BOINC@AUSTRALIA FORUM
September 24, 2019, 06:28:34 AM *
Welcome, Guest. Please login or register.

Login with username, password and session length
News:
 
   Home   Help Login Register  
Pages: [1]   Go Down
  Print  
Author Topic: Running two tasks on one GPU  (Read 3803 times)
Dingo
Founder BOINC@AUSTRALIA
ADMINISTRATOR
Super Hero
*****

Aussie Karma: +2166/-0
Offline Offline

Posts: 9217



WWW
« on: November 28, 2017, 01:09:46 PM »

I have been running one task on my new GTX1080Ti and it only takes between 465 - 479 ms to do a task and getting 3465 credits or about 26,825 per hour.

A short time ago I changed the setting on the Einstein to do two tasks on the GPU at the same time.   The time jumped up to between 730 and 1207 and the same credit. for 1207 that is 10,334 per hour X 2 because I am running two tasks = 20,668.

In the best case  of 730 ms that is 17,087 per hour X 2 = 34,174 so I will keep it running two till I get more results to see if it is worth while.


https://einsteinathome.org/host/12593401/tasks/2/0
Logged


Have a look at the BOINC@AUSTRALIA Facebook Page and join.

Proud Founder and member of BOINC@AUSTRALIA

HERE are my Primes from Primegrid

Have a look at my  Web Cam of Parliament House Ottawa, CANADA
chooka03
Information Member
Super Hero
*****

Aussie Karma: +587/-2
Offline Offline

Posts: 1783



« Reply #1 on: November 28, 2017, 05:30:58 PM »

After much advice from Gary Roberts, here what I was using Dingo -

<app_config>
    <app>
        <name>hsgamma_FGRPB1G</name>
        <gpu_versions>
            <gpu_usage>0.33</gpu_usage>
            <cpu_usage>0.2</cpu_usage>
        </gpu_versions>
    </app>
</app_config>

I think I found this optimal for my GPU.

The DOWNSIDE with this though was setting my CPU usage to 25% so I can only run 1 CPU task. Sucks a bit. Einstein uses 1 core per WU.

« Last Edit: November 28, 2017, 05:33:08 PM by chooka03 » Logged

Dingo
Founder BOINC@AUSTRALIA
ADMINISTRATOR
Super Hero
*****

Aussie Karma: +2166/-0
Offline Offline

Posts: 9217



WWW
« Reply #2 on: November 28, 2017, 07:52:18 PM »

@Chooka-   it looks like I am doing a task between 720 - 730 ms so if they were all 720ms that is 17325 credits per hour X 2 as I am running two tasks or 34,650 per hour.  On one GPU I can't complain and I have 8 cores and 16 threads on this new machine so I am running 12 CPU tasks on POGS and the 2 Einstein on the GPU Big Grin  https://pogs.theskynet.org/pogs/results.php?hostid=832792&offset=0&show_names=0&state=4&appid=
« Last Edit: November 28, 2017, 07:58:28 PM by Dingo » Logged


Have a look at the BOINC@AUSTRALIA Facebook Page and join.

Proud Founder and member of BOINC@AUSTRALIA

HERE are my Primes from Primegrid

Have a look at my  Web Cam of Parliament House Ottawa, CANADA
chooka03
Information Member
Super Hero
*****

Aussie Karma: +587/-2
Offline Offline

Posts: 1783



« Reply #3 on: November 29, 2017, 04:55:25 AM »

Cool!
This is going to be a bit of an issue for me though. Having to dedicate 3 out of my 4 cores on one box to GPU work means I can't do too well at the CPU crunching for POGS.

Oh well....... that's how the cookie crumbles I guess :)
Logged

zzrhardy
Full Member
***

Aussie Karma: +28/-0
Offline Offline

Posts: 86


« Reply #4 on: October 06, 2018, 02:36:50 PM »

I had been running 1 WU at a time on my Einstein@Home box as a way of reducing load/heat a bit, but just retested multiple WUs again to see if tuning had changed. On both cards (Titan V and Vega 64 LC) the 2nd WU gave modest overall gains, the 3rd gave very small overall gains and going further yielded no further gains, it just increased completion time linearly.

So it looks like 3x WU at a time per GPU is all that is required to completely load down a decent GPU.

For reference this was the appconfig file contents:

Code:
<app_config>
    <app>
        <name>hsgamma_FGRPB1G</name>
        <gpu_versions>
            <gpu_usage>0.33</gpu_usage>
            <cpu_usage>1</cpu_usage>
        </gpu_versions>
    </app>
</app_config>
Logged



Current builds: Kuroneko and Chibineko
Dataman
Forum Moderator
Super Hero
*****

Aussie Karma: +1620/-0
Offline Offline

Posts: 2882


Madness Takes Its Toll - Have Exact Change Ready


« Reply #5 on: October 07, 2018, 11:36:27 AM »

I had been running 1 WU at a time on my Einstein@Home box as a way of reducing load/heat a bit, but just retested multiple WUs again to see if tuning had changed. On both cards (Titan V and Vega 64 LC) the 2nd WU gave modest overall gains, the 3rd gave very small overall gains and going further yielded no further gains, it just increased completion time linearly.

So it looks like 3x WU at a time per GPU is all that is required to completely load down a decent GPU.

For reference this was the appconfig file contents:

Code:
<app_config>
    <app>
        <name>hsgamma_FGRPB1G</name>
        <gpu_versions>
            <gpu_usage>0.33</gpu_usage>
            <cpu_usage>1</cpu_usage>
        </gpu_versions>
    </app>
</app_config>

I have been running 0.25 on most of mine. I tried some at 0.20 but with 2 x GPU's on an 8 thread machine it does not have enough CPU. Even on a 12 thread it munches up 10 CPU's. It does work well on the 32 thread machines.  Wave
Logged


chooka03
Information Member
Super Hero
*****

Aussie Karma: +587/-2
Offline Offline

Posts: 1783



« Reply #6 on: October 07, 2018, 05:09:53 PM »

0.33 is what I run with. Except with the RX480 because it can't do more than 1 WU at a time. Stupid.
Apparently it's a Win 10 issue or something.
Logged

zzrhardy
Full Member
***

Aussie Karma: +28/-0
Offline Offline

Posts: 86


« Reply #7 on: October 07, 2018, 08:49:54 PM »

Ugh, I forgot why I first switched to single WU on the Vega 64s - sometimes random work units would take up to 5 hours whilst all the others would proceed at normal speed. Throughput on Chibineko actually decreased over last 24 hours due to this, so I am back to single WU on this box.

No issues with 3x WU on the Titan Vs tho.
Logged



Current builds: Kuroneko and Chibineko
Pages: [1]   Go Up
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.11 | SMF © 2006-2009, Simple Machines LLC Valid XHTML 1.0! Valid CSS!
Page created in 0.04 seconds with 18 queries.