Welcome to MilkyWay@home

Checkpoints for the GPU-applications


Advanced search

Message boards : Application Code Discussion : Checkpoints for the GPU-applications
Message board moderation

To post messages, you must log in.

AuthorMessage
jotun263

Send message
Joined: 24 Aug 09
Posts: 5
Credit: 519,653
RAC: 0
500 thousand credit badge12 year member badge
Message 31656 - Posted: 28 Sep 2009, 7:54:45 UTC

Hi!

Only a suggestion, but how about 1 or 2 checkpoints in the GPU-apps?!?!

I use the BOINC-Client 6.10.4 (dev. edition) 'cause the 6.6.38 made some troubles. Now I have the problem that every time the client gets new WUs, the active GPU-unit break up the calculation and the last new downloaded one starts. So a lot of precious calc time and energy (GPUs aren't energy savers!) get lost and some units restart up to 15 times until they get finished. Calc time up to 1 hour!!!

Or is there a workaround to prevent this behaviour?!

Greetz!
jotun263
ID: 31656 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfilePaul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
100 million credit badge13 year member badge
Message 31657 - Posted: 28 Sep 2009, 8:52:48 UTC - in response to Message 31656.  

Hi!

Only a suggestion, but how about 1 or 2 checkpoints in the GPU-apps?!?!

I use the BOINC-Client 6.10.4 (dev. edition) 'cause the 6.6.38 made some troubles. Now I have the problem that every time the client gets new WUs, the active GPU-unit break up the calculation and the last new downloaded one starts. So a lot of precious calc time and energy (GPUs aren't energy savers!) get lost and some units restart up to 15 times until they get finished. Calc time up to 1 hour!!!

Or is there a workaround to prevent this behaviour?!

Greetz!
jotun263

Move to 6.10.7 which fixes most of the issues with 6.10.4/.5/.6 ...
ID: 31657 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
50 million credit badge13 year member badgeextraordinary contributions badge
Message 31662 - Posted: 28 Sep 2009, 15:01:00 UTC
Last modified: 28 Sep 2009, 15:01:11 UTC

As Paul said, updating to a newer client solves this problem.

It doesn't make sense to checkpoint during the current WUs. Even a HD3870 completes the longest WUs in just over 2 minutes. A lot of people have set their preferences to 5 minutes as the minimum time between checkpoints, so there wouldn't be one written either way. Even a setting of only 60 seconds would be already too high for the HD4800 line, as they complete the WUs usually faster. And I'm not going to write checkpoints more often then set by the user (generally, checkpointing is done at discretion of the app and cannot be enforced by the BOINC client).
ID: 31662 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
jotun263

Send message
Joined: 24 Aug 09
Posts: 5
Credit: 519,653
RAC: 0
500 thousand credit badge12 year member badge
Message 31682 - Posted: 29 Sep 2009, 11:32:22 UTC - in response to Message 31662.  

Many thanks for the fast replies!

I will update to 6.10.7 when I come home and see what will happen.

Even a HD3870 completes the longest WUs in just over 2 minutes.
2 minutes?! I own a GTX280 and the calculation time of most WUs is around 6 minutes...
ID: 31682 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Emanuel

Send message
Joined: 18 Nov 07
Posts: 280
Credit: 2,442,757
RAC: 0
2 million credit badge14 year member badge
Message 31684 - Posted: 29 Sep 2009, 12:29:44 UTC
Last modified: 29 Sep 2009, 12:30:24 UTC

The latest version is 6.10.10 (6.10.9 on Linux); get it here. And yeah, AMD/ATI have been working on double precision cards for longer than nvidia.. upcoming cards will probably be a lot faster, especially considering nvidia's public statements that say they want to focus on GPGPU.
ID: 31684 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
jotun263

Send message
Joined: 24 Aug 09
Posts: 5
Credit: 519,653
RAC: 0
500 thousand credit badge12 year member badge
Message 31805 - Posted: 1 Oct 2009, 12:51:35 UTC - in response to Message 31684.  

I updated the client to 6.10.11 and everything work fine now. Even the calculation errors, which happen after WU restart, are gone. Thanks a lot!

upcoming cards will probably be a lot faster, especially considering nvidia's public statements that say they want to focus on GPGPU.
I'm waiting impatiently for the new NVIDIA-series. Extremly curious about the performance and the new features this series can provide for us... ;)
ID: 31805 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Application Code Discussion : Checkpoints for the GPU-applications

©2021 Astroinformatics Group