Nvidia Volta - Titan V thread

Message boards : Number crunching : Nvidia Volta - Titan V thread
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · Next

AuthorMessage
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1907408 - Posted: 16 Dec 2017, 4:47:57 UTC - in response to Message 1907403.  

See the card in the Host's Detail, but no tasks yet on it. See you are running stock too which means BOINC it going to send you every flavor of gpu app to test. It would go faster if you put the system on the Anonymous platform using the latest Lunatics Installer available at Crunchers Anonymous or Mike's World and select the SoG app for the gpu.

Lunatics Installer v0.45b6 (64-bit)
Mike's World
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1907408 · Report as offensive
hasherati

Send message
Joined: 9 Oct 17
Posts: 11
Credit: 13,749,520
RAC: 25
United States
Message 1907411 - Posted: 16 Dec 2017, 5:02:23 UTC - in response to Message 1907408.  

[quote]See the card in the Host's Detail, but no tasks yet on it. See you are running stock too which means BOINC it going to send you every flavor of gpu app to test. It would go faster if you put the system on the Anonymous platform using the latest Lunatics Installer available at Crunchers Anonymous or Mike's World and select the SoG app for the gpu.

Thanks for the tip. Installed and processing, standby.
ID: 1907411 · Report as offensive
hasherati

Send message
Joined: 9 Oct 17
Posts: 11
Credit: 13,749,520
RAC: 25
United States
Message 1907415 - Posted: 16 Dec 2017, 5:11:46 UTC - in response to Message 1907303.  
Last modified: 16 Dec 2017, 5:12:15 UTC

Ok, boys, here's a work unit on the Titan V:
https://setiathome.berkeley.edu/result.php?resultid=6236392827
ID: 1907415 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1907417 - Posted: 16 Dec 2017, 5:14:24 UTC - in response to Message 1907411.  
Last modified: 16 Dec 2017, 5:16:46 UTC

[quote]See the card in the Host's Detail, but no tasks yet on it. See you are running stock too which means BOINC it going to send you every flavor of gpu app to test. It would go faster if you put the system on the Anonymous platform using the latest Lunatics Installer available at Crunchers Anonymous or Mike's World and select the SoG app for the gpu.

Thanks for the tip. Installed and processing, standby.

And to really make it work, i'd suggest trying these command line values in the
mb_cmdline_win_x86_SSE3_OpenCL_NV_SoG.txt
file in the project data directory.

-hp -period_iterations_num 1 -high_perf -high_prec_timer -sbs 2048 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64

No need to restart or anything like that.
With the values in that file, they will be picked up when each WU starts.

I'd let it run for a couple of hours- see what processing times you get for Arecibo & GBT work. With that number of Compute Units available (80! 80 $%$#%# Compute Units!), 2 or even 3 (or maybe, maybe even 4) WUs at a time will probably give you the most work per hour. But let it do 1 at a time for a while just to see what the base line is.


EDIT- and did you pick AVX for the CPU when running the installer? That will give the best output from it as well.
Grant
Darwin NT
ID: 1907417 · Report as offensive
hasherati

Send message
Joined: 9 Oct 17
Posts: 11
Credit: 13,749,520
RAC: 25
United States
Message 1907420 - Posted: 16 Dec 2017, 5:26:10 UTC - in response to Message 1907417.  
Last modified: 16 Dec 2017, 5:27:11 UTC


And to really make it work, i'd suggest trying these command line values in the
mb_cmdline_win_x86_SSE3_OpenCL_NV_SoG.txt
file in the project data directory.

-hp -period_iterations_num 1 -high_perf -high_prec_timer -sbs 2048 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64

No need to restart or anything like that.
With the values in that file, they will be picked up when each WU starts.

I'd let it run for a couple of hours- see what processing times you get for Arecibo & GBT work. With that number of Compute Units available (80! 80 $%$#%# Compute Units!), 2 or even 3 (or maybe, maybe even 4) WUs at a time will probably give you the most work per hour. But let it do 1 at a time for a while just to see what the base line is.


EDIT- and did you pick AVX for the CPU when running the installer? That will give the best output from it as well.


Done. Standby for more work units to pour in. This thing is flying through units like no ones business.

EDIT- Can't recall if I selected AVX for CPU when running the installer.
ID: 1907420 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1907422 - Posted: 16 Dec 2017, 5:31:00 UTC - in response to Message 1907420.  

EDIT- Can't recall if I selected AVX for CPU when running the installer
Just look at your task manager to see if the AVX app is running.
ID: 1907422 · Report as offensive
hasherati

Send message
Joined: 9 Oct 17
Posts: 11
Credit: 13,749,520
RAC: 25
United States
Message 1907424 - Posted: 16 Dec 2017, 5:37:17 UTC - in response to Message 1907422.  

Just look at your task manager to see if the AVX app is running.


No AVX is not running, but those setting you gave me cut the time in half for a unit:
https://setiathome.berkeley.edu/result.php?resultid=6236430352
ID: 1907424 · Report as offensive
hasherati

Send message
Joined: 9 Oct 17
Posts: 11
Credit: 13,749,520
RAC: 25
United States
Message 1907429 - Posted: 16 Dec 2017, 5:40:20 UTC - in response to Message 1907424.  

Card just tanked. May need to reboot.
ID: 1907429 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1907451 - Posted: 16 Dec 2017, 7:59:38 UTC

Haven't seen any BLC tasks yet, but the times with the suggested tuning parameters for the r3557 SoG app are equalling the compute time for Arecibo shorties that the special app gets in Linux. Damn impressive.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1907451 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1907455 - Posted: 16 Dec 2017, 8:26:55 UTC - in response to Message 1907429.  

Card just tanked. May need to reboot.

Watch your temps. Nvidia always is way too conservative with fan profiles to accommodate low noise profiles and sacrifice card performance. The card could have crashed because of high temps or more likely immature drivers doing real compute.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1907455 · Report as offensive
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 1907498 - Posted: 16 Dec 2017, 14:25:53 UTC
Last modified: 16 Dec 2017, 14:28:40 UTC

Impressive times & -unroll 80 ????? Does the software even work with that numbers?
And that is under Windows SoG, now imagine with Linux & CUDA90. Ready for Hypersonic speeds.

You give me another reason to win the lottery. Imagine what a 4x Titan V Linux host could do?

Soon, very soon they will need to make changes in the way the work is distributed to feed this new babies.
With this times a single Titan V could easelly crunch the entire 100 WU cache 1 - 1 1/2 hrs.
ID: 1907498 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1907503 - Posted: 16 Dec 2017, 14:40:25 UTC - in response to Message 1907417.  
Last modified: 16 Dec 2017, 14:41:14 UTC

Nevermind, lol
ID: 1907503 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1907535 - Posted: 16 Dec 2017, 17:16:16 UTC

lol Naa, that's what they have warranties for!

ID: 1907535 · Report as offensive
hasherati

Send message
Joined: 9 Oct 17
Posts: 11
Credit: 13,749,520
RAC: 25
United States
Message 1907568 - Posted: 16 Dec 2017, 21:08:33 UTC - in response to Message 1907505.  

I guess $3000 went up in smoke, after just a couple of SoG tasks.
Ah well, I guess he could afford it.


No, she lives. I couldn't resist it and opened up the other box and just installed the twin brother.
I'm going to burn it in with some Ethereum mining and might not have time to run some SETI tests until later tonight PST time.

https://drive.google.com/file/d/1Fmo6jLkq88pvMiq29-q_DySbA5UoYPIh/view?usp=sharing
ID: 1907568 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1907574 - Posted: 16 Dec 2017, 21:40:24 UTC - in response to Message 1907568.  

I that a 760W PSU I see? Might be a tad small ...
ID: 1907574 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1907575 - Posted: 16 Dec 2017, 21:57:32 UTC - in response to Message 1907574.  

what's you thinking 850? I'd probably go with 1000, overkill I know but....
ID: 1907575 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1907576 - Posted: 16 Dec 2017, 22:00:18 UTC

Maybe Raistmer will notice the thread and chime in. My question, AFAIK SBS=2048 is the largest value you can pass in parameter tuning and the stderr.txt shows that. However the stderr.txt also shows 3072 MB actually allocated for the task. Is this a case where the parameters are just ignored and the OpenCL part of the driver simply sets the memory used to be the 25% of the cards memory as has been established in other threads? So 12GB/4=3072 MB.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1907576 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1907578 - Posted: 16 Dec 2017, 22:04:21 UTC - in response to Message 1907575.  

The Titan V has a TDP of 250W and the user probably is just letting it run stock with no overclock so I would expect the card to hew to its TDP. So ~500 watts total for both cards plus whatever the Xeon server needs.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1907578 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1907580 - Posted: 16 Dec 2017, 22:12:49 UTC - in response to Message 1907578.  

The Titan V has a TDP of 250W and the user probably is just letting it run stock with no overclock so I would expect the card to hew to its TDP. So ~500 watts total for both cards plus whatever the Xeon server needs.

So 850W bare minimum to allow for full loading of the GPUs & CPU.
Grant
Darwin NT
ID: 1907580 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1907581 - Posted: 16 Dec 2017, 22:16:36 UTC - in response to Message 1907575.  

what's you thinking 850? I'd probably go with 1000, overkill I know but....

Given a switch mode supply is most efficient around 50% of it's rated load, and up to 75% is still pretty good, a 1kW PSU would be just about right. 25% headroom is nice, 15% isn't enough IMHO.
Grant
Darwin NT
ID: 1907581 · Report as offensive
Previous · 1 · 2 · 3 · Next

Message boards : Number crunching : Nvidia Volta - Titan V thread


 
©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.