Advanced search

Message boards : News : New WUs of TRYP (*adTRYP*)

Author Message
ignasi
Send message
Joined: 10 Apr 08
Posts: 254
Credit: 16,836,000
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 24207 - Posted: 2 Apr 2012 | 13:03:47 UTC

We have submitted a new batch of TRYP (*adTRYP*), significantly longer than the latest TRYP (met*, eth*, pro*,...).

The latest TRYP were a real production attempt to straighforwardly obtain binding for small inhibitor molecules. It finally proved harder than first expected, therefore we are improving the underlying analysis method as well as implementing a cleverer scheme to obtain the data. That is what *adTRYP* are for, to prepare a systematic methodological study on how to produce efficiently many events of binding for small protein inhibitors. See experiments section for more info.

Cheers

5pot
Send message
Joined: 8 Mar 12
Posts: 411
Credit: 2,083,882,218
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 24209 - Posted: 2 Apr 2012 | 14:37:30 UTC

Are these long runs? I see roughly 900 long runs have been added overnight. And off topic, but any new word on how the 680 is performing?

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 7,520
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 24211 - Posted: 2 Apr 2012 | 16:38:48 UTC - in response to Message 24207.

I have one of these *adTRYP* running on my GTX480@800MHz. It's computation reached 10% in 55 minutes and 40 seconds. It's running with 75% GPU usage (Swan_sync=0 applied, WindowsXP 32bit). So it will finish on my GTX480@800MHz in approximately 9 hours and 17 minutes.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 24212 - Posted: 2 Apr 2012 | 16:43:20 UTC - in response to Message 24209.
Last modified: 2 Apr 2012 | 16:49:33 UTC

Yes, Ignasi's adTRYP tasks are Long; I'm expecting one to finish in ~13.5h (3h 38 @ 28%) on a mildly overclocked GTX470 (680MHz), 1 CPU thread free, but not using SWAN_SYNC. Getting ~66% GPU utilization (2003 R2 x64 server). I guess SWAN_SYNC would do more for these tasks, as might freeing an additional thread. These are not for entry level cards!

OT, the app is being tested in house.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

5pot
Send message
Joined: 8 Mar 12
Posts: 411
Credit: 2,083,882,218
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 24215 - Posted: 2 Apr 2012 | 17:51:40 UTC

on a 570 (non-OC), on W7 I too am only getting 56-58% utilization, with one core free.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 24217 - Posted: 2 Apr 2012 | 22:11:29 UTC - in response to Message 24215.

W7 tends to cost you 11% or more.

I turned SWAN_SYNC on, restarted and freed up 3 CPU threads (really just 2 as SWAN eats one up). Utilization rose to ~72%. Total run time should now be around 12.5h.
When I freed up all the CPU threads the GPU utilization rose to 79%.
72/66 is about 9% faster and 79/66 is around 19% faster, so the system can be tweaked to improve GPU performance, if you so desire.
An OC'ed CPU with HT switched off might improve GPU performance further, but you have to find your own balance in these situations. I'm happy to free 2 or 3 cores but not to switch off HT, OC the CPU, or free up all the CPU cores. If I had a couple of GTX 580's in the system, I expect all of the above would be implemented.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

wiyosaya
Send message
Joined: 22 Nov 09
Posts: 114
Credit: 589,114,683
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 24218 - Posted: 3 Apr 2012 | 2:55:49 UTC

I am interested in checking thread / cpu utilization, however, I have no info on how to do so. Is there a link on how to check thread utilization? Thanks.

I have one of these running on my GTX 460. I expect it to complete in a little over 20 hours.
____________

wiyosaya
Send message
Joined: 22 Nov 09
Posts: 114
Credit: 589,114,683
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 24242 - Posted: 4 Apr 2012 | 16:49:53 UTC

My GTX 460 did complete the WU in about 20-hours.

It then took approximately another 10 hours for the WU to successfully upload. It only successfully uploaded after I returned home from work, and sat and pressed the retry button for about a half-hour.

This is why I find the 24-hour limit on first tier bonus credits distasteful as I feel this penalizes people in my situation, and people in my situation have no control over problems with uploading results. I've done everything in my power to correct this without success as evidenced by this thread - http://www.gpugrid.net/forum_thread.php?id=2713

http://www.gpugrid.net/workunit.php?wuid=3318070

____________

Michael Kingsford Gray
Avatar
Send message
Joined: 28 Nov 11
Posts: 21
Credit: 121,646,463
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwat
Message 24265 - Posted: 6 Apr 2012 | 5:56:37 UTC

178-IBUCH_adTRYP_120402-0-5-RND1445_5
Worked quite OK on my machine, after having previously failed on 4 others machines.
And it only took about 9.6 hours to run to completion.
Do you think that my (unusually huge) amount of GPU RAM (3Gb on *each* GTX580) might be the clue here?
Or could it be the 12 cores, or the 24Gb of CPU RAM?
Dunno.
But no "large" task has failed to run to completion for some months now, if ever.
I must be getting *something* right that differs from others'?

Profile dskagcommunity
Avatar
Send message
Joined: 28 Apr 11
Posts: 456
Credit: 817,865,789
RAC: 0
Level
Glu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 24266 - Posted: 6 Apr 2012 | 8:26:04 UTC

Perhaps the Drivers?
____________
DSKAG Austria Research Team: http://www.research.dskag.at



Michael Kingsford Gray
Avatar
Send message
Joined: 28 Nov 11
Posts: 21
Credit: 121,646,463
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwat
Message 24267 - Posted: 6 Apr 2012 | 9:34:05 UTC - in response to Message 24266.

Possibly.
I have resisted the urgings of nVidia and kept them as they are (296.10)
If it ain't broken, don't fix it.

On checking: two of the four failed tasks are running on the same revision drivers.

That cannot be it?

Michael Kingsford Gray
Avatar
Send message
Joined: 28 Nov 11
Posts: 21
Credit: 121,646,463
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwat
Message 24268 - Posted: 6 Apr 2012 | 9:35:36 UTC - in response to Message 24267.

If you are interested enough, check workunit:
www.gpugrid.net/workunit.php?wuid=3317664

Michael Kingsford Gray
Avatar
Send message
Joined: 28 Nov 11
Posts: 21
Credit: 121,646,463
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwat
Message 24269 - Posted: 6 Apr 2012 | 9:42:37 UTC - in response to Message 24268.

I just noticed something passing strange about driver revisions:
GPUGrid reports
"[2] NVIDIA GeForce GTX 580 (3034MB) driver: 280.26"

Nah, forget my previous blatherings about the driver being 296.10!
I was reading the incorrect row.
They are 280.26, which NONE of the failed workstation have.

You might be correct, after all!

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 24270 - Posted: 6 Apr 2012 | 11:20:30 UTC - in response to Message 24268.

If you are interested enough, check workunit:
www.gpugrid.net/workunit.php?wuid=3317664


So why did they fail?

5196126 122094 2 Apr 2012 | 15:35:32 UTC 2 Apr 2012 | 15:38:09 UTC Error while computing 4.21 0.05 ---
5197855 122149 2 Apr 2012 | 16:23:21 UTC 2 Apr 2012 | 16:25:54 UTC Error while computing 4.10 0.03 ---
5198024 32158 2 Apr 2012 | 17:17:13 UTC 3 Apr 2012 | 18:22:26 UTC Error while computing 13,789.31 1,136.44 ---
5202073 122222 3 Apr 2012 | 19:00:42 UTC 4 Apr 2012 | 18:04:59 UTC Error while computing 17,690.41 1,820.02 ---
5205771 122321 4 Apr 2012 | 19:38:28 UTC 5 Apr 2012 | 1:11:27 UTC Error while computing 16,493.75 2,073.30 ---
5206989 112231 5 Apr 2012 | 3:03:03 UTC 5 Apr 2012 | 22:02:13 UTC Completed and validated 34,510.66 34,305.45 64,200.00

From top down,
Coprocessors NVIDIA GeForce GTS 450 (1024MB) driver: 295.73
Operating System Microsoft Windows 7 (295 driver on W7)
Coprocessors NVIDIA GeForce GTX 580 (1536MB) driver: 296.10
Operating System Microsoft Windows 7 (295 driver on W7)
Coprocessors NVIDIA GeForce 8800 GT (512MB) driver: 285.62
Operating System Microsoft Windows Vista (CC1.1 GPU, shouldn't be attached in my opinion, let alone crunching long tasks)
Coprocessors NVIDIA GeForce 8800 GT (1023MB) driver: 296.10
Operating System Microsoft Windows XP (ditto)
The last task completed because it ran on a good card and used a good driver.

If you look here you will see 114 successful tasks, and no failures!

Contrast this system. An all too common bad setup; CC1.1 card, running long tasks, keeping a high cache, failing most tasks, using an alpha Boinc client, also using W7 and one of the unrecommended drivers.

This system, is counterproductive, as is this one.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Michael Kingsford Gray
Avatar
Send message
Joined: 28 Nov 11
Posts: 21
Credit: 121,646,463
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwat
Message 24271 - Posted: 6 Apr 2012 | 11:32:49 UTC - in response to Message 24270.

If you look here you will see 114 successful tasks, and no failures!

Quite.
That is my pooter.

Michael Kingsford Gray
Avatar
Send message
Joined: 28 Nov 11
Posts: 21
Credit: 121,646,463
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwat
Message 24272 - Posted: 6 Apr 2012 | 11:40:45 UTC - in response to Message 24270.

But...

Operating System Microsoft Windows 7 (295 driver on W7)
Coprocessors NVIDIA GeForce GTX 580 (1536MB) driver: 296.10

Is the failure down to the driver revision, or the fact that the card has 1/3 the memory of mine?
Or both?

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 24273 - Posted: 6 Apr 2012 | 11:47:05 UTC - in response to Message 24272.

The driver is probably the reason. Memory tends to not be such an issue, especially on high end cards (1GB+ GDDR). I think the card has 1/2 the memory of your cards.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Michael Kingsford Gray
Avatar
Send message
Joined: 28 Nov 11
Posts: 21
Credit: 121,646,463
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwat
Message 24274 - Posted: 6 Apr 2012 | 11:56:19 UTC - in response to Message 24273.

Might I suggest then, that operations might be enhanced by encouraging participants who have buggy-drivers, (No, not those with horse-and-cart!), to adopt a more friendly version?
As you know who they are, as well as their contact details, surely it would not be too onerous to create a script that 'points' each user to the appropriate download page?
And advising those with inappropriate task-selection/Graphics-card combinations to change? Or changing their selections of task types automatically, even?

GPUGrid may well be missing a golden opportunity by not so doing.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 24275 - Posted: 6 Apr 2012 | 12:24:14 UTC - in response to Message 24274.

All good positive suggestions, but all down to the researchers to move forward/implement, or not.
I'm just a forum mod and tester, I don't have access to emails and I can't alter users settings. I did suggest that moderators be allowed to alter settings under certain circumstances, such as continuous failures or being asked to by the cruncher. However, even if we could do these things we could not install drivers. We can't select task types here, so when we get failures on one task type, they just keep coming! I think a server side system to prevent this should be considered. We don't loose a lot of credit when a task fails in the first few seconds, but long tasks in particular do eat up lots of 'premium' bandwidth.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Post to thread

Message boards : News : New WUs of TRYP (*adTRYP*)

//