Advanced search

Message boards : Number crunching : New WU's

Author Message
flashawk
Send message
Joined: 18 Jun 12
Posts: 297
Credit: 3,327,731,397
RAC: 3,051,704
Level
Arg
Scientific publications
watwatwatwatwatwatwatwat
Message 51529 - Posted: 20 Feb 2019 | 3:42:54 UTC
Last modified: 20 Feb 2019 | 3:45:01 UTC

These guys are big, 9 hours 421,000 points, 137MB. It's going to be tough on the guys with slower cards.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1104
Credit: 6,101,732,079
RAC: 1,085
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51542 - Posted: 21 Feb 2019 | 19:03:53 UTC - in response to Message 51529.

These guys are big, 9 hours 421,000 points, 137MB. It's going to be tough on the guys with slower cards.

Yep, they're big-uns. They're running between 14:50 and 15:50 on my various 1060 GPUs. Had to lower my BOINC cache size a bit to make sure they completed in under 24 hours. The GTX 1060s on the Ryzen 7 machines are faster than the other boxes even though they're running 3x 1060 cards and 13x Universe CPU WUs. BTW, on the Ryzen X370/X470 MBs the 1060 GPUs run just as fast in the 3rd PCIe 2.0 x4 slot as in the two PCIe 3.0 x8 slots. This might not be true though with faster GPUs...

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1104
Credit: 6,101,732,079
RAC: 1,085
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51548 - Posted: 22 Feb 2019 | 15:40:52 UTC

Yikes, these had been kind of rare, but just got 4 of them in the last few hours.

Matt Kowal
Avatar
Send message
Joined: 27 May 14
Posts: 9
Credit: 89,194,351
RAC: 216,512
Level
Thr
Scientific publications
wat
Message 51551 - Posted: 22 Feb 2019 | 21:17:44 UTC - in response to Message 51542.

BTW, on the Ryzen X370/X470 MBs the 1060 GPUs run just as fast in the 3rd PCIe 2.0 x4 slot as in the two PCIe 3.0 x8 slots. This might not be true though with faster GPUs...


Good information. Thank you for sharing.

flashawk
Send message
Joined: 18 Jun 12
Posts: 297
Credit: 3,327,731,397
RAC: 3,051,704
Level
Arg
Scientific publications
watwatwatwatwatwatwatwat
Message 51554 - Posted: 23 Feb 2019 | 14:37:55 UTC

Ya, I've completed 7 and have 4 more being crunched right now, I don't mind doing them at all. It's nice getting almost half a million points for 9 hours of work, I've noticed that Pablo gets a very high utilization of the shader cores, above 90 %. His always run a lot warmer than Adria's work units.

Zalster
Avatar
Send message
Joined: 26 Feb 14
Posts: 175
Credit: 4,013,368,076
RAC: 10,143
Level
Arg
Scientific publications
watwatwat
Message 51555 - Posted: 23 Feb 2019 | 16:07:12 UTC - in response to Message 51554.

e1s100_I450-PABLO_ADAPTIVE_goal_KIX_CMYB-1-4-RND1090_0 Are these those guys?

5.5 hrs on 1080ti

biodoc
Send message
Joined: 26 Aug 08
Posts: 158
Credit: 1,405,900,597
RAC: 137,721
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51558 - Posted: 23 Feb 2019 | 18:42:06 UTC - in response to Message 51555.

e1s100_I450-PABLO_ADAPTIVE_goal_KIX_CMYB-1-4-RND1090_0 Are these those guys?

5.5 hrs on 1080ti


Yep and also the PABLO_UCB_goal tasks.

Your 1080 Ti is a lot faster than mine (6.72 hours).

Zalster
Avatar
Send message
Joined: 26 Feb 14
Posts: 175
Credit: 4,013,368,076
RAC: 10,143
Level
Arg
Scientific publications
watwatwat
Message 51559 - Posted: 23 Feb 2019 | 19:36:12 UTC - in response to Message 51558.

e1s100_I450-PABLO_ADAPTIVE_goal_KIX_CMYB-1-4-RND1090_0 Are these those guys?

5.5 hrs on 1080ti


Yep and also the PABLO_UCB_goal tasks.

Your 1080 Ti is a lot faster than mine (6.72 hours).


I think it's more to do with the MoBo. My 1080Tis are hybrids but the MoBo is ASUS X99 with PCIe x16 on the 4 main PCIe.
____________

Profile ServicEnginIC
Avatar
Send message
Joined: 24 Sep 10
Posts: 21
Credit: 1,154,625,676
RAC: 659,856
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51562 - Posted: 23 Feb 2019 | 20:18:51 UTC
Last modified: 23 Feb 2019 | 20:21:07 UTC

e1s100_I450-PABLO_ADAPTIVE_goal_KIX_CMYB-1-4-RND1090_0 Are these those guys?

5.5 hrs on 1080ti




Yep and also the PABLO_UCB_goal tasks.

Your 1080 Ti is a lot faster than mine (6.72 hours).


A heavy reason for this is due to biodoc and Zalster WU's are running in different modes.
biodoc: # CUDA Synchronisation mode: BLOCKING
Zalster: # CUDA Synchronisation mode: SPIN
SPIN mode is clearly faster than BLOCKING one.
This is profusely discussed in Retvari Zoltan thread http://www.gpugrid.net/forum_thread.php?id=4813#50180

biodoc
Send message
Joined: 26 Aug 08
Posts: 158
Credit: 1,405,900,597
RAC: 137,721
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51567 - Posted: 24 Feb 2019 | 20:03:13 UTC

@ServicEnginIC, Thanks for the link to the swan-sync/CUDA Synchronisation mode thread. I switched my computers to "spin" mode and now with an n=1, I completed a PABLO_ADAPTIVE task in 5.7 hours on my 1080 Ti, In "block" with an n=1, a PABLO_ADAPTIVE task completed in 6.72 hours. Nice!

@Zalster, my 2 1080 Ti's are running on an Asus x79 board that has 2 PCIE 3.0 x16 slots so I think I'm ok on PCIE bandwidth.

Profile ServicEnginIC
Avatar
Send message
Joined: 24 Sep 10
Posts: 21
Credit: 1,154,625,676
RAC: 659,856
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51568 - Posted: 24 Feb 2019 | 20:25:27 UTC
Last modified: 24 Feb 2019 | 20:28:01 UTC

Congratulations, @biodoc.
This is specially notorious in this kind of long WU's. About one hour saved in execution time is very nice!
I'm very curious about some change you may have noted in GPU's temperatures... (?)

biodoc
Send message
Joined: 26 Aug 08
Posts: 158
Credit: 1,405,900,597
RAC: 137,721
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51569 - Posted: 24 Feb 2019 | 22:07:19 UTC - in response to Message 51568.

Congratulations, @biodoc.
This is specially notorious in this kind of long WU's. About one hour saved in execution time is very nice!
I'm very curious about some change you may have noted in GPU's temperatures... (?)


According to nvidia-smi output, temps, power consumption and GPU utilization went up. Unfortunately, I didn't save the numbers before I made the change.

My 1080 Ti PNY card (single blower founders-like cooling) is running very hot.

+-----------------------------------------------------------------------------+ | NVIDIA-SMI 390.77 Driver Version: 390.77 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce GTX 108... Off | 00000000:01:00.0 Off | N/A | | 66% 75C P2 206W / 250W | 706MiB / 11177MiB | 96% Default | +-------------------------------+----------------------+----------------------+ | 1 GeForce GTX 108... Off | 00000000:02:00.0 Off | N/A | | 70% 82C P2 212W / 250W | 694MiB / 11178MiB | 98% Default | +-------------------------------+----------------------+----------------------+

Zalster
Avatar
Send message
Joined: 26 Feb 14
Posts: 175
Credit: 4,013,368,076
RAC: 10,143
Level
Arg
Scientific publications
watwatwat
Message 51570 - Posted: 25 Feb 2019 | 1:51:17 UTC - in response to Message 51569.

My hybrids are running 52C
____________

Post to thread

Message boards : Number crunching : New WU's