Advanced search

Message boards : News : New TRYP WUs

Author Message
ignasi
Send message
Joined: 10 Apr 08
Posts: 254
Credit: 16,836,000
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 23487 - Posted: 16 Feb 2012 | 15:53:52 UTC
Last modified: 16 Feb 2012 | 15:54:38 UTC

I am submitting 700 WUs in the long queue as part of a new experiment.
You'll note that they have the TRYP tag that was used for our publication in PNAS 2011:
metTRYP
ethTRYP
proTRYP
isoTRYP
butTRYP
penTRYP
hexTRYP

We are now extending the study to 7 different ligands to see if we can reconstruct their bindings too.

cheers,
i

Profile K1atOdessa
Send message
Joined: 25 Feb 08
Posts: 249
Credit: 370,186,977
RAC: 0
Level
Asp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 23504 - Posted: 17 Feb 2012 | 15:48:07 UTC - in response to Message 23487.

I've seen a few of these "TRYP" WU's come through my machines. They appear in terms of credit on par with the "TAKE" WU's, giving approximately 1 point per second of run time on my GTX570. This is considerably less than the "NATHAN" WU's who tend to give about 2 points per second of run time.

It appears, though, that while these "NATHAN" WU's use 90%+ of the gpu, the new "TRYP" WU"s hover around 50% utilization. That probably explains the credit -- the WU's give less credit because they appear to not be as efficient at utilizing the gpu. The same time is required, but only about 1/2 the actual processing is completed.

I don't care either way, but others may become upset if their RAC drops due to less efficient tasks than we're used to seeing recently.

ignasi
Send message
Joined: 10 Apr 08
Posts: 254
Credit: 16,836,000
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 23505 - Posted: 17 Feb 2012 | 16:18:20 UTC - in response to Message 23504.

True.

My WUs have been traditionally less efficient due to extra CPU-bound calculations during runtime. These TRYP Wus are no exception. We are applying forces to some atoms as part of complex protocols. Now, unfortunately and in relation to our crediting system based on FLOPS not on runtime, these WUs tend to give less credits per unit of time.

In exchange, I promise to produce amazing videos out of these simulations. So you can show them to your buddies. ;)

cheers,
i

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,835,229,774
RAC: 291,018
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 23507 - Posted: 17 Feb 2012 | 20:47:02 UTC - in response to Message 23505.

I'm seeing 75% GPU utilization from a GTX470 @680MHz on a 2003R2 x64 i7-2600 (HT on, 2 threads free) and using SWAN_SYNC=0, so ~90% of the CPU is actual being used. Expected runtime is ~6h40min.
____________
FAQ's

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

Profile K1atOdessa
Send message
Joined: 25 Feb 08
Posts: 249
Credit: 370,186,977
RAC: 0
Level
Asp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 23510 - Posted: 17 Feb 2012 | 22:42:10 UTC - in response to Message 23505.

Deal!

Simba123
Send message
Joined: 5 Dec 11
Posts: 147
Credit: 69,970,684
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwat
Message 23526 - Posted: 18 Feb 2012 | 23:47:07 UTC

lol! :)
don't forget to post a link up for those videos when you get them done!

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,835,229,774
RAC: 291,018
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 23686 - Posted: 28 Feb 2012 | 11:06:37 UTC - in response to Message 23507.
Last modified: 28 Feb 2012 | 11:27:08 UTC

I'm seeing 75% GPU utilization from a GTX470 @680MHz on a 2003R2 x64 i7-2600 (HT on, 2 threads free) and using SWAN_SYNC=0, so ~90% of the CPU is actual being used. Expected runtime is ~6h40min.

Saw 81% GPU use for 188-IBUCH_metTRYP-1-5-RND9663_1, when not using the CPU to crunch, but dropped back down to ~73% when crunching on 6 CPU threads. The difference is not directly down to the use of the CPU by the GPUGrid task. More likely competition for system resources. I would like to know where the bottleneck is so I can avoid CPU tasks that exasperate the issue. An 8% GPU performance loss is >> than a CPU thread, or two.
____________
FAQ's

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

Post to thread

Message boards : News : New TRYP WUs