Windows CUDA 4.0 and CUDA 4.2 Versions
log in

Advanced search

Message boards : News : Windows CUDA 4.0 and CUDA 4.2 Versions

1 · 2 · Next
Author Message
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 14412 - Posted: 9 Jul 2012, 18:35:18 UTC

Due to a PBCAK (problem between chair and keyboard), the CUDA 4.2 apps were actually CUDA 4.0 and the wrong CUDA runtime DLL was included in the optimized applications zip file. This has now been corrected resulting in both CUDA 4.0 and CUDA 4.2 applications being available for 32 and 64-bit Windows.

Profile Overtonesinger
Avatar
Send message
Joined: 16 Jul 10
Posts: 21
Credit: 142,646,892
RAC: 2,366
Message 14424 - Posted: 10 Jul 2012, 19:20:20 UTC - in response to Message 14412.

Why Collatz suddenly needs new version of BOINC ??? I hate it...

The 6.12.34 is the best version!

Remember THIS: I had to uninstall the version 7 because it has STOPPED all GPU work units forever when any CPU work was running in high priority mode!!

Ok, ok, I will give the Boinc client version SEVEN one last try!... :)))
____________
Melwen - child of the Fangorn Forest

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 14425 - Posted: 10 Jul 2012, 21:20:55 UTC - in response to Message 14424.

Why Collatz suddenly needs new version of BOINC ??? I hate it...

The 6.12.34 is the best version!

Remember THIS: I had to uninstall the version 7 because it has STOPPED all GPU work units forever when any CPU work was running in high priority mode!!

Ok, ok, I will give the Boinc client version SEVEN one last try!... :)))


You shouldn't need to upgrade. The minimum boinc version was set to 66000 (6.60.00) instead of 60600 (6.06.00) when I added the latest CUDA apps.

It should now require 6.0.0 for CUDA, 6.6.0 for ATI, and 7.0.0 for OpenCL. If you continue to get the "please upgrade" messages, let me know.

Profile Nikolay A. Saharov
Send message
Joined: 17 Jun 09
Posts: 13
Credit: 40,327,207
RAC: 51,220
Message 14428 - Posted: 11 Jul 2012, 10:59:10 UTC

I have errors with the new application when BOINC suspends it and switches to another WU.

Please see my results http://boinc.thesonntags.com/collatz/result.php?resultid=120497285
____________

Morten
Avatar
Send message
Joined: 6 Oct 10
Posts: 2
Credit: 4,608,832
RAC: 0
Message 14442 - Posted: 12 Jul 2012, 18:32:28 UTC
Last modified: 12 Jul 2012, 19:28:03 UTC

Tasks are not paused gracefully when switching between projects - they are terminated with "computation error" status.

Just suspending the project renders the active tasks as lost....
____________
Morten Ross

Mark Henderson
Send message
Joined: 11 Aug 09
Posts: 6
Credit: 25,791,075
RAC: 0
Message 14444 - Posted: 12 Jul 2012, 21:53:11 UTC

Same here, suspending them at any point crashes them.

Ashley
Send message
Joined: 22 Jun 12
Posts: 11
Credit: 1,028,851
RAC: 0
Message 14445 - Posted: 13 Jul 2012, 1:56:05 UTC

Hi I'm trying the 64bit nivida app it seems to be working fine no problems, i tried to make it crash by suspending it and it did not crash but then I'm only running collatz on my nivida card.

sounds like the "playing nice with others" projects is not working if you don't have the latest drivers upgrade them then report what happens other wise there is a problem.

oh yeah my 9800gt is only work at 75% load why is it only working that much the old app worked at 90% - 95% load.

Ashley

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 14446 - Posted: 13 Jul 2012, 2:25:25 UTC - in response to Message 14444.

Same here, suspending them at any point crashes them.


So close.... I have a version working that now fixes the crashing but it is not resuming where where it left off.

Ashley
Send message
Joined: 22 Jun 12
Posts: 11
Credit: 1,028,851
RAC: 0
Message 14450 - Posted: 13 Jul 2012, 7:38:21 UTC

I finished one it did not crash

Anonymous platform (NVIDIA GPU cuda42)
Total time 9643.55 seconds
GPU time 8381.75 seconds

collatz v2.05 (cuda31)
Total 8,035.38 seconds
GPU 6,789.84 seconds

don't ask me if its faster these are the reported times.

The cuda42 had a huge stderr_txt output file its pages and pages of it you may want to look at it.

here is the link
http://boinc.thesonntags.com/collatz/result.php?resultid=120628831

Ashley

Morten
Avatar
Send message
Joined: 6 Oct 10
Posts: 2
Credit: 4,608,832
RAC: 0
Message 14459 - Posted: 13 Jul 2012, 22:46:08 UTC - in response to Message 14446.
Last modified: 13 Jul 2012, 22:46:43 UTC

Same here, suspending them at any point crashes them.


So close.... I have a version working that now fixes the crashing but it is not resuming where where it left off.


What version of what?

You've hidden your computers so we don't know if you're running latest drivers (301.42), 304.79 beta or a later science app than collatz_3.11_windows_x86_64__cuda42.exe.

As for the resume state - it's normal that it will not resume at the same percentage complete as when you pause - it rolls back to last commit/checkpoint.
____________
Morten Ross

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 14473 - Posted: 16 Jul 2012, 20:01:08 UTC - in response to Message 14459.

Same here, suspending them at any point crashes them.


So close.... I have a version working that now fixes the crashing but it is not resuming where where it left off.


What version of what?

You've hidden your computers so we don't know if you're running latest drivers (301.42), 304.79 beta or a later science app than collatz_3.11_windows_x86_64__cuda42.exe.

As for the resume state - it's normal that it will not resume at the same percentage complete as when you pause - it rolls back to last commit/checkpoint.


True, but I hate to have people lose an average of 2.5 minutes of crunching just because BOINC didn't set the time_to_checkpoint flag prior to killing the app so i'm trying to get it to do a checkpoint prior to letting BOINC kill the app to switch to another project. For an old 8400GS which takes hours to complete, that could reduce the elapsed time and GPU time by 15 minutes. Fast GPUs won't see as big a difference as some take only 20 minutes to comlete a WU which means only one in three WUs get suspended (if switching every 60 minutes) which means less than a minute lost on average.

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 14474 - Posted: 16 Jul 2012, 22:06:19 UTC

I built new CUDA 4.2 apps for Windows x86 and x64. Version is 3.13. It should fix the suspend/resume crash as well as checkpoint on suspend so it will pick up exactly where it left off when it resumes.

Ashley
Send message
Joined: 22 Jun 12
Posts: 11
Credit: 1,028,851
RAC: 0
Message 14479 - Posted: 18 Jul 2012, 1:15:51 UTC

where do you download the new app i can't find it in the Power (Optimized) Applications page.

Ashley

Profile arkayn
Volunteer tester
Avatar
Send message
Joined: 30 Aug 09
Posts: 219
Credit: 676,877,192
RAC: 21,486
Message 14480 - Posted: 18 Jul 2012, 3:26:47 UTC - in response to Message 14479.

where do you download the new app i can't find it in the Power (Optimized) Applications page.

Ashley


Looks like he has released it as the stock build.

http://boinc.thesonntags.com/collatz/download/collatz_3.13_windows_intelx86__cuda42.exe
http://boinc.thesonntags.com/collatz/download/collatz_3.13_windows_x86_64__cuda42.exe

To use them you will have to edit the app_info and change the numbers.
____________

Ashley
Send message
Joined: 22 Jun 12
Posts: 11
Credit: 1,028,851
RAC: 0
Message 14481 - Posted: 18 Jul 2012, 4:44:14 UTC

oh, he could have said so.

Ashley

All good.

ph2000
Send message
Joined: 1 Nov 10
Posts: 4
Credit: 939,974,286
RAC: 1,281,011
Message 14506 - Posted: 23 Jul 2012, 21:13:52 UTC

Some days ago I got automatically some CUDA 4.2 tasks for mini_collatz, then it changed back to the slower cuda23.
Is it possible to change the default mini_collatz application to 4.2?

Profile Robert Pick
Send message
Joined: 17 Oct 10
Posts: 13
Credit: 41,134,618
RAC: 632
Message 14507 - Posted: 23 Jul 2012, 22:22:54 UTC

After running Einstein for several Mo. I decided for a change, Collatz came in first.In between runs I upgraded Nvidia to (301.42) and Boincto 7.28. Everything else I run didn't seem to mind the changes. This afternoon when I made the decision to run Collatz I hit the download button and got 77 WU'ts. I let E@H finish it's last WU and let Collatz rip. Oh, What A Mess!!!! They all error out before they even start to count. What's up? Pick

diederiks
Send message
Joined: 17 Feb 10
Posts: 4
Credit: 1,956,705
RAC: 0
Message 14510 - Posted: 24 Jul 2012, 12:38:18 UTC

Same problem here on same driver en boinc version, runing on Win XP with a GF460SE All other GPU project have no problem (seti(beta),GPU Grid, Einstein, primegrid, milkyway all running without a failing WU)

pvan3l
Send message
Joined: 15 Sep 09
Posts: 1
Credit: 167,442,188
RAC: 315,976
Message 14512 - Posted: 24 Jul 2012, 16:45:08 UTC

Same here with boinc 7.0.28 XP 32bit SP3 Nvidia GTX275 Driver 301.42 Cuda 4.2.1.
All WU Crash.

No problems with ATI cards and drivers

Profile Robert Pick
Send message
Joined: 17 Oct 10
Posts: 13
Credit: 41,134,618
RAC: 632
Message 14514 - Posted: 25 Jul 2012, 0:16:25 UTC
Last modified: 25 Jul 2012, 0:18:35 UTC

Tried once again. Got 63 to crunch. Same result!!! (7/24/2012 17:04:00 | Collatz Conjecture | Sending scheduler request: Requested by user.
7/24/2012 17:04:00 | Collatz Conjecture | Requesting new tasks for NVIDIA
7/24/2012 17:04:04 | Collatz Conjecture | Scheduler request completed: got 63 new tasks
7/24/2012 17:06:20 | Collatz Conjecture | Scheduler request completed
7/24/2012 17:06:20 | Collatz Conjecture | [error] garbage_collect(); still have active task for acked result collatz_2376194880730634627432_103079215104_0; state 5
7/24/2012 17:06:21 | Collatz Conjecture | Computation for task collatz_2376194880730634627432_103079215104_0 finished
7/24/2012 17:06:21 | Collatz Conjecture | Output file collatz_2376194880730634627432_103079215104_0_0 for task collatz_2376194880730634627432_103079215104_0 absent

Note;[error] garbage_collect. Hummmm! Wonder what that means? I just aborted all of them and will wait for some kind of fix before attempting to run anymore. Pick

1 · 2 · Next
Post to thread

Message boards : News : Windows CUDA 4.0 and CUDA 4.2 Versions


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.