techPowerUp! Forums

techPowerUp! Forums (http://www.techpowerup.com/forums/index.php)
-   World Community Grid (WCG) (http://www.techpowerup.com/forums/forumdisplay.php?f=68)
-   -   Guide to running multiple GPU work units (http://www.techpowerup.com/forums/showthread.php?t=174138)

KieX Oct 21, 2012 09:48 PM

Guide to running multiple GPU work units
 
DISCLAIMER: Use this at your own risk! It is entirely down to each user to update as needed!
  • Users of app_info.xml method: This method will not be supported by WCG in the future. Please check the prerequisites list below and use the new setup guide. This is much simpler, with no AMD/NVIDIA specific code and avoids a lot of issues we encountered in the past


Index:
1) Why run multiple GPU work units?
2) Prerequisites
3) Auto-setup utility for app_config.xml
4) Manual setup guide for app_config.xml
5) Troubleshooting
6) Old app_info method


1) Why run multiple GPU work units?
By default World Community Grid is setup to run a single GPU WU on your machine. This guide will show you how to get better performance from your GPU by running more than 1 WU at a time. This can increase the output of both your PPD (point per day) and the runtime per day. This is achieved by creating a file called app_config.xml to detail how to use your hardware


2) Prerequisites
Before creating the app_config.xml file you will need:
Quote:

*NOTE* If you were using app_info.xml before:
Please hit "No new tasks" and complete your WU. Once your task list is clear, please remove the app_info.xml file and update BOINC to version 7.0.40 or above. Do not update before you clear the tasks as you may encounter problems submitting finished work units.
  1. Install BOINC version 7.0.40 or above. Currently 7.0.42 is available as BETA but should be stable.
    Download page for all versions: http://boinc.berkeley.edu/download_all.php

  2. Check you have the latest NVIDIA or AMD drivers for your graphics card. (Complete install is best to make sure all needed files are there)
    Download page for NVIDIA: http://www.nvidia.co.uk/Download/index.aspxDownload page for AMD: http://support.amd.com/us/gpudownload/Pages/index.aspx
    For AMD crunchers under Windows 8: You will need to use 13.2beta drivers or higher (13.2b4 works for sure) to avoid instant WU errors.

  3. Make sure your graphics card is supported! The link below shows a list of cards that are NOT compatible:
    https://secure.worldcommunitygrid.or...rtName=GPU#610

  4. Login to the WCG website and check your project lists include HCC and that the option to do GPU work is ticked on the profile assigned to your computer.


3) Auto-setup utility for app_config.xml
The fastest way to setup the file is using m&m's easy to use utility. Enter your desired numbers and everything will be created for you in the right place. Remember to restart your BOINC Manager aftwerwards in order for the file to be read correctly:

Link to post / file download link: Utility to run multiple GPU work units
Quote:

Originally Posted by m&m's (Post 2813484)
Hello everybody, I did a utility to set all the settings required faster.
The utility works with Vista/7/8 and with the new app_config.xml (BOINC version 7.0.40 or above).

Basically, it creates the app_config.xml with all your settings.


4) Manual setup guide for app_config.xml
If you would prefer to make all changes manually you can follow the steps listed below. (If you've already used the auto-setup in previous point, you can skip this entirely).
  1. Alternatively if you prefer to manually edit your file the nextOpen notepad or other text editor and copy the following code into it:
    Quote:

    Originally Posted by agent00skid (Post 2805712)
    Code:

    <app_config>
      <app>
          <name>hcc1</name>
          <max_concurrent>4</max_concurrent>
          <gpu_versions>
              <gpu_usage>.25</gpu_usage>
              <cpu_usage>1</cpu_usage>
          </gpu_versions>
      </app>
    </app_config>


  2. Edit the following according to your setup:
    • max_concurrent: This specifies the maximum number of WU the app can run at a time. Best to set this to the number of total HCC1 WU your computer will be doing at a time.

    • gpu_usage: This specifies the % use of the GPU per WU. Set this according to how many WU you want to run per GPU.
      1 / Desired WU per GPU = gpu_usage

    • cpu_usage: This specifies the % use of CPU threads per GPU WU. The default value is 1, but if you want to run more GPU WU than you have CPU cores edit it accordingly.
      Total CPU Threads / Total GPU WU = cpu_usage

  3. Save the file as "app_config.xml" into your WCG project folder.
    (Default Windows 7: C:\Program Data\BOINC\Data\projects\www.worldcommunitygrid.org)
    (Default Windows XP: C:\Documents and Settings\All Users\Application Data\BOINC\Data\projects\www.worldcommunitygrid.org)

  4. Open BOINC and you're done!
Check the event log (ctrl+shift+e) for any errors. I found that in switching from app_info it sometimes doesn't download any WU but once you hit the "update" button a few times it all looks normal. If you get an error about app_name not found in file, check the numbers you entered are correct and perhaps reset the project.

max_concurrent can be used to limit the HCC WU. For example if you have 8 CPU threads and want to use 2 for other projects you can set the value for HCC max_concurrent to 6.


5) Troubleshooting
  • How many WU to run?
    Your mileage may vary, as factors like CPU speed and numer of threads have an impact on performance. The best way to determine what works best for you is to find out how long a single WU takes then see how much longer it takes to complete multiple WU. Below is a simple list for ballpark figures:
    - Most NVIDIA: 2-3 WU
    - AMD 6970: 4 WU
    - AMD 7770: 3-4 WU
    - AMD 79xx: 8-12 WU

  • Mixed/Multi GPU
    If you're using mixed cards by default BOINC uses the best one, and in some cases may not use all your GPU even if they're the same. To use more than one GPU in the same machine go to the BOINC data folder (normally: C:\Program Data\BOINC) and look for file "cc_config.xml". If it doesn't exist, create it but the contents should have the following:

    Code:

    <cc_config>
    <options>
    <use_all_gpus>1</use_all_gpus>
    </options>
    </cc_config>

  • Driver crahses:
    If you are having driver crahses, the following registry modification might prevent that:
    (Source: Bun-Bun from XS)
    Code:

    Windows Registry Editor Version 5.00

    [HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Watchdog]
    "DisableBugCheck"="1"

    [HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Watchdog\Display]
    "EaRecovery"="0"



6) Old app_info method
This is for reference only. Not supported in future.
Spoiler


Updates to this tutorial
03-Feb-'13: Updates to utility link and prerequisites
03-Jan-'13: Setup utility by m&m included as preferred method
22-Dec-'12: Full OP update for app_config method.
12-Dec-'12: Updated the NVIDIA code to correct default settings
10-Dec-'12: Added info for multiple GPU setups
26-Nov-'12: Added links to "all-project" app_info files
14-Nov-'12: Added HCC GPU & CPU code
10-Nov-'12: Updated Version number on all templates
10-Nov-'12: Added CPU settings information

Sources:
http://www.worldcommunitygrid.org/fo...3962_offset,90
http://www.xtremesystems.org/forums/...app_info-files
(Credit to original posters, this is is just a simplified version to share with this community.)

[Ion] Oct 21, 2012 09:53 PM

Thank you for posting this here, it looks useful! :toast: :respect:

I'll add it to my Essentials thread :)

KieX Oct 21, 2012 09:55 PM

Quote:

Originally Posted by [Ion] (Post 2753208)
Thank you for posting this here, it looks useful! :toast: :respect:

I'll add it to my Essentials thread :)

Thank you [Ion], you did after all help me with finding some of this. :toast:

Bow Oct 21, 2012 10:07 PM

:respect:

stinger608 Oct 21, 2012 10:10 PM

That is awesome information KieX!!! Thanks a ton for sharing all the info!

This should be a sticky for sure!!!

KieX Oct 21, 2012 10:13 PM

This is what the .5 setting looks like on a single 7770:

http://img.techpowerup.org/121021/Capture.png

Roughly the same time it used to take for a single WU, but doing 2. Hopefully if this works for others it would also mean a huge boost for TPU. :rockout:

ThE_MaD_ShOt Oct 21, 2012 10:17 PM

Just be careful loading up the wu's. You only want to load the gpu to around 95%. If you load it to much you will start erroring out the wu's. You just want to get the gpu to use it's idle time so you can maximize the crunching power. Also make sure you have good case air flow as the card is not going to get the break in between the wu's to cool down. It is going to steady be at 95 or so % no down time. And as Keix stated use at your own risk.

Bow Oct 21, 2012 10:20 PM

Whats the heat increase like on the cards?

[Ion] Oct 21, 2012 10:22 PM

Quote:

Originally Posted by KieX (Post 2753216)
This is what the .5 setting looks like on a single 7770:

http://img.techpowerup.org/121021/Capture.png

Roughly the same time it used to take for a single WU, but doing 2. Hopefully if this works for others it would also mean a huge boost for TPU. :rockout:

That's amazing!
If I get an ATI card for the i7-920 build I'll have to give this a try!

KieX Oct 21, 2012 10:28 PM

Quote:

Originally Posted by ThE_MaD_ShOt (Post 2753220)
Just be careful loading up the wu's. You only want to load the gpu to around 95%. If you load it to much you will start erroring out the wu's. You just want to get the gpu to use it's idle time so you can maximize the crunching power. Also make sure you have good case air flow as the card is not going to get the break in between the wu's to cool down. It is going to steady be at 95 or so % no down time. And as Keix stated use at your own risk.

It doesn't seem to be giving me a steady pattern, still loads, and unloads as before.. just loads higher each time. Will add your advice to the OP though.. that's a pretty good point regarding the errors and cooling.

Quote:

Originally Posted by Bow (Post 2753221)
Whats the heat increase like on the cards?

For my particular 7770s they were crunching at 45C load, and now they're 50C. GPU load went from about 70% to 91%


Of course.. each card will vary some may have a higher impact. ThE_MaD_ShOt's advice above regarding cooling is very much valid.

Norton Oct 21, 2012 11:13 PM

Works like a charm- running 3x wu's on my 7870 atm :toast:

Wondering if I set the count to .125 if I can run all 8 cores on gpu wu's?

KieX Oct 21, 2012 11:24 PM

Quote:

Originally Posted by Norton (Post 2753250)
Works like a charm- running 3x wu's on my 7870 atm :toast:

Wondering if I set the count to .125 if I can run all 8 cores on gpu wu's?

From all the posts I've seen, the maximum anyone used is 6 WU on a 7970. No idea if 8 will work or not, this is testing after all. But I'd recommend allowing a few hours in each setting to at least see if it gives any problems.

[Ion] Oct 21, 2012 11:26 PM

Quote:

Originally Posted by Norton (Post 2753250)
Works like a charm- running 3x wu's on my 7870 atm :toast:

Wondering if I set the count to .125 if I can run all 8 cores on gpu wu's?

Wow...3 on a HD7870....

How long are they taking to finish?

Norton Oct 21, 2012 11:29 PM

Quote:

Originally Posted by [Ion] (Post 2753259)
Wow...3 on a HD7870....

How long are they taking to finish?

About 20 seconds longer than single wu's (appr. 01:45 vs 01:25 to 99.415%)

[Ion] Oct 21, 2012 11:36 PM

Quote:

Originally Posted by Norton (Post 2753262)
About 20 seconds longer than single wu's (appr. 01:45 vs 01:25 to 99.415%)

o_O

KieX Oct 21, 2012 11:42 PM

Quote:

Originally Posted by Norton (Post 2753262)
About 20 seconds longer than single wu's (appr. 01:45 vs 01:25 to 99.415%)

So looks like you'll be reclaiming cherry pie then :roll:

brandonwh64 Oct 21, 2012 11:47 PM

I dont have this folder?

C:\ProgramData\boinc\projects\www.worldcommunitygrid.org

Norton Oct 22, 2012 12:03 AM

Quote:

Originally Posted by KieX (Post 2753277)
So looks like you'll be reclaiming cherry pie then :roll:

Possibly tomorrow but not today..... 2 of my remote rigs report in the am so I have a shot at 100k tomorrow :D

KieX Oct 22, 2012 12:05 AM

Quote:

Originally Posted by brandonwh64 (Post 2753281)
I dont have this folder?

C:\ProgramData\boinc\projects\www.worldcommunitygrid.org

Is BOINC installed with default settings? In that case the ProgramData folder may be hidden. Enter the address into an explorer window.

If you can see the ProgramData folder but can't see the boinc folders.. then you probably installed it somewhere else like another drive?

Norton Oct 22, 2012 12:19 AM

We need to get t_ski to give these a try. iirc he's running 3x7970's in his main rig so he has a shot at the daily single cpu crunching record :cool:

ThE_MaD_ShOt Oct 22, 2012 12:21 AM

Quote:

Originally Posted by Norton (Post 2753250)
Works like a charm- running 3x wu's on my 7870 atm :toast:

Wondering if I set the count to .125 if I can run all 8 cores on gpu wu's?

When I was reading up on this last week there was a guy running 8 on a 7970 and he started getting errors. He had to back it down to 4. It was recommended to add 1 wu at a time until you get close to 95% load and stop there. Once you hit optimal and it takes sometime and patience and skill because you may have to start each wu manually, but you can get it to where one is starting right when another is finishing so there is no down time on the gpu. Also if you run more then on you can have it so there is one at 50% when the one is finishing and one is starting. I haven't tried it because my 6850's are already at 93% load with just one. But the 7850 is fair game as it is just over 80% load with one. When I get some time to play with it I will load that gpu up with more wu's.

brandonwh64 Oct 22, 2012 12:26 AM

Ok I got 4 WU on each card but it shows it taking 38 minutes per WU?

ThE_MaD_ShOt Oct 22, 2012 12:28 AM

Quote:

Originally Posted by brandonwh64 (Post 2753319)
Ok I got 4 WU on each card but it shows it taking 38 minutes per WU?

Card is loaded to hard. What is the load percentage? Also be careful of errors.

Norton Oct 22, 2012 12:28 AM

Quote:

Originally Posted by ThE_MaD_ShOt (Post 2753315)
When I was reading up on this last week there was a guy running 8 on a 7970 and he started getting errors. He had to back it down to 4. It was recommended to add 1 wu at a time until you get close to 95% load and stop there. Once you hit optimal and it takes sometime and patience and skill because you may have to start each wu manually, but you can get it to where one is starting right when another is finishing so there is no down time on the gpu. Also if you run more then on you can have it so there is one at 50% when the one is finishing and one is starting. I haven't tried it because my 6850's are already at 93% load with just one. But the 7850 is fair game as it is just over 80% load with one. When I get some time to play with it I will load that gpu up with more wu's.

I'm peaking @87% load and 60C max temps so I can definitely fit in a 4th wu. The peak load% only lasts for 5-6 seconds and then it drops into the 70's due to the stagger on the wu's.

I think I can squeeze in a 5th wu without causing errors but I don't think I'll push it THAT hard... :o

brandonwh64 Oct 22, 2012 12:30 AM

They have dropped to around 7m per wu. is that good?


All times are GMT. The time now is 05:55 AM.

Powered by vBulletin® Version 3.8.6
Copyright ©2000 - 2013, Jelsoft Enterprises Ltd.