the past day and it varies between 95 and 99%, so i will not try to run multiple WU's. One WU at a time seems enough load :)
Temperature hovers between 74-80 degrees.

I would like to experiment more with Raistmer's settings for low-performance path. I will try to find them in the forums.

Thanks,
Kets
ID: 1861611 · Report as offensive     Reply Quote
Kets_One

Send message
Joined: 5 Jun 03
Posts: 9
Credit: 975,168
RAC: 0
Netherlands
Message 1861635 - Posted: 15 Apr 2017, 11:11:20 UTC
Last modified: 15 Apr 2017, 11:44:31 UTC

Ive found the following http://lunatics.kwsn.info/index.php/topic,1808.msg60931.html#msg60931from Raistmer.

Based on the info in the forum there, ive ried out several settings:

Original:
-high_prec_timer -sbs 1024 -period_iterations_num 20 -tt 300 -use_sleep

Attempt 1:
-high_prec_timer -sbs 512 -period_iterations_num 1 -tt 300 -use_sleep
results: strange client behavior (WUs on hold halfway through completion, then start another WU)
High temperatures of GPU (>90 degrees C), high framebuffer usage
High GPU utilization (99%)

Attempt 2:
-high_prec_timer -sbs 1024 -period_iterations_num 10 -tt 250 -use_sleep
Ive reduced -tt a bit and increased -period_iterations_num
Lower GPU utilization (fluctuating between 90 and 99%)
WU completion speed improved compared to original.
Running a bit less hot (85 degrees C), better :)

Attempt 3:
-high_prec_timer -sbs 1024 -period_iterations_num 15 -tt 300 -use_sleep
after reading stderr.out and Raistmers instructions on how to interpret, i am trying these settings.
GPU utilization around 95%
No noticeable changes in WU speed
running a bit less hot (75-80 degrees C), even better still :)
ID: 1861635 · Report as offensive     Reply Quote
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1861656 - Posted: 15 Apr 2017, 13:34:28 UTC - in response to Message 1861635.  
Last modified: 15 Apr 2017, 13:47:34 UTC

Ive found the following http://lunatics.kwsn.info/index.php/topic,1808.msg60931.html#msg60931from Raistmer.

Based on the info in the forum there, ive ried out several settings:

Original:
-high_prec_timer -sbs 1024 -period_iterations_num 20 -tt 300 -use_sleep

Attempt 1:
-high_prec_timer -sbs 512 -period_iterations_num 1 -tt 300 -use_sleep
results: strange client behavior (WUs on hold halfway through completion, then start another WU)
High temperatures of GPU (>90 degrees C), high framebuffer usage
High GPU utilization (99%)

Attempt 2:
-high_prec_timer -sbs 1024 -period_iterations_num 10 -tt 250 -use_sleep
Ive reduced -tt a bit and increased -period_iterations_num
Lower GPU utilization (fluctuating between 90 and 99%)
WU completion speed improved compared to original.
Running a bit less hot (85 degrees C), better :)

Attempt 3:
-high_prec_timer -sbs 1024 -period_iterations_num 15 -tt 300 -use_sleep
after reading stderr.out and Raistmers instructions on how to interpret, i am trying these settings.
GPU utilization around 95%
No noticeable changes in WU speed
running a bit less hot (75-80 degrees C), even better still :)


. . The command lines in green have I believe the wrong iteration/tt ratios. With iterations of 10 tt should be 300 or even 400, with iterations of 15 tt should be about 250 to 300. The fewer iterations you are running the longer the max kernel runtimes need to be. You may need to tweak it a little to find the optimum ratios. And I still think that since the app defaulted sbs to 512 when you had it overspecified at 2048 that is the number I would go with. The stop start behaviour you observed is fairly common when you over commit the VRAM. Use_sleep should only be necessary if you are a) running multiple GPU tasks b) have limited CPU resources or c) are experiencing unacceptable mouse/screen/keyboard lag. If you do not intend to change your mind about crunching on the CPU then your should have sufficient CPU core resources and should not experience lag once well set up. With attempt #1 the iterations number is waaayyy too aggressive for that GPU, and tt should have been at least 1000.

. . Can I ask you to humour an old man and try -

-high_prec_timer -sbs 512 -period_iterations_num 10 -tt 400

. . as that would be about the upper limit of what I think that hardware will happily support

. . If that seems also too aggressive (runs hot or flaky) try

-high_prec_timer -sbs 512 -period_iterations_num 12 -tt 350

. . Unless temps are too high or behaviour is flaky, you need to let each setting run for a day or two to confirm performance on Several WUs.

. . Also does your Laptop BIOS allow you to crank up the fans at all?

Stephen

?
ID: 1861656 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1861679 - Posted: 15 Apr 2017, 16:49:30 UTC - in response to Message 1861611.  

See if you can find and use a GPU fan control program and set a more aggressive fan profile. The stock default fan profile on the Nvidia cards these days are set for low noise and not high performance. That is why you are running at 75-80° C. The cooler you can keep the GPU the more clock speed is set by Nvidia's GPUBoost feature. The card will automatically overclock itself when it is given thermal headroom. That will improve your crunch times.

You can read about Raistmer's app over at Lunatics in the forums.
Some considerations regarding OpenCL MultiBeam app tuning from algorithm view
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1861679 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1861680 - Posted: 15 Apr 2017, 17:04:39 UTC

Based on the last validated GPU task in your list, I agree with Stephen's recommendation. It looks like it was run with a -SBS size of 512 along with reasonable -TT of 300. What I always like to use is Raistmer's tuning runs that are printed at the end of the stderr.txt. If you look for the minimum crunch time, a delta of 1 and the highest N value, the value of the FftLength size tells you your optimum SBS buffer setting. Of course you have to consider the overall size of RAM on you video cards. You can't crunch with a SBS size of 2048KB on a card with a maximum video capacity of 512KB. Also consider the total amount of RAM used if you try to run multiple tasks on the card concurrently.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1861680 · Report as offensive     Reply Quote
Profile Le Meunier

Send message
Joined: 23 Dec 02
Posts: 1
Credit: 1,062,364
RAC: 0
France
Message 1863354 - Posted: 23 Apr 2017, 22:49:06 UTC

I received this warning :
"SETI@home: Notice from server
NVIDIA GPU: Veuillez mettre à jour votre système avec les derniers drivers pour pouvoir calculer avec le GPU
23/04/2017 12:07:26"

I upgraded my PC to a Gigabyte GEFORCE GTX 750Ti with up to date drivers.
Numbers crunching with gpu looks running well at the tasks windows.

Is there a "guru" around to clarify ?
Regards

PS : I run cpdnboinc, SETI, Asteroids, Milkyway and Enigma (currently suspended) with no problem as it seems !
ID: 1863354 · Report as offensive     Reply Quote
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1863374 - Posted: 24 Apr 2017, 0:37:03 UTC - in response to Message 1863354.  
Last modified: 24 Apr 2017, 0:38:02 UTC

Your computer shows:
NVIDIA GeForce GTX 750 Ti (2048MB) driver: 341.95 OpenCL: 1.1

Driver 367.57 works well with that card as well as v375

Get the drivers from nvidia.com and do a CLEAN Install.
ID: 1863374 · Report as offensive     Reply Quote
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 . . . 23 · Next

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


 
©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.