Advanced search

Message boards : Wish list : Test Work Unit.

Author Message
Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 16452 - Posted: 20 Apr 2010 | 9:53:35 UTC

It would be useful if the team could put together a permanent test application/WU(s) for users setting up & testing systems; say a ten minute task (roughly) that users could run to test system stability or functionality before crunching the live data.

Perhaps a check box in GPUGRID Preferences, http://www.gpugrid.net/prefs.php?subset=project,
ACEMD
ACEMD Ver 2
ACEMD Beta
TEST ACEMD
TEST ACEMD Ver 2 (probably not needed)

That way we could choose Test Only and download the test work unit, crunch it for 10min or so and know if we have a stable system or not.
- Useful for testing New Systems, odd configurations, Over Clocking and of course diagnosing problems.

If these work units did not count towards the computers ratings and stats it would be free to fail many times (while being configured) and then still allow users to run the normal tasks afterwards. Perhaps this could be done so that the task just needs to be downloaded once; unless you think it might be a useful networking diagnostic tool, or the size would be so small that it would not matter to the server. Though, I would suggest a cap of say 15tasks per hour and perhaps 150 per day.

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1925
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 16455 - Posted: 20 Apr 2010 | 10:47:51 UTC - in response to Message 16452.

The only way would be to create a new project.
A bit too much of work.

gdf

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 16457 - Posted: 20 Apr 2010 | 11:34:26 UTC - in response to Message 16455.

Just a suggestion. If it would take days to setup I understand its not worth it.

Another suggestion,

Could this be included in an app?

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

It might push things along a bit and would help people that dont have either the skills or time to write a cc_config file themselves. I had to set 6 systems to report results immediately and one to use all cards.

Perhaps the cache would be better at 0.05days for GPUGRID than the default 0.25days?

Again, just suggestions.
Thanks,

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1925
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 16462 - Posted: 20 Apr 2010 | 13:34:50 UTC - in response to Message 16457.

Not really.
We can have a message thread for the best configuration for GPUGRID in the FAQ.
Post this text there in a new thread.

gdf

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2185
Credit: 15,826,317,025
RAC: 536,970
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18590 - Posted: 9 Sep 2010 | 21:40:40 UTC
Last modified: 9 Sep 2010 | 21:46:15 UTC

There is a need for some testing tool. It shouldn't be under BOINC at all. Just a couple of exe files (which represent the actual applications in GPUgrid) we could download, and set the desired running time (or number of test runs) at start. Even the result can be fixed, and checked at the end of each round.

Profile liveonc
Avatar
Send message
Joined: 1 Jan 10
Posts: 292
Credit: 41,567,650
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwat
Message 18592 - Posted: 9 Sep 2010 | 22:03:43 UTC - in response to Message 18590.
Last modified: 9 Sep 2010 | 22:04:48 UTC

A previous thread mentioned the possibility to choose long or short WU's, there already is a "Run test applications" option & there used to be an option to accept Beta WU's. The server already only sends Fermi WU's only to Fermi GPU's. So if there's an option for long, short, Beta & the server sends WU's according to CC1.1/1.3 or CC=<2.0 that's already plenty.
____________

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18596 - Posted: 9 Sep 2010 | 23:07:17 UTC - in response to Message 18592.

By "Test Work Unit" in this case we mean an app to test the stability of the crunchers system (mainly the GPU at various clock settings). Just really for crunchers to see if their system is stable crunching a GPUGrid task, before downloading and trying to tackle live data. Different from test applications for GPUGrid.
The idea is to save on the bandwidth, time and to prevent messing up your systems credibility here; if you get too many failures you stop getting new work, for a full day.
There are some programs you can use to do this to some extent, I posted one up some time ago, but you basically just need to run a GPU task/benchmark tool alongside GPUZ (or other tool) to check temperatures and so on. GPU failures might mean shutting down, start up again and change OC settings or you may just want to change the fan speeds. You could run the small MW tasks to test your setup, but it’s not quite the same, and has no actual GPUGrid benchmark.
Perhaps something could be done to attach Boinc to what is essentially a benchmarking project, for the purpose of testing, or perhaps GPUGrid could have a small separate stability-test task that you could choose to run; a task the just runs and never uploads, so as not to mess up your credit and stop you getting new work. If it just cycles through a loop (say 1min) it could be used to help tune your card and give a benchmark (one that could be used by reviewers).

Profile liveonc
Avatar
Send message
Joined: 1 Jan 10
Posts: 292
Credit: 41,567,650
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwat
Message 18597 - Posted: 9 Sep 2010 | 23:21:52 UTC - in response to Message 18596.

Wouldn't that still require GPUGRID.net to make a sub-project option since I've never heard of the boinc-client being able to run a task that isn't attached to a project & if that means that people will download this/these WU's, wouldn't the GPUGRID server have to cancel the upload once finished? But what about not permitting a finished WU from being uploaded & therefore would require a manual abort? Would that be convenient?
____________

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 18606 - Posted: 10 Sep 2010 | 11:08:20 UTC - in response to Message 18597.

There are probably several ways to do this.
Perhaps the easiest would be to create a benchmark/stability-test application, separate it from the rest of the GPUGrid projects (WRT points, uploading, and tracking success), allow the app to be downloaded (as a one off) with one work unit that is kept locally on the crunchers system. After that, if a cruncher configures their preferences via the web to only crunch the stability-test WU and did a GPUGrid update from the Boinc Projects Tab they would just run that WU repeatedly, with some details in the Boinc messages Tab - until they changed their preferences and did an update. A small, highly optimized task could allow crunchers to set good and reliable fixed fan speeds and overclock settings on their cards. It would also allow for an easy comparison of GPU card performances (on the one system), and lead to better optimization by the crunchers.
For crunchers with several systems, one web profile could be used for this; set the system you are testing to use School, for example, and the other systems to use different profiles. Configure School to just run the benchmark/stability-test application. Once you are finished, just move it to a different profile and update GPUGrid from the Boinc Projects Tab.

Post to thread

Message boards : Wish list : Test Work Unit.