Welcome to MilkyWay@home

Run Multiple WU's on Your GPU

Message boards : Number crunching : Run Multiple WU's on Your GPU
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 . . . 11 · Next

AuthorMessage
Profile Cliff
Avatar

Send message
Joined: 28 Nov 14
Posts: 51
Credit: 86,696,721
RAC: 0
Message 64196 - Posted: 19 Dec 2015, 19:05:39 UTC - in response to Message 64186.  

What's wrong with rereading the configuration? Works for me.

Boinc Manager > Options > Read configuration. I suppose that's the item in English, my Manager spells them in Dutch.


Because it doesn't work!

Did that got the T shirt. The only way to sort the problem out is either reset the project, and dump your cache, OR do what I did and edit client state files.
Regards,
Cliff.
--
Been there Done That, still no Damn T-Shirt
ID: 64196 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,939,765
RAC: 22,862
Message 64197 - Posted: 20 Dec 2015, 11:28:52 UTC - in response to Message 64195.  

Hi Mikey,
yeah, makes sense, but I just wish someone in the programming dept at BOINC would makes those sorts of changes immediate.. Save a lot of faffing about to sort out a problem, particularly if you have a large[ish] cache of WU.

Regards


I believe this is a MW 'tweak' they made to their Server software, it isn't true at all the other projects.
ID: 64197 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Sotho

Send message
Joined: 12 Nov 11
Posts: 3
Credit: 19,247,595
RAC: 0
Message 65264 - Posted: 25 Sep 2016, 20:08:08 UTC

Just for fun :) HD6970
ID: 65264 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jake Weiss
Volunteer moderator
Project developer
Project tester
Project scientist

Send message
Joined: 25 Feb 13
Posts: 580
Credit: 94,200,158
RAC: 0
Message 65278 - Posted: 26 Sep 2016, 16:37:27 UTC

Hey Everyone,

Just noticed this awesome thread and decided to make it sticky since it is very useful. Thank you all for this.

Jake
ID: 65278 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Sotho

Send message
Joined: 12 Nov 11
Posts: 3
Credit: 19,247,595
RAC: 0
Message 65285 - Posted: 26 Sep 2016, 18:27:36 UTC

1WU ~18s/WU: ~128300 credit/day


4WUs ~12.5s/WU: ~184800 credit/day



It is +44% , less than i expected. (~74%)

13.1 driver, maybe the new is better...
ID: 65285 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Sotho

Send message
Joined: 12 Nov 11
Posts: 3
Credit: 19,247,595
RAC: 0
Message 65288 - Posted: 26 Sep 2016, 19:41:42 UTC
Last modified: 26 Sep 2016, 19:43:19 UTC

12.10 good old driver is the best till now,~200.000 credits/day with my HD6970 when 2WU run.
ID: 65288 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Sutaru Tsureku

Send message
Joined: 30 Apr 09
Posts: 99
Credit: 29,853,513
RAC: 1,056
Message 65387 - Posted: 5 Oct 2016, 18:29:58 UTC
Last modified: 5 Oct 2016, 18:35:56 UTC

Maybe it would be helpful to make an overview (new/other sticky thread) how much WUs/GPU simultaneously give the highest output referred to the VGA cards?


I start with...

After a few fast tests - more will follow - the current result:

On my AMD Radeon R9 Fury X's I let run 3 WUs/GPU with:

<app_config>
<app_version>
<app_name>milkyway</app_name>
<plan_class>opencl_ati_101</plan_class>
<avg_ncpus>0.01</avg_ncpus>
<ngpus>0.33</ngpus>
<cmdline>--non-responsive --gpu-target-frequency 1 --gpu-polling-mode -1 --gpu-wait-factor 0 --process-priority 4 --gpu-disable-checkpointing</cmdline>
</app_version>
<app_version>
<app_name>milkyway_separation_modified__fit</app_name>
<plan_class>opencl_ati_101</plan_class>
<avg_ncpus>0.01</avg_ncpus>
<ngpus>0.33</ngpus>
<cmdline>--non-responsive --gpu-target-frequency 1 --gpu-polling-mode -1 --gpu-wait-factor 0 --process-priority 4 --gpu-disable-checkpointing</cmdline>
</app_version>
</app_config>


On my NVIDIA GeForce GT 730 I let run 2 WUs/GPU with:

<app_config>
<app_version>
<app_name>milkyway</app_name>
<plan_class>opencl_nvidia_101</plan_class>
<avg_ncpus>0.01</avg_ncpus>
<ngpus>0.5</ngpus>
<cmdline>--non-responsive --gpu-target-frequency 1 --gpu-polling-mode 0 --gpu-wait-factor 0 --process-priority 4</cmdline>
</app_version>
<app_version>
<app_name>milkyway_separation_modified__fit</app_name>
<plan_class>opencl_nvidia_101</plan_class>
<avg_ncpus>0.01</avg_ncpus>
<ngpus>0.5</ngpus>
<cmdline>--non-responsive --gpu-target-frequency 1 --gpu-polling-mode 0 --gpu-wait-factor 0 --process-priority 4</cmdline>
</app_version>
</app_config>


(strange, (pre) and (code) don't work properly - is the website broken?)


On both PCs no idle CPU-Cores, CPUs fully loaded with CPU WUs.
ID: 65387 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
iwajabitw

Send message
Joined: 16 Nov 14
Posts: 16
Credit: 335,683,507
RAC: 0
Message 65438 - Posted: 13 Oct 2016, 14:38:25 UTC

How about an a Gtx 980 OC. How many GPU tasks should I be able to run and what what would be the proper edit. On SETI I usually have 3 going to use 99% of the GPU cuda. Thanks.
ID: 65438 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,142,956
RAC: 2
Message 65456 - Posted: 16 Oct 2016, 21:34:42 UTC - in response to Message 65438.  

I just experimented with my Radeon R9 290.

One at a time, it does a WU in 15 seconds.
Two at a time, it does two WUs in 25 seconds.
Three at a time, it does three WUs in 35 seconds.

So there is a speed advantage, but after running three for about 5 minutes, my screens all went off for a second, and all the WUs stopped. I restarted BOINC and they still stayed partially completed. I aborted the WUs and it carried on ok with the next ones.

I'll stick to one at a time.

I also noticed it did three together, starting and stopping them all at the same time, so I wasn't removing the gap between GPU 100% usage completely, just making it happen less often.

Maybe they should be giving out bigger WUs? Even my CPU cores can do them in 15 minutes. Or maybe only the GPU units could be bigger? Not sure what's possible at the Milkyway server end.
ID: 65456 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Vortac

Send message
Joined: 22 Apr 09
Posts: 95
Credit: 4,808,181,963
RAC: 0
Message 65458 - Posted: 16 Oct 2016, 22:22:24 UTC - in response to Message 65456.  
Last modified: 16 Oct 2016, 22:59:56 UTC

I also noticed it did three together, starting and stopping them all at the same time, so I wasn't removing the gap between GPU 100% usage completely, just making it happen less often

Well, in first run all workunits will start and finish almost simultaneously. But soon enough, they'll start diverging, creating a more balanced load. After a few runs and with 3 or 4 workunits running together, chances that they will all finish at the same time are rather small and your GPU load will almost never be zero.
ID: 65458 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,142,956
RAC: 2
Message 65459 - Posted: 17 Oct 2016, 0:41:59 UTC - in response to Message 65458.  

I waited for 15 minutes, but they seemed to converge rather than diverge. With three at once, one of them was 5 seconds off, but it soon became equal. Then it crashed.
ID: 65459 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Vortac

Send message
Joined: 22 Apr 09
Posts: 95
Credit: 4,808,181,963
RAC: 0
Message 65462 - Posted: 17 Oct 2016, 7:14:37 UTC
Last modified: 17 Oct 2016, 7:16:14 UTC

I have 4 GPUs each running 4 WUs, this is how they look in my BOINC. No two progress bars are the same:

http://www.upslike.net/images/2016/10/17/87e38f55b1922c5461bb4457ccb8a573.jpg
ID: 65462 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
mmonnin

Send message
Joined: 2 Oct 16
Posts: 162
Credit: 1,004,359,532
RAC: 16,224
Message 65464 - Posted: 17 Oct 2016, 17:13:05 UTC

^^ Top computer in MW.

Try pausing a task for a second or 2 to offset one of them. Running 3 at once I would sometimes get 2 WUs at 100% so I moved it to 4x as well.
ID: 65464 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,142,956
RAC: 2
Message 65466 - Posted: 17 Oct 2016, 17:25:54 UTC - in response to Message 65464.  
Last modified: 17 Oct 2016, 17:26:26 UTC

They seem to converge anyway if they're offset. Besides, I'm sticking to one to prevent it from crashing. I have three monitors on the card aswell, maybe it gets all confused with 6 things to do. The increase isn't x3 anyway, more like x1.3, not worth locking it up and losing loads of computing time.
ID: 65466 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
iwajabitw

Send message
Joined: 16 Nov 14
Posts: 16
Credit: 335,683,507
RAC: 0
Message 65467 - Posted: 18 Oct 2016, 0:40:59 UTC

So after some testing there is no real benefit to running multiple WU on the GTX980. With a max setting in the config file for 20 wu, it takes 12 minutes to process avg, all the tasks. 1 wu setting, takes 40seconds avg. Power % of the GPU never exceeds 58 even at 20 tasks at once. So simple math, I can do 20wu in 12 minutes or 1 in 40 seconds yielding 30 tasks avg complete over the same 12 minutes.
ID: 65467 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bluestang

Send message
Joined: 13 Oct 16
Posts: 112
Credit: 1,174,293,644
RAC: 0
Message 65468 - Posted: 18 Oct 2016, 2:35:41 UTC - in response to Message 65467.  

Doesn't matter if you can run 1 or 20 WUs if this project can't properly optimise the DB, scheduling, connections, etc.

Extreme mishandling and excessive server load.
ID: 65468 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,142,956
RAC: 2
Message 65470 - Posted: 18 Oct 2016, 10:37:49 UTC - in response to Message 65468.  

What problems? I get plenty WUs.
ID: 65470 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,142,956
RAC: 2
Message 65481 - Posted: 18 Oct 2016, 21:13:15 UTC - in response to Message 56955.  

As per the original post, this works to double up GPU WUs:

<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.05</cpu_usage>
</gpu_versions>
</app>
</app_config>

But how do I make this work for the CPU requirements? I want certain projects to leave more CPU cores free when running GPU apps, as they need them. Specifically Einstein. I put "2" in the cpu usage, but it doesn't take effect. I still see "0.5C + 1 AMD/ATI GPU in the task list. Is <name>milkyway</name> important? I just substituted "einstein".
ID: 65481 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Rymorea

Send message
Joined: 6 Oct 14
Posts: 46
Credit: 20,017,425
RAC: 0
Message 65484 - Posted: 18 Oct 2016, 21:50:10 UTC

I use this app_config for Einstein AMD uses %11-14 and NVIDIA %3-7 for BRP4G


<app_config>
<app>
<name>einsteinbinary_BRP4G</name>
<max_concurrent>2</max_concurrent>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>0.4</cpu_usage>
</gpu_versions>
</app>
</app_config>

ID: 65484 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mr P Hucker
Avatar

Send message
Joined: 5 Jul 11
Posts: 990
Credit: 376,142,956
RAC: 2
Message 65485 - Posted: 18 Oct 2016, 22:22:19 UTC - in response to Message 65484.  

Thanks, that worked. I put "1" into CPU usage, and it now takes a CPU core away from CPU WUs if Einstein is on the GPU.

Where did you get "einsteinbinary_BRP4G" from? Is it just the first bit of the "executable" in properties for the task?
ID: 65485 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 . . . 11 · Next

Message boards : Number crunching : Run Multiple WU's on Your GPU

©2024 Astroinformatics Group