Are some gpu tasks longer now?

Message boards : Number crunching : Are some gpu tasks longer now?
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · Next

AuthorMessage
Miklos M.

Send message
Joined: 5 May 99
Posts: 955
Credit: 136,115,648
RAC: 73
Hungary
Message 1794668 - Posted: 9 Jun 2016, 9:26:24 UTC

I have fairly fast gpu cards, but the running time used to be 7-9 minutes, some now take over 1/2 hour. Just curious.
ID: 1794668 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34249
Credit: 79,922,639
RAC: 80
Germany
Message 1794672 - Posted: 9 Jun 2016, 9:40:21 UTC
Last modified: 9 Jun 2016, 9:40:45 UTC

Not longer but different type (Guppi) and angle range (VLAR).
Plenty of threads about it here.


With each crime and every kindness we birth our future.
ID: 1794672 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1794689 - Posted: 9 Jun 2016, 12:16:56 UTC

Yes, the vlars that are now running on your card will run about 4x slower than non-vlar MBs, as you have noticed. I have seen the same on my GTX 980s. That's why vlars were excluded from NVIDIA graphics cards in the past - the app for computing them slows way down on them. However, with the new Green Bank data, turns out most of the MBs from them ARE vlars, so it was decided to put vlars on NVIDIA cards anyway.

Kinda makes you wonder why you laid out the $$$ for really good graphics cards, don't it?
ID: 1794689 · Report as offensive
Miklos M.

Send message
Joined: 5 May 99
Posts: 955
Credit: 136,115,648
RAC: 73
Hungary
Message 1794706 - Posted: 9 Jun 2016, 13:42:42 UTC - in response to Message 1794689.  

You are right, it makes little sense to me. Perhaps the powers to be will change the programs and not schedule vlars in the near future. Is there a way to tell which are vlars?
ID: 1794706 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14644
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1794708 - Posted: 9 Jun 2016, 13:45:04 UTC - in response to Message 1794689.  

Yes, the vlars that are now running on your card will run about 4x slower than non-vlar MBs, as you have noticed. I have seen the same on my GTX 980s. That's why vlars were excluded from NVIDIA graphics cards in the past - the app for computing them slows way down on them. However, with the new Green Bank data, turns out most of the MBs from them ARE vlars, so it was decided to put vlars on NVIDIA cards anyway.

Kinda makes you wonder why you laid out the $$$ for really good graphics cards, don't it?

I'm now finding that the newer blc6 and blc7 VLARs now run about 2x slower on my GTX 750Ti with cuda50 - it was closer to your 4x at the start of the run.

But I have a suspicion that the newer and larger the GPU, the greater the slowdown. I'll try and test that next time I have a gap between GPUGrid tasks on my GTX 970.

If true, it does suggest that possibly the really big cards are overkill for the tasks we run here, which reinforces your question.
ID: 1794708 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14644
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1794710 - Posted: 9 Jun 2016, 13:46:40 UTC - in response to Message 1794706.  

Is there a way to tell which are vlars?

The last four letters in the task name. You may need to scroll right or click 'properties' for see them.
ID: 1794710 · Report as offensive
Profile Mr. Kevvy Crowdfunding Project Donor*Special Project $250 donor
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 15 May 99
Posts: 3776
Credit: 1,114,826,392
RAC: 3,319
Canada
Message 1794711 - Posted: 9 Jun 2016, 13:52:45 UTC - in response to Message 1794706.  
Last modified: 9 Jun 2016, 14:09:09 UTC

Perhaps the powers to be will change the programs and not schedule vlars in the near future.


Dr. Korpela indicated that it was tried on Beta to add a setting to the Project Preferences to allow/disallow GPU VLAR, but that this resulted in no GPU work at all. I'm puzzled by this, as the code with no option to disallow them was in placed and worked just fine for a while; something must have been done incorrectly so I'm still hopeful that they are working on it and a solution is forthcoming. I keep seeing that it's a non-issue because non-VLAR Arecibo work is going away, but I don't see any evidence of this (edit: Richard H. posted some above... damn) and the team has indicated otherwise (about equal amounts of work will be distributed from each source.)

I'm now finding that the newer blc6 and blc7 VLARs now run about 2x slower on my GTX 750Ti with cuda50 - it was closer to your 4x at the start of the run.


I noticed today that my RAC started rising after over a month of fall. It should have bottomed out around 260K. So it does appear that some (welcome) change was made. (Edit: maybe it was just the recent AP run.)
ID: 1794711 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34249
Credit: 79,922,639
RAC: 80
Germany
Message 1794715 - Posted: 9 Jun 2016, 13:58:40 UTC - in response to Message 1794708.  
Last modified: 9 Jun 2016, 13:59:17 UTC

Yes, the vlars that are now running on your card will run about 4x slower than non-vlar MBs, as you have noticed. I have seen the same on my GTX 980s. That's why vlars were excluded from NVIDIA graphics cards in the past - the app for computing them slows way down on them. However, with the new Green Bank data, turns out most of the MBs from them ARE vlars, so it was decided to put vlars on NVIDIA cards anyway.

Kinda makes you wonder why you laid out the $$$ for really good graphics cards, don't it?

I'm now finding that the newer blc6 and blc7 VLARs now run about 2x slower on my GTX 750Ti with cuda50 - it was closer to your 4x at the start of the run.

But I have a suspicion that the newer and larger the GPU, the greater the slowdown. I'll try and test that next time I have a gap between GPUGrid tasks on my GTX 970.

If true, it does suggest that possibly the really big cards are overkill for the tasks we run here, which reinforces your question.


Thats definitely not the case.


With each crime and every kindness we birth our future.
ID: 1794715 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14644
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1794717 - Posted: 9 Jun 2016, 14:00:50 UTC - in response to Message 1794711.  

Eric told us last night (private email to developer team) that "ALFA observations have been occurring at a greatly reduced rate." - SETI's Arecibo tasks are recorded using the ALFA receiver.

Also, see the News at Beta.
ID: 1794717 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14644
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1794719 - Posted: 9 Jun 2016, 14:06:31 UTC - in response to Message 1794715.  

Thats definitely not the case.

Speaking from my own experience with NVidia cards and CUDA applications. GBT 'guppi' VLARs are settling around 53 minutes (two at a time) on GTX 750Ti. Will start GTX 970 test in about two hours' time.

Before anyone asks, I'm not interested in running applications which require 100% CPU support (except as tests) - that would prevent me contributing to CPU-only projects.
ID: 1794719 · Report as offensive
Profile Mr. Kevvy Crowdfunding Project Donor*Special Project $250 donor
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 15 May 99
Posts: 3776
Credit: 1,114,826,392
RAC: 3,319
Canada
Message 1794720 - Posted: 9 Jun 2016, 14:11:32 UTC - in response to Message 1794719.  
Last modified: 9 Jun 2016, 14:12:24 UTC

Thats definitely not the case.

Speaking from my own experience with NVidia cards and CUDA applications. GBT 'guppi' VLARs are settling around 53 minutes (two at a time) on GTX 750Ti. Will start GTX 970 test in about two hours' time.


980 11-13 mins., 970 13-16 mins. with 0.2 core one at a time. Does about the same regardless of the CPU or PCIe revision. Non-VLArs are/were 3-4 mins.
ID: 1794720 · Report as offensive
Miklos M.

Send message
Joined: 5 May 99
Posts: 955
Credit: 136,115,648
RAC: 73
Hungary
Message 1794721 - Posted: 9 Jun 2016, 14:14:46 UTC - in response to Message 1794710.  

Richard, what am I looking for in the "last four letters?"
ID: 1794721 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14644
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1794722 - Posted: 9 Jun 2016, 14:16:34 UTC - in response to Message 1794721.  

Richard, what am I looking for in the "last four letters?"

V - L - A - R

As in blc6_2bit_guppi_57403_HIP11048_0006.15090.831.22.45.11.vlar_0
ID: 1794722 · Report as offensive
Miklos M.

Send message
Joined: 5 May 99
Posts: 955
Credit: 136,115,648
RAC: 73
Hungary
Message 1794768 - Posted: 9 Jun 2016, 17:41:08 UTC - in response to Message 1794722.  

I do not have any vlar wu', yet many are taking the aforementioned 1/2-3/4 hour each. It does not matter whether I turn off any other apps.
ID: 1794768 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14644
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1794802 - Posted: 9 Jun 2016, 20:56:02 UTC - in response to Message 1794720.  

Speaking from my own experience with NVidia cards and CUDA applications. GBT 'guppi' VLARs are settling around 53 minutes (two at a time) on GTX 750Ti. Will start GTX 970 test in about two hours' time.

980 11-13 mins., 970 13-16 mins. with 0.2 core one at a time. Does about the same regardless of the CPU or PCIe revision. Non-VLArs are/were 3-4 mins.

OK, some timings from host 4292666. That link drops you in at offset 40, which will probably still be right for the next 12 hours or so. All tasks referenced are on that page or the next 20.

Guppi VLAR tasks, runtime in minutes

	GTX 970		GTX 750Ti
			
	51:45		58:41
	51:50		58:27
	50:54		58:02
	52:18		58:51
	49:20		58:44
	52:10		58:55
	52:36		58:13
	50:53		59:09
	51:50		58:42
			58:16
			
Average	51:31		58:36

Arecibo mid-AR tasks

	13:50		24:22
	13:52		23:49
	12:40		24:14
	13:51		23:56
			
Average	13:33		24:05

All tasks run between ~15:30 and ~20:30 UTC today, with the same cuda50 application. Timings were taken with each card running two instances of the stated task type. The guppies were a mixture of blc6 and blc7, and (going by deadline only) the Arecibo were all pretty close to AR=0.428594

The only conclusion I'm drawing from that is that the bigger card (GTX 970) was slowed down more dramatically by the 'double guppi' loading.
ID: 1794802 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1794808 - Posted: 9 Jun 2016, 21:21:42 UTC - in response to Message 1794719.  

Before anyone asks, I'm not interested in running applications which require 100% CPU support (except as tests) - that would prevent me contributing to CPU-only projects.

Ignorance dominates the ability of adaptation?
-use_sleep would be easiest change to be happy with CPU-only projects.
ID: 1794808 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34249
Credit: 79,922,639
RAC: 80
Germany
Message 1794818 - Posted: 9 Jun 2016, 21:47:05 UTC - in response to Message 1794808.  

Before anyone asks, I'm not interested in running applications which require 100% CPU support (except as tests) - that would prevent me contributing to CPU-only projects.

Ignorance dominates the ability of adaptation?
-use_sleep would be easiest change to be happy with CPU-only projects.


I was going to say the same but i`m tired of it :(


With each crime and every kindness we birth our future.
ID: 1794818 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14644
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1794821 - Posted: 9 Jun 2016, 21:55:42 UTC

Just tell us when it's finished, please, and works out of the box.
ID: 1794821 · Report as offensive
bluestar

Send message
Joined: 5 Sep 12
Posts: 6979
Credit: 2,084,789
RAC: 3
Message 1794823 - Posted: 9 Jun 2016, 21:58:38 UTC - in response to Message 1794668.  
Last modified: 9 Jun 2016, 22:01:19 UTC

Anyway, Miklos M.

I happen to know that you are closing in on the double red somewhere else.

Guess some people are missing or not noticing that fact.
ID: 1794823 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1794824 - Posted: 9 Jun 2016, 21:59:09 UTC - in response to Message 1794821.  

Just tell us when it's finished, please, and works out of the box.


Amen!
ID: 1794824 · Report as offensive
1 · 2 · 3 · Next

Message boards : Number crunching : Are some gpu tasks longer now?


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.