Typical time for a GTX 980 to crunch a work unit?

Message boards : Number crunching : Typical time for a GTX 980 to crunch a work unit?
Message board moderation

To post messages, you must log in.

AuthorMessage
krem1234

Send message
Joined: 1 Mar 03
Posts: 19
Credit: 3,087,359
RAC: 0
Italy
Message 1711058 - Posted: 10 Aug 2015, 18:23:01 UTC

Thinking about starting off with one and was just curious how long it generally takes for it to process a work unit. I calculated a figure based on the card's GFLOPS compared to my CPU and how long the WUs take on my laptop but it seems faster than actual comparing that to the top hosts board. Thanks
ID: 1711058 · 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 1711062 - Posted: 10 Aug 2015, 18:31:49 UTC - in response to Message 1711058.  
Last modified: 10 Aug 2015, 18:32:46 UTC

It depends on how many instances of work you do per card. Most users tend to do more than 1 work unit per card so the times to complete may look longer with those users. But when you calculate how long it takes to say 3 work units at a time and divide the time by 3, it's actually faster than doing 1 work unit at a time.

Without getting into details, when you are only doing 1 work unit at a time with higher end, you aren't using all of the resources of that card. When you run more than 1, you make better utilization of the card and the program that is analyzing the work. Jason can go into details if you want.
ID: 1711062 · Report as offensive
krem1234

Send message
Joined: 1 Mar 03
Posts: 19
Credit: 3,087,359
RAC: 0
Italy
Message 1711068 - Posted: 10 Aug 2015, 18:38:55 UTC - in response to Message 1711062.  
Last modified: 10 Aug 2015, 19:04:27 UTC

Thanks, and I'd be interested in any additional details Jason may have.

Do you know if there's a general optimal # of WUs to be processing simultaneously on this card? Or is that more system-specific?
ID: 1711068 · 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 1711091 - Posted: 10 Aug 2015, 19:01:22 UTC - in response to Message 1711068.  
Last modified: 10 Aug 2015, 19:27:41 UTC

Loaded question there.

It's going to be based on what your system can handle. Usually we tell people to start with 1 work unit at time and run it for some time. (couple of hours) figure out the average time.

Then go to 2 at a time and repeat for couple more hours. divide by 2 to get the time.

Then 3 and repeat.

What you are looking for is a point where the average time after you divide is going up rather than down, then you know you went too far. Like a inverted bell curve. You want to be on the downslope.

Somewhere along that curve is where the best performance is. But your system will also play a big part. It may turn out that your system may lock up or freeze before you find that ultimate point. In which case, you go with lesser amounts per card to keep the system stable.

There tends to be a lot of fine tuning that occurs. Modification of the mbcuda.cfg or the commandline for the astropulses.

Sometimes the applications seize control of the computer and it becomes unresponsive ( i use process lasso to keep the programs in check and not running amuck)

As a generalization, you can do at least 3 work units at a time without too much lagging on your system. Again, depending on what your Motherboard, CPU chip, etc, your mileage may vary.
ID: 1711091 · 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 1711094 - Posted: 10 Aug 2015, 19:03:50 UTC - in response to Message 1711068.  

I always use eFMer's Seti Performance tool whenever I change to a new card type to determine the most efficient use and highest output per day.
Seti Performance

I do three tasks per card with my 970's, likely the 980 will do the same. Run the test and see where it falls.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1711094 · Report as offensive
Profile Todderbert
Avatar

Send message
Joined: 17 Jun 99
Posts: 221
Credit: 53,153,779
RAC: 0
United States
Message 1711316 - Posted: 11 Aug 2015, 4:57:45 UTC

I run three at a time on my 970s/980s, but I have it setup to run no more then 2APs at a time, while a MB will always be three per card. I'm sure the GPUs will run three APs at a time but I'm going for stability rather then speed. So I use .34 for the APs and .32 for the MBs, works smoothly for me.
ID: 1711316 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1711324 - Posted: 11 Aug 2015, 5:26:09 UTC

I've been running a pair of gtx980 for a time, and have spent some time getting the optimal loading right I've found 2AP to be the realistic limit as compute times increase quite dramatically when going to three.

I keep my app_config.xml figures simple - the version below allows 2 APs or 3 MBs:

<app_config>
<app>
<name>astropulse_v7</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.7</cpu_usage>
</gpu_versions>
</app>
<app>
<name>setiathome_v7</name>
<gpu_versions>
<gpu_usage>.33</gpu_usage>
<cpu_usage>.1</cpu_usage>
</gpu_versions>
</app>
</app_config>


Key thing when tuning - take TIME, unless a configuration is obviously defective, let it run for days or even weeks before making the next change. This way you will hopefully eliminate the vagaries of the SETI data set.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1711324 · Report as offensive
Darth Beaver Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 20 Aug 99
Posts: 6728
Credit: 21,443,075
RAC: 3
Australia
Message 1711325 - Posted: 11 Aug 2015, 5:28:55 UTC

My 680 will do 2 MB's in aprox 14 mins per units and 2 AP's in aprox 24 mins

It can do 4 units at a time but is to slow to be worth doing that many at a time
ID: 1711325 · Report as offensive
Andrew Scharbarth
Volunteer tester

Send message
Joined: 29 May 07
Posts: 40
Credit: 5,984,436
RAC: 0
United States
Message 1711378 - Posted: 11 Aug 2015, 9:37:57 UTC
Last modified: 11 Aug 2015, 9:40:42 UTC

I've got a 970 that averages about 6 minutes on work units and about 20 minutes on ap's.
ID: 1711378 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1712561 - Posted: 13 Aug 2015, 22:17:33 UTC
Last modified: 13 Aug 2015, 22:21:06 UTC

To get an idea of run times take a look at my results. I run one at a time with all my GPUs (2 x GTX780 + 2 x GTX980).

Linux1 with 2 x GTX 780 and 2 x GTX980
To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones
ID: 1712561 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1713036 - Posted: 14 Aug 2015, 21:05:09 UTC - in response to Message 1712561.  

To get an idea of run times take a look at my results. I run one at a time with all my GPUs (2 x GTX780 + 2 x GTX980).

Linux1 with 2 x GTX 780 and 2 x GTX980


A correction. AP's run two at a time.
To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones
ID: 1713036 · Report as offensive

Message boards : Number crunching : Typical time for a GTX 980 to crunch a work unit?


 
©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.