CUDA only work.
log in

Advanced search

Message boards : Number crunching : CUDA only work.

Author Message
riptide
Send message
Joined: 7 Aug 09
Posts: 54
Credit: 1,060,610
RAC: 0
Message 442 - Posted: 8 Aug 2009, 0:08:52 UTC

Hi. I'm looking to possibly run CUDA (1.10) only work. By the looks of things to get CUDA I have to enable CPU work in teh preferences (thereby getting CPU WU's aswell (1.11). Am i right?

riptide
Send message
Joined: 7 Aug 09
Posts: 54
Credit: 1,060,610
RAC: 0
Message 541 - Posted: 13 Aug 2009, 4:08:27 UTC - in response to Message 442.

Well this was established in the other thread as true. Pity really.

Anyways, also I'd like to be able to download more than 10 WU's at any given time. I get 10GPU CUDA WU's and they are done in like just over 2 hours. :( And the client won't update itself I have to do it manually... and sometimes I have to sleep also. :D

Profile STE\/E
Avatar
Send message
Joined: 12 Jul 09
Posts: 581
Credit: 761,710,729
RAC: 0
Message 545 - Posted: 13 Aug 2009, 10:50:14 UTC - in response to Message 541.
Last modified: 13 Aug 2009, 11:02:34 UTC

Hi. I'm looking to possibly run CUDA (1.10) only work. By the looks of things to get CUDA I have to enable CPU work in teh preferences (thereby getting CPU WU's aswell (1.11). Am i right?


You may get a few of the CPU Wu's too, I know I did yesterday filling my Caches again but I just Aborted the few I got and haven't got any more since.


Well this was established in the other thread as true. Pity really.

Anyways, also I'd like to be able to download more than 10 WU's at any given time. I get 10GPU CUDA WU's and they are done in like just over 2 hours. :( And the client won't update itself I have to do it manually... and sometimes I have to sleep also. :D


I take what your saying is that you get 10 Wu's & then the Client won't call for more Wu on it's own. If that's the case I would think you have to either Up your Resource Share to the project or up your Contact Time.

All my box's that are running the Collatz Project are Quads & get their full Cache of 120 Wu's on their own without me doing anything because I have my Resource & Contact Times set high enough.

Rabinovitch
Avatar
Send message
Joined: 8 Aug 09
Posts: 20
Credit: 22,606,527
RAC: 19,737
Message 553 - Posted: 13 Aug 2009, 15:56:39 UTC

My great ask to project developers is to make at last possible run only cuda Wus, like in SETI. Then I could run cuda Wus constantly on my GTX260, without any magig with enable/disable CPU Wus... Is it so difficult?

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 556 - Posted: 13 Aug 2009, 17:38:44 UTC - in response to Message 553.

My great ask to project developers is to make at last possible run only cuda Wus, like in SETI. Then I could run cuda Wus constantly on my GTX260, without any magig with enable/disable CPU Wus... Is it so difficult?


Not if we dropped ATI support. ;-) Getting both ATI and CUDA to work well is taking more time than expected and where time is concerned, my family and two jobs take precedence. That doesn't leave much time for dealing with support questions, server and database administration, network administration, building and testing new apps, analyzing the data that has been returned, and re-writing the CUDA only server code so it supports both CUDA and ATI.

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3242
Credit: 1,691,599,600
RAC: 5,679,927
Message 560 - Posted: 13 Aug 2009, 21:06:10 UTC - in response to Message 541.
Last modified: 13 Aug 2009, 21:08:44 UTC

Well this was established in the other thread as true. Pity really.

Anyways, also I'd like to be able to download more than 10 WU's at any given time. I get 10GPU CUDA WU's and they are done in like just over 2 hours. :( And the client won't update itself I have to do it manually... and sometimes I have to sleep also. :D


You need to change Boinc so it either caches more units or connects more often. To do the later go into the Boinc Manager and change the Connect About every X days to something like 0.10 days. That way it connects every time it needs to instead of every 1.0 days for instance. I do this thru Advanced, Preferences, network usage. Be sure to say okay, this WILL overwrite any preferences you have setup on the website. So if you normally use the website to make changes do it there instead. It is under Your Account, Computing Preferences and then under Network Usage the first setting. Mine is set there to zero days on the 1st line and 0.25 on the 2nd line. I am having no troubles getting all the units I can handle even though most are finishing in under 5 minutes!
____________

riptide
Send message
Joined: 7 Aug 09
Posts: 54
Credit: 1,060,610
RAC: 0
Message 562 - Posted: 13 Aug 2009, 21:39:00 UTC - in response to Message 560.
Last modified: 13 Aug 2009, 21:40:22 UTC

Well AFIAK i takes BOINC up to 2 weeks to work out the projects "debt" so it might work later. I changed it earlier today to 0.1 day, but it still downloads 10 WU's.

Anyway, I haz a solution to BOTH problems.

I run rivatuner... so I go to the 'Schedule' tab on Rivatuner and I set it up to start a batch file when the GPU core temps drops below 60C (ie when all CUDA WU's are processed). Boinc will already be setup to NOT accept new taks.

The batch file that Rivatuner starts goes like this

chdir C:\program files\boinc\
boinccmd --project http://boinc.thesonntags.com/collatz allowmorework
boinccmd --project http://boinc.thesonntags.com/collatz update
ping 1.1.1.1 -n 1 -w 15000>nul
boinccmd --project http://boinc.thesonntags.com/collatz nomorework


That
1) Allows more work
2) Updates Project (ie. reports finished WU's + Download WU's)
3) Ping command serves as a delay of 15 seconds which is enough time to download my alotted WU's
4)Disallow new work.

The secret is that when ALL CUDA WU's are run, the 1st scheduler to operate is the GPU scheduler to download CUDA WU's. It takes about a minute for the CPU scheduler to kick in by which time the 'nomorework' command has kicked in. Hence I only ever get CUDA work.

That's what you call a Neat Kludge fix. :D

And it works for me, tested.

Have a nice day.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 582 - Posted: 14 Aug 2009, 18:08:01 UTC

I moved a new version of the scheduler into production. It should honor the "no cpu" preference setting so that it runs CUDA only. It will run both CUDA and GPU. It will run ATI. It will NOT run ATI and CPU.

The 6.5.4 client seems to do "generic" requests and I haven't figured out yet how to get it to download both ATI and CPU tasks. Using the 6.2.0 beta client, it requests 0 seconds of CPU and 0 seconds of CAL work. I think if it requested work for both like the CUDA version does, it would send both CPU and CAL work.

riptide
Send message
Joined: 7 Aug 09
Posts: 54
Credit: 1,060,610
RAC: 0
Message 583 - Posted: 14 Aug 2009, 18:18:11 UTC

Good news!

Bymark
Avatar
Send message
Joined: 28 Jul 09
Posts: 78
Credit: 587,923,886
RAC: 1,151,043
Message 584 - Posted: 14 Aug 2009, 18:32:20 UTC - in response to Message 582.

Thanks!
____________

Profile KWSN Checklist
Avatar
Send message
Joined: 25 Jul 09
Posts: 123
Credit: 1,287,549,992
RAC: 0
Message 588 - Posted: 14 Aug 2009, 22:26:16 UTC

Works as advertised here.

riptide
Send message
Joined: 7 Aug 09
Posts: 54
Credit: 1,060,610
RAC: 0
Message 736 - Posted: 28 Aug 2009, 15:34:49 UTC
Last modified: 28 Aug 2009, 15:35:28 UTC

Currently I'm getting no CUDA work.


EDIT: Ok now I am. Its back ok now.


Post to thread

Message boards : Number crunching : CUDA only work.


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.