S@H GPU client (opencl) not making full use of GPU ?

Message boards : Number crunching : S@H GPU client (opencl) not making full use of GPU ?
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Seneca Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Oct 02
Posts: 51
Credit: 10,114,348
RAC: 272
Germany
Message 1836248 - Posted: 16 Dec 2016, 8:53:24 UTC
Last modified: 16 Dec 2016, 9:11:19 UTC

Just curious ...

I'm running BOINC with mainly S@H on a Windows 10 machine with 32GB RAM and a NVIDIA GTX570 GPU in a PCIe x 16 slot.

While inspecting the GPU usage with Sysinternals ProcessMonitor, the display in the GPU engine history (System Information -> GPU -> Engines) I get 16 engines displayed, but there are not more than 4 of them used ... the others remain a zero percent load ...

Of the active engines only one or two are working at about full load, the others are only spiky with approx 30% averaged load at max. GPU temp is reported with around 65-70°C by RealTemp.

Is there any misconfiguration, or is it expected that the GPU doesn't look to be working at full load ?

Curious greetings

Seneca
0=0
ID: 1836248 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1836252 - Posted: 16 Dec 2016, 9:05:45 UTC - in response to Message 1836248.  
Last modified: 16 Dec 2016, 9:08:10 UTC

Of the active engines only one or two are working at about full load, the others are only spiky with approx 30% averaged load at max. GPU temp is reported with around 66°C by RealTemp.

Is there any misconfiguration, or is it expected that the GPU doesn't look to be working at full load ?

I don't know how Process Explorer determines what is or isn't a GPU engine and how many of them there are, however for my GTX 1070s Process Explorer shows 30 GPU engines, and there are 15 Compute Units on each of my video cards.

One of the methods for improving the present applications is to make more use of all available Compute Units, as well as make better use of them. There is still a way to go with both of those objectives.
The older applications are very limited in just how much of the GPUs resources they can make use of.
Grant
Darwin NT
ID: 1836252 · 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 1836257 - Posted: 16 Dec 2016, 9:27:30 UTC

The easiest way to increase workload on your GPU is to add comand line params to the comand line text file in your project folder.
Search for a file called ****cmdline***.txt and add the following

-sbs 512 -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.
On next task the comandline values wil be used.

The other thing is to run multiple instances on your GPU.

Du kannst mich auch gerne in deutsch fragen.


With each crime and every kindness we birth our future.
ID: 1836257 · Report as offensive
Profile Seneca Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Oct 02
Posts: 51
Credit: 10,114,348
RAC: 272
Germany
Message 1836294 - Posted: 16 Dec 2016, 14:59:18 UTC - in response to Message 1836257.  

Hmmm ...

ich habe hier

    C:\ProgramData\BOINC\projects\setiathome.berkeley.edu\ap_cmdline_7.03_win_x86_64_SSE2_CPU.txt
    C:\ProgramData\BOINC\projects\setiathome.berkeley.edu\ap_cmdline_7.03_win_x86_SSE_CPU.txt
    C:\ProgramData\BOINC\slots\5\mb_cmdline.txt



In welchen ?

Vielleicht hat sich die Frage auch irgendwie erledigt ... das Programm GPUTemp zeigt

    GPU Core Temp 69°C
    GPU Core Load 98%
    GPU Memory Controller 45% ... 78%
    GPU Memory 50%



an ... das sieht eher korrekt aus ...

Viele Grüße

Seneca
0=0

ID: 1836294 · 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 1836296 - Posted: 16 Dec 2016, 15:54:28 UTC

Normal solltest du etwas mit **Nvidia_SOG**.txt haben.
Du kannst es mal in die mb_cmdline.txt kopieren so heißt die Datei bei mir.
Ich lasse aber unter anonymous Plattform laufen.
Es kann sein dass die Namensgebung für stock apps geändert wurde.

Auf deine Temps solltest du aber achten.
Evtl Lüfter höher drehen.


With each crime and every kindness we birth our future.
ID: 1836296 · Report as offensive
Profile Seneca Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Oct 02
Posts: 51
Credit: 10,114,348
RAC: 272
Germany
Message 1836837 - Posted: 18 Dec 2016, 21:34:12 UTC - in response to Message 1836296.  
Last modified: 18 Dec 2016, 21:35:35 UTC

Ich denke mittlerweile, das sollte so ganz OK sein ... die Temperatur geht bis max. 72°C hoch und ist i.d.R. so um die 65°C.

Die Lüftersteuerung steht auf "Auto" - gerade habe ich 70°C, Core 99% und der Lüfter dreht bei (automatischen) 60%

Ach ja - ein Wert fehlt noch: Die Video Engine steht i.d.R. bei 0%, was insgesamt die Auslastungsanzeige erklären könnte.

By the way: Meine CPU köchelt derweil bei dezenten 78°C vor sich hin ... da sind maximal 98°C zulässig, und ich denke konstant knapp 80 sind CPU-freundlicher als ständiges auf und ab ...

Grüße

Seneca
0=0[/img]
ID: 1836837 · Report as offensive

Message boards : Number crunching : S@H GPU client (opencl) not making full use of GPU ?


 
©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.