SETI@home v8.22 Windows GPU applications support thread

Message boards : Number crunching : SETI@home v8.22 Windows GPU applications support thread
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 17 · 18 · 19 · 20

AuthorMessage
Profile Michael Christoffersen
Avatar

Send message
Joined: 12 Nov 02
Posts: 3
Credit: 1,640,144
RAC: 16
Denmark
Message 1999906 - Posted: 27 Jun 2019, 12:25:30 UTC

Call me insane (I'm ok with that!)

But... I run 8 WU's on my 7850/2gb (8 x 40min) and I run 4 WU's on my GTX 680/2gb (4 x 90min)

Surely we're not going to argue which card does the job better.

As far as I have read (like anywhere) Team green is capped at 4 WU's, and Team red is capped at 8-12 WU's.
But I can be wrong, so correct me if I am wrong :)

Anyway, have a nice day and all that :)
ID: 1999906 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3387
Credit: 177,776,486
RAC: 833,823
United States
Message 1999909 - Posted: 27 Jun 2019, 12:33:25 UTC - in response to Message 1999906.  
Last modified: 27 Jun 2019, 12:36:35 UTC

Call me insane (I'm ok with that!)

But... I run 8 WU's on my 7850/2gb (8 x 40min) and I run 4 WU's on my GTX 680/2gb (4 x 90min)

Surely we're not going to argue which card does the job better.

As far as I have read (like anywhere) Team green is capped at 4 WU's, and Team red is capped at 8-12 WU's.
But I can be wrong, so correct me if I am wrong :)

Anyway, have a nice day and all that :)


You did confirm that each increase in running gpu tasks in parallel increased your total production on the card?
Sometimes if you will notice when you go from say 1 task to 2 tasks, the amount of time each task takes to run is more than double the time it takes 1 task to run.
In that case, running tasks in parallel actually lowers your production.

Each system/card is different. Experimentation and recording results has the final say.

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1999909 · Report as offensive     Reply Quote
Profile Michael Christoffersen
Avatar

Send message
Joined: 12 Nov 02
Posts: 3
Credit: 1,640,144
RAC: 16
Denmark
Message 1999918 - Posted: 27 Jun 2019, 13:02:34 UTC - in response to Message 1999909.  

I did notice that, and I am no math gene, but if I remember correctly, 1 WU took about 20min on the 7850, and now 8 WU's takes about 40min each. I can deal with that. On my GTX 680, well... Lets just say that my 680 card isnt really up for the task(s) hehe :)
ID: 1999918 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3387
Credit: 177,776,486
RAC: 833,823
United States
Message 1999988 - Posted: 27 Jun 2019, 21:15:46 UTC - in response to Message 1999918.  

I did notice that, and I am no math gene, but if I remember correctly, 1 WU took about 20min on the 7850, and now 8 WU's takes about 40min each. I can deal with that. On my GTX 680, well... Lets just say that my 680 card isnt really up for the task(s) hehe :)


If my experience applies, it sounds like you should scale up 20 X 8 = 160 minutes if there was no loss in speed. It sounds like we are talking about 40 X 8 = 320 minute total. If I am not confused that is about a 50% reduction in processing speed per task.

(1440 / 20 ) * 1 = 72 tasks a day.
(1440 / 40 ) * 8 = 288 tasks a day.

So if the number you gave me are fairly accurate you are seeing a very large increase in production.

Congratulations!

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1999988 · Report as offensive     Reply Quote
Stephen "Heretic" Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 4574
Credit: 137,653,277
RAC: 246,725
Australia
Message 1999998 - Posted: 27 Jun 2019, 22:52:27 UTC - in response to Message 1999988.  
Last modified: 27 Jun 2019, 23:02:06 UTC

I did notice that, and I am no math gene, but if I remember correctly, 1 WU took about 20min on the 7850, and now 8 WU's takes about 40min each. I can deal with that. On my GTX 680, well... Lets just say that my 680 card isnt really up for the task(s) hehe :)

If my experience applies, it sounds like you should scale up 20 X 8 = 160 minutes if there was no loss in speed. It sounds like we are talking about 40 X 8 = 320 minute total. If I am not confused that is about a 50% reduction in processing speed per task.
(1440 / 20 ) * 1 = 72 tasks a day.
(1440 / 40 ) * 8 = 288 tasks a day.
So if the number you gave me are fairly accurate you are seeing a very large increase in production.
Congratulations!
Tom


. . Hi Tom,

. . If one at a time takes 20 mins that is 20 mins/task, if 8 at once takes 40 mins then that is 40/8 =5 mins/task, that is a 75% reduction in processing time or a 400% increase in speed (20 mins = 4 tasks instead of 1). That seems an unlikely improvement unless it was very badly configured when running 1 at a time. If running onesies under SoG the GPU tends to run at over 90% efficiency but to achieve that level of improvement it would have had to be less than 25% efficiency.

. . Considering he is running Cuda42 on the Nvidia card I am guessing it is also very badly configured and would achieve much better results if he changed over to SoG and ran onesies. It would probably just about halve the run times per task on that rig.

Stephen

? ?
ID: 1999998 · Report as offensive     Reply Quote
Stephen "Heretic" Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 4574
Credit: 137,653,277
RAC: 246,725
Australia
Message 2000002 - Posted: 27 Jun 2019, 23:07:47 UTC - in response to Message 1999906.  

Call me insane (I'm ok with that!)

But... I run 8 WU's on my 7850/2gb (8 x 40min) and I run 4 WU's on my GTX 680/2gb (4 x 90min)

Surely we're not going to argue which card does the job better.

As far as I have read (like anywhere) Team green is capped at 4 WU's, and Team red is capped at 8-12 WU's.
But I can be wrong, so correct me if I am wrong :)


Anyway, have a nice day and all that :)


. . When testing I have run 8 at a time on GTX970s with 4GB without trouble. So the cap is only a matter of how far you want to push it and how much you are willing to reduce the operating efficiency. Generally it is better to run SoG and just do the one or maybe 2 at a time.

Stephen

:(
ID: 2000002 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5161
Credit: 411,875,319
RAC: 1,954,423
United States
Message 2000012 - Posted: 28 Jun 2019, 0:11:26 UTC - in response to Message 2000002.  

Generally it is better to run SoG and just do the one or maybe 2 at a time.

Stephen

:(


No, that is not true. Sorry Stephen....

The rule is, how much Memory does the GPU have and how much of it is available for use in OpenCl applications.

The 970s only has 3.5GB. Yes we all know it says 4 on the box but it's really only 3.5

Nvidia cards are limited to 27% of available RAM for use in OpenCl. Intel and ATI cards are up around 50-67% of the RAM of the card.

So when setting the commandlines for SoG, you have to figure in how much RAM per work and then figure out how many parallel work units you can run.

With 980ti you can run 3 instances and with the new 2080Ti you can run4 instances.

Since I've never had an ATI card, I can't tell you how many you can do but it's fairly obviously that it will be more than what can run on Nvidia.

Good luck.

Z
ID: 2000012 · Report as offensive     Reply Quote
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 11484
Credit: 167,798,825
RAC: 106,915
Australia
Message 2000044 - Posted: 28 Jun 2019, 4:56:25 UTC - in response to Message 2000012.  
Last modified: 28 Jun 2019, 4:57:44 UTC

So when setting the commandlines for SoG, you have to figure in how much RAM per work and then figure out how many parallel work units you can run.

All I care about it what produces the most work per hour. For CUDA50, depending on your hardware, 2, 3 or even 4 WUs at a time would produce the most work per hour.
For SoG, no matter what settings I have used, running 1 WU at a time has always produced the most work per hour.
Just because a card can run more than 1 WU at a time doesn't mean it's a good idea.
Grant
Darwin NT
ID: 2000044 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5161
Credit: 411,875,319
RAC: 1,954,423
United States
Message 2000095 - Posted: 28 Jun 2019, 13:38:05 UTC - in response to Message 2000044.  

For me running 3 at a time under SoG was much faster than 1.

The person I know currently running 2080Ti, it's 4 SoG at a time.

It's the GPU that makes the difference.
ID: 2000095 · Report as offensive     Reply Quote
Previous · 1 . . . 17 · 18 · 19 · 20

Message boards : Number crunching : SETI@home v8.22 Windows GPU applications support thread


 
©2019 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.