How do I configure seti@home to run multiple WU on one GPU?

Message boards : Number crunching : How do I configure seti@home to run multiple WU on one GPU?
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Profile RAUERTECH

Send message
Joined: 7 May 09
Posts: 6
Credit: 575,382
RAC: 2
United States
Message 1421963 - Posted: 29 Sep 2013, 12:32:03 UTC

Hey fellow Crunchers,

I'm new to Seti@Home and I'm in love with the cause. I am currently running on my newest machine.
Specs are below...

Core I5 3570k quad core @ 4.0GHz
16 GB of ddr3 1866Mhz
128GB SSD
EVGA GTX 770 2GB edition.
Windows 7 pro 64-bit

I have been doing some thread searching for days now and I need your help. I have installed lunatics app and configured the install per my pc specs. I restarted my computer and ran bionc and I'm still only doing 1 WU on my GPU. Now I know I'm suppose to edit app info but I cannot locate this. Please help! Thanks and let me know if this noob needs to give you more info.
ID: 1421963 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1421967 - Posted: 29 Sep 2013, 12:44:37 UTC - in response to Message 1421963.  
Last modified: 29 Sep 2013, 12:49:44 UTC

I have been doing some thread searching for days now and I need your help. I have installed lunatics app and configured the install per my pc specs. I restarted my computer and ran bionc and I'm still only doing 1 WU on my GPU. Now I know I'm suppose to edit app info but I cannot locate this. Please help! Thanks and let me know if this noob needs to give you more info.

The app_info.xml will be in the the Setiathome project folder, by default the whole Boinc Data directory is a hidden directory, you'll be to unhide it, or paste the full directory into explorer,
the default Boinc Data directory location for Windows 7 is: C:\ProgramData\BOINC\ and the full location for the project folder is: C:\ProgramData\BOINC\projects\setiathome.berkeley.edu

To change the number of tasks running per GPU, edit the app_info.xml with Notepad, and change each <count>1</count> to <count>0.5</count> for two tasks per GPU, or <count>0.33</count> for 3 tasks per GPU.

Restart Boinc.

There is a second method of making Boinc run two tasks per GPU, this is by making an app_config.xml (with Notepad) with the following in it:

<app_config>
 <app>
 <name>astropulse_v6</name>
 <gpu_versions>
 <gpu_usage>0.5</gpu_usage>
 <cpu_usage>0.04</cpu_usage>
 </gpu_versions>
 </app>
 <app>
 <name>setiathome_v7</name>
 <gpu_versions>
 <gpu_usage>0.5</gpu_usage>
 <cpu_usage>0.04</cpu_usage>
 </gpu_versions>
 </app>
</app_config>


Put it in your project directory, and do a 'Read config files', no restart of Boinc required.

Claggy
ID: 1421967 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1421969 - Posted: 29 Sep 2013, 12:45:52 UTC

First, let it run with one on the GPU for a couple of days - this will give you a chance to see if there are any issues with the hardware or drivers.
Then you need to edit the app_info.xml file. This is a plain text file, so only use a simple editor... You need to edit each occurance of "<CUDA>1</CUDA>" to "<CUDA>0.5</CUDA>" - and restart BOINC, this will give you two simultaneous tasks on your GPU, setting the value to 0.33 will give you three tasks, but I'd let the system settle before going up a notch. It can be worth increasing the CPU use from the default 0.04 to say 0.1
It is also worth freeing a CPU core - this is best done from the BOINC manager tool - set the "use xx% of CPU to free one or two cores - a bit of simple arithmetic to work out what xx% should be, as that depends on the total number of cores (real and virtual) your CPU has.
If you leave a few days between each change you cans see what the effect of each is, and can work out what the optimum for your system is.

I should have said (assuming you are using a recent version of windows) is in the "C:\ProgramData\BOINC\projects\setiathome.berkeley.edu" folder by default.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1421969 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1421976 - Posted: 29 Sep 2013, 13:23:17 UTC - in response to Message 1421969.  
Last modified: 29 Sep 2013, 13:39:39 UTC

You need to edit each occurance of "<CUDA>1</CUDA>" to "<CUDA>0.5</CUDA>" - and restart BOINC, this will give you two simultaneous tasks on your GPU, setting the value to 0.33 will give you three tasks, but I'd let the system settle before going up a notch.

The entry that needs changing is the count value:

From:
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>

To:
<coproc>
<type>CUDA</type>
<count>0.5</count>
</coproc>


It can be worth increasing the CPU use from the default 0.04 to say 0.1

That won't change how much CPU is used, Boinc uses that value to manage how many tasks it runs, once 10 of those 0.1 CPUs are in use, Boinc will just run one less CPU task

It is also worth freeing a CPU core - this is best done from the BOINC manager tool - set the "use xx% of CPU to free one or two cores - a bit of simple arithmetic to work out what xx% should be, as that depends on the total number of cores (real and virtual) your CPU has.

That is only required for the OpenCL apps, since NRauer has never run Astropulse v6 either on CPU or GPU, he doesn't really need to do that.

What he can do is change the priority of the v7 Cuda app from the default 'Below Normal' to 'Normal', or 'Above Normal' by editing the mbcuda.cfg removing the ; from the processpriority line,
and changing the priority on that line to what he wants, For Example:

;;; This configuration file is for optional control of Cuda Multibeam x41zc
;;; Currently, the available options are for
;;; application process priority control (without external tools), and
;;; per gpu priority control (useful for multiple Cuda GPU systems)
[mbcuda]
;;;;; Global applications settings, to apply to all Cuda devices
;;; You can uncomment the processpriority line below, by removing the ';', to engage machine global priority control of x41x
;;; possible options are 'belownormal' (which is the default), 'normal', 'abovenormal', or 'high'
;;; For dedicated crunching machines, 'abovenormal' is recommended
;;; raising global application priorities above the default
;;; may have system dependant usability effects, and can have positive or negative effects on overall throughput
processpriority = normal
;;; Pulsefinding: Advanced options for long pulsefinds (affect display usability & long kernel runs)
;;; defaults are conservative.
;;; WARNING: Excessive values may induce display lag, driver timeout & recovery, or errors.
;;; pulsefinding blocks per multiprocessor (1-16), default is 1 for Pre-Fermi, 4 for Fermi or newer GPUs
;pfblockspersm = 8
;;; pulsefinding maximum periods per kernel launch (1-1000), default is 100, as per 6.09
;pfperiodsperlaunch = 200

;[bus1slot0]
;;; Optional GPU specifc control (requires Cuda 3.2 or newer app), example
;processpriority = abovenormal
;pfblockspersm = 8
;pfperiodsperlaunch = 200


Claggy
ID: 1421976 · Report as offensive
Profile RAUERTECH

Send message
Joined: 7 May 09
Posts: 6
Credit: 575,382
RAC: 2
United States
Message 1422099 - Posted: 29 Sep 2013, 19:55:14 UTC
Last modified: 29 Sep 2013, 19:56:54 UTC

Thank you! ok so I have it set to .33 and I am running 3 GPU WU @ 0.04CPUs + .33 NVIDIA GPUs. I checked my GPU Memory ands its at about 900MB. My card is a 2GB Card. So could I run 4WU? If so what number do I set it to?
ID: 1422099 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1422107 - Posted: 29 Sep 2013, 20:11:06 UTC

Its not about GPU memory, its about how much parallel processing the GPU can do, and how well it can transfer data around the whole system. On some systems (mine being an example) increasing the default 0.04CPU improves the performance - you have to try things and see what works best for you.
If you start off low (1 per GPU) and get an idea over several days how things are doing, then do one change by setting two per GPU, allow that a few days to settle you will get a feeling for when things are reaching the limits. For most systems the optimum is between two and three WU per GPU.

(BTW - Claggy's correction of "COUNT" is the correct parameter to change - I realised what I'd written was wrong while heading away from home, and got home to see Claggy had posted his correction - thanks)
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1422107 · Report as offensive
Profile RAUERTECH

Send message
Joined: 7 May 09
Posts: 6
Credit: 575,382
RAC: 2
United States
Message 1422123 - Posted: 29 Sep 2013, 21:07:22 UTC - in response to Message 1422107.  

Is this what I need to change to .1? <avg_ncpus>0.04</avg_ncpus>
ID: 1422123 · Report as offensive
Profile RAUERTECH

Send message
Joined: 7 May 09
Posts: 6
Credit: 575,382
RAC: 2
United States
Message 1422127 - Posted: 29 Sep 2013, 21:14:45 UTC

Ok so I did edit the mbcuda file. it was already set to abovenormal but I removed the ; from the beginning of the line. Can someone look into making a video on youtube on how this is all done. or I will once I figure it all out.
ID: 1422127 · Report as offensive
Profile Fred E.
Volunteer tester

Send message
Joined: 22 Jul 99
Posts: 768
Credit: 24,140,697
RAC: 0
United States
Message 1422134 - Posted: 29 Sep 2013, 21:34:15 UTC
Last modified: 29 Sep 2013, 21:35:28 UTC

Is this what I need to change to .1? <avg_ncpus>0.04</avg_ncpus>

Yes.
Ok so I did edit the mbcuda file. it was already set to abovenormal but I removed the ; from the beginning of the line. Can someone look into making a video on youtube on how this is all done. or I will once I figure it all out.

The semicolon made it a comment line and the line was ignored until you removed it, so the tasks were running at below normal (default) before your change. I think a video would be useful and have not heard that one already esists.
Another Fred
Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop.
ID: 1422134 · Report as offensive
Profile RAUERTECH

Send message
Joined: 7 May 09
Posts: 6
Credit: 575,382
RAC: 2
United States
Message 1422735 - Posted: 30 Sep 2013, 23:22:26 UTC

So I rAN WITH 3 wu per GPU. before the change I was averaging about 5000 PPD day. After installing Lunatics app and changing the gpu to .33 and changing it to above average I was getting around 3500 PPD. What am I doing wrong. Possible im using the wrong boinc? any tips would be great. Im going to be running my machine 24/7 for the cause and want to get the best bang for my buck. I might be buying another video card but I think I should be doing over 10,000 a day with my current setup. User MSATTlER is getting about 47,000 average credits per day on a machine with a core 2 quad and 2 gtx 580's. HOW THE HECK IS THIS POSSIBLE AND IF IT IS PLEASE HELP! lol thanks!
ID: 1422735 · Report as offensive
Profile Fred E.
Volunteer tester

Send message
Joined: 22 Jul 99
Posts: 768
Credit: 24,140,697
RAC: 0
United States
Message 1422757 - Posted: 1 Oct 2013, 0:35:07 UTC

So I rAN WITH 3 wu per GPU. before the change I was averaging about 5000 PPD day. After installing Lunatics app and changing the gpu to .33 and changing it to above average I was getting around 3500 PPD. What am I doing wrong. Possible im using the wrong boinc? any tips would be great. Im going to be running my machine 24/7 for the cause and want to get the best bang for my buck. I might be buying another video card but I think I should be doing over 10,000 a day with my current setup. User MSATTlER is getting about 47,000 average credits per day on a machine with a core 2 quad and 2 gtx 580's. HOW THE HECK IS THIS POSSIBLE AND IF IT IS PLEASE HELP! lol thanks!

Your machine is indeed capable of a higher daily average (usually called RAC for Recent Average Credit). Mine has a 670 and a 6 core cpu, so you should be able to reach the same level. It takes a while for RAC to build - it is a fairly long term average.

Nothing wrong with the BOINC version, I use the same one. Wonder why you're not crunching Astropulse - your applications detail page shows none attempted. Generally, AP pays more and that's how I've built up my numbers. But I can run 24k without AP. Maybe someone else will have some suggestions, but I think you're on the right track and just need some time and that 24/7 crunching.
Another Fred
Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop.
ID: 1422757 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1422759 - Posted: 1 Oct 2013, 0:39:25 UTC - in response to Message 1422735.  

So I rAN WITH 3 wu per GPU. before the change I was averaging about 5000 PPD day. After installing Lunatics app and changing the gpu to .33 and changing it to above average I was getting around 3500 PPD. What am I doing wrong. Possible im using the wrong boinc? any tips would be great. Im going to be running my machine 24/7 for the cause and want to get the best bang for my buck. I might be buying another video card but I think I should be doing over 10,000 a day with my current setup. User MSATTlER is getting about 47,000 average credits per day on a machine with a core 2 quad and 2 gtx 580's. HOW THE HECK IS THIS POSSIBLE AND IF IT IS PLEASE HELP! lol thanks!

Because of those Abandoned tasks you have the servers have restricted the amount of work that you get so just be patient and that will sort itself out, plus now that you are producing work faster your Pendings will also increase.

It'll take a few weeks for your rig to settle down so just relax and be patient as these things take time.

Cheers.
ID: 1422759 · Report as offensive
Profile RAUERTECH

Send message
Joined: 7 May 09
Posts: 6
Credit: 575,382
RAC: 2
United States
Message 1422765 - Posted: 1 Oct 2013, 0:58:29 UTC - in response to Message 1422759.  

ohh thanks I did not realize that? so I changed it back to just 1 wu per gpu. should I change it back then?
ID: 1422765 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1422790 - Posted: 1 Oct 2013, 22:17:45 UTC - in response to Message 1422765.  

ohh thanks I did not realize that? so I changed it back to just 1 wu per gpu. should I change it back then?

I'd change it up to doing 2 for a 2-3 weeks and see how it goes, then I'd up it to 3 for another 2-3 weeks and then I'd compare the results to work out what is the best for your rig.

Cheers.
ID: 1422790 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1422795 - Posted: 1 Oct 2013, 22:29:37 UTC

BTW, you don't need a free CPU core to run your card while just running SETI V7 work it.

My own 3570K runs 2 GTX660's, each running 3 V7 workunits, but I don't run AP's on them, though I'll take as many AP's (and V7's) as I can to keep all 4 CPU cores busy (freeing a CPU core on my setup drops production here).

Cheers.
ID: 1422795 · Report as offensive
Profile perryjay
Volunteer tester
Avatar

Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 20,676,751
RAC: 0
United States
Message 1422818 - Posted: 1 Oct 2013, 23:41:49 UTC

As they said, just be patient. I'm running an I-5 3330 with a GTX 650 SC and you can see where my RAC is now. I'm running optimized of course and two V7s or one AP and one V7 at a time on my GPU. The AP work really does bump you up if you can get them. :)

As to Mark Sattler, his machines are nice to aspire to but he does a lot to get them up where they are. I'm not sure what he has been doing lately but he has most of his machines super cooled and clocked to the max.


PROUD MEMBER OF Team Starfire World BOINC
ID: 1422818 · Report as offensive
j tramer

Send message
Joined: 6 Oct 03
Posts: 242
Credit: 5,412,368
RAC: 0
Canada
Message 1423260 - Posted: 2 Oct 2013, 20:32:57 UTC

I have been trying to do this also...I cant find these files to edit?

I have changed my files show I see all hidden files, I still cant find these files? what am I doing wrong here?
ID: 1423260 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1423272 - Posted: 2 Oct 2013, 20:44:30 UTC - in response to Message 1423260.  
Last modified: 2 Oct 2013, 20:45:25 UTC

I have been trying to do this also...I cant find these files to edit?

I have changed my files show I see all hidden files, I still cant find these files? what am I doing wrong here?

Look at your Boinc startup from the Event Log, that will tell you where your Boinc Data directory is, it should be where i posted earlier:

the default Boinc Data directory location for Windows 7 is: C:\ProgramData\BOINC\ and the full location for the project folder is: C:\ProgramData\BOINC\projects\setiathome.berkeley.edu


Claggy
ID: 1423272 · Report as offensive
j tramer

Send message
Joined: 6 Oct 03
Posts: 242
Credit: 5,412,368
RAC: 0
Canada
Message 1423275 - Posted: 2 Oct 2013, 20:47:49 UTC

got it.... ;0

thxs for the help ;0
ID: 1423275 · Report as offensive
j tramer

Send message
Joined: 6 Oct 03
Posts: 242
Credit: 5,412,368
RAC: 0
Canada
Message 1423345 - Posted: 2 Oct 2013, 23:37:16 UTC

now my gpu is using about 98 percent vs 65 percent.....had to adjust my fans on my gpu to spin faster
ID: 1423345 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : How do I configure seti@home to run multiple WU on one 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.