The GTX750(Ti) Thread

Message boards : Number crunching : The GTX750(Ti) Thread
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · Next

AuthorMessage
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34258
Credit: 79,922,639
RAC: 80
Germany
Message 1620235 - Posted: 29 Dec 2014, 16:52:45 UTC
Last modified: 29 Dec 2014, 16:53:01 UTC

The unroll might be to high for your host.

Try the following.

-use_sleep -unroll 6 -oclfft_plan 256 16 256 -ffa_block 2830 -ffa_block_fetch 2830 -tune 1 64 4 1 -tune 2 64 4 1

Finnish a few tasks so i can check it later.


With each crime and every kindness we birth our future.
ID: 1620235 · Report as offensive
WezH
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 576
Credit: 67,033,957
RAC: 95
Finland
Message 1620246 - Posted: 29 Dec 2014, 17:59:53 UTC - in response to Message 1620235.  
Last modified: 29 Dec 2014, 18:27:19 UTC

The unroll might be to high for your host.

Try the following.

-use_sleep -unroll 6 -oclfft_plan 256 16 256 -ffa_block 2830 -ffa_block_fetch 2830 -tune 1 64 4 1 -tune 2 64 4 1

Finnish a few tasks so i can check it later.


Ok, cmd-line options are in place, running

<app>
<name>astropulse_v7</name>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>


Let's see, using this host as I normally do.

What I did understand from readme was that -unroll will increase applications GPU memory consumption.

EDIT: First one to run with these parameters: http://setiathome.berkeley.edu/result.php?resultid=3879941524. Not looking fast, running 27 minutes and 32% done.
"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -
ID: 1620246 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34258
Credit: 79,922,639
RAC: 80
Germany
Message 1620262 - Posted: 29 Dec 2014, 18:49:23 UTC

First i need to see how app behaves.
We will increase unroll step by step.

As soon this task finnished successfully increase unroll to 8.
But only unroll please.


With each crime and every kindness we birth our future.
ID: 1620262 · Report as offensive
WezH
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 576
Credit: 67,033,957
RAC: 95
Finland
Message 1620264 - Posted: 29 Dec 2014, 18:54:39 UTC - in response to Message 1620262.  

First i need to see how app behaves.
We will increase unroll step by step.

As soon this task finnished successfully increase unroll to 8.
But only unroll please.


OK Mike, I will do it.

And: Thank You.
"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -
ID: 1620264 · Report as offensive
WezH
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 576
Credit: 67,033,957
RAC: 95
Finland
Message 1620288 - Posted: 29 Dec 2014, 19:31:50 UTC - in response to Message 1620262.  
Last modified: 29 Dec 2014, 19:33:28 UTC

First i need to see how app behaves.
We will increase unroll step by step.

As soon this task finnished successfully increase unroll to 8.
But only unroll please.


Task http://setiathome.berkeley.edu/result.php?resultid=3879941524 completed, now unroll 8 now with task http://setiathome.berkeley.edu/result.php?resultid=3880025588
"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -
ID: 1620288 · Report as offensive
Highlander
Avatar

Send message
Joined: 5 Oct 99
Posts: 167
Credit: 37,987,668
RAC: 16
Germany
Message 1620303 - Posted: 29 Dec 2014, 20:06:32 UTC

WezH, how is your memory config? Single or Dual-Channel setup (aka 1x 2GB or 2x 1GB MemModule)?
- Performance is not a simple linear function of the number of CPUs you throw at the problem. -
ID: 1620303 · Report as offensive
JarrettH

Send message
Joined: 14 Nov 02
Posts: 97
Credit: 25,385,250
RAC: 95
Canada
Message 1620331 - Posted: 29 Dec 2014, 22:05:06 UTC
Last modified: 29 Dec 2014, 22:06:03 UTC

Hey Mike, maybe you know the answer to this while we're on the topic of configuring GPUs. I asked what commands I should use for a GT 730 here:

http://setiathome.berkeley.edu/forum_thread.php?id=75996&postid=1618100

Is processing one task optimal? My 550 Ti does two fastest.

It's the newer 730 with specs as on far right:

http://www.geforce.com/hardware/desktop-gpus/geforce-gt-730/specifications

Thanks
ID: 1620331 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34258
Credit: 79,922,639
RAC: 80
Germany
Message 1620333 - Posted: 29 Dec 2014, 22:27:08 UTC - in response to Message 1620331.  

Hey Mike, maybe you know the answer to this while we're on the topic of configuring GPUs. I asked what commands I should use for a GT 730 here:

http://setiathome.berkeley.edu/forum_thread.php?id=75996&postid=1618100

Is processing one task optimal? My 550 Ti does two fastest.

It's the newer 730 with specs as on far right:

http://www.geforce.com/hardware/desktop-gpus/geforce-gt-730/specifications

Thanks


I need to know how many CU`s the 730 has.
One instance on this card should be faster.


With each crime and every kindness we birth our future.
ID: 1620333 · 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 1620347 - Posted: 29 Dec 2014, 22:49:54 UTC - in response to Message 1620335.  

Hey Wez,

Sorry been gone all day, my day off so almost never online. Saw Mike is here, best person to help you. Hope it all gets sorted out.

Zalster
ID: 1620347 · Report as offensive
JarrettH

Send message
Joined: 14 Nov 02
Posts: 97
Credit: 25,385,250
RAC: 95
Canada
Message 1620362 - Posted: 29 Dec 2014, 23:10:37 UTC - in response to Message 1620333.  

Hey Mike, maybe you know the answer to this while we're on the topic of configuring GPUs. I asked what commands I should use for a GT 730 here:

http://setiathome.berkeley.edu/forum_thread.php?id=75996&postid=1618100

Is processing one task optimal? My 550 Ti does two fastest.

It's the newer 730 with specs as on far right:

http://www.geforce.com/hardware/desktop-gpus/geforce-gt-730/specifications

Thanks


I need to know how many CU`s the 730 has.
One instance on this card should be faster.


It has two CU
ID: 1620362 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34258
Credit: 79,922,639
RAC: 80
Germany
Message 1620366 - Posted: 29 Dec 2014, 23:16:48 UTC - in response to Message 1620362.  

Hey Mike, maybe you know the answer to this while we're on the topic of configuring GPUs. I asked what commands I should use for a GT 730 here:

http://setiathome.berkeley.edu/forum_thread.php?id=75996&postid=1618100

Is processing one task optimal? My 550 Ti does two fastest.

It's the newer 730 with specs as on far right:

http://www.geforce.com/hardware/desktop-gpus/geforce-gt-730/specifications

Thanks


I need to know how many CU`s the 730 has.
One instance on this card should be faster.


It has two CU


Something to start with.
-use_sleep -unroll 4 -oclfft_plan 256 16 512.
Maybe some further tuning possible.
But i want to see a few results first.


With each crime and every kindness we birth our future.
ID: 1620366 · Report as offensive
JarrettH

Send message
Joined: 14 Nov 02
Posts: 97
Credit: 25,385,250
RAC: 95
Canada
Message 1620377 - Posted: 29 Dec 2014, 23:34:26 UTC

I'll try that. It is going in the i3 2100 machine if you're wondering. What is -use_sleep for?
ID: 1620377 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34258
Credit: 79,922,639
RAC: 80
Germany
Message 1620378 - Posted: 29 Dec 2014, 23:38:09 UTC - in response to Message 1620377.  

I'll try that. It is going in the i3 2100 machine if you're wondering. What is -use_sleep for?


It reduces CPU usage.


With each crime and every kindness we birth our future.
ID: 1620378 · Report as offensive
JarrettH

Send message
Joined: 14 Nov 02
Posts: 97
Credit: 25,385,250
RAC: 95
Canada
Message 1620406 - Posted: 30 Dec 2014, 0:35:08 UTC

Do you think my 550 Ti machine is doing ok as well? I've haven't tinkered with it in a while. One core on an E6600 is free and the 550 Ti crunches two tasks.

I believe it's using this:

-unroll 10 -ffa_block 6144 -ffa_block_fetch 1536 -hp

Are those settings still current? The -oclFFT_plan switch is new to me. Could that be added? And finally, does ffa_block and fetch not apply to the 730?

Thanks for all your help, Mike.
ID: 1620406 · Report as offensive
WezH
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 576
Credit: 67,033,957
RAC: 95
Finland
Message 1620611 - Posted: 30 Dec 2014, 9:40:41 UTC - in response to Message 1620288.  

First i need to see how app behaves.
We will increase unroll step by step.

As soon this task finnished successfully increase unroll to 8.
But only unroll please.


Task http://setiathome.berkeley.edu/result.php?resultid=3879941524 completed, now unroll 8 now with task http://setiathome.berkeley.edu/result.php?resultid=3880025588


Unroll 8 has been working over night, but runtimes are still high.

What I should try next?
"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -
ID: 1620611 · Report as offensive
WezH
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 576
Credit: 67,033,957
RAC: 95
Finland
Message 1620612 - Posted: 30 Dec 2014, 9:41:22 UTC - in response to Message 1620303.  

WezH, how is your memory config? Single or Dual-Channel setup (aka 1x 2GB or 2x 1GB MemModule)?


Dual-channel, 2*1GB
"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -
ID: 1620612 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34258
Credit: 79,922,639
RAC: 80
Germany
Message 1620618 - Posted: 30 Dec 2014, 10:11:57 UTC - in response to Message 1620611.  
Last modified: 30 Dec 2014, 10:12:55 UTC

First i need to see how app behaves.
We will increase unroll step by step.

As soon this task finnished successfully increase unroll to 8.
But only unroll please.


Task http://setiathome.berkeley.edu/result.php?resultid=3879941524 completed, now unroll 8 now with task http://setiathome.berkeley.edu/result.php?resultid=3880025588


Unroll 8 has been working over night, but runtimes are still high.

What I should try next?


Now increase unroll to 10.
But only change unroll again.


With each crime and every kindness we birth our future.
ID: 1620618 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34258
Credit: 79,922,639
RAC: 80
Germany
Message 1620620 - Posted: 30 Dec 2014, 10:23:45 UTC - in response to Message 1620406.  

Do you think my 550 Ti machine is doing ok as well? I've haven't tinkered with it in a while. One core on an E6600 is free and the 550 Ti crunches two tasks.

I believe it's using this:

-unroll 10 -ffa_block 6144 -ffa_block_fetch 1536 -hp

Are those settings still current? The -oclFFT_plan switch is new to me. Could that be added? And finally, does ffa_block and fetch not apply to the 730?

Thanks for all your help, Mike.


It could do better.
First you are suffering from high CPU usage, this requires -use_sleep switch.
-oclfft_plan is a new method to optimze fft kernel planning.
Read the read me`s.

Try

-use_sleep -unroll 10 -oclfft_plan 256 16 256 -ffa_block 2830 -ffa_block_fetch 2830 -tune 1 64 4 1 -tune 2 64 4 1.
Maybe unroll 12 is possible on the 550Ti.

On the 730 ffa_block values wont give much improvement.
We can still add it after you finnished a couple units.


With each crime and every kindness we birth our future.
ID: 1620620 · Report as offensive
WezH
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 576
Credit: 67,033,957
RAC: 95
Finland
Message 1620713 - Posted: 30 Dec 2014, 13:12:19 UTC - in response to Message 1620618.  

Now increase unroll to 10.
But only change unroll again.


So far so good, no freezes.

http://setiathome.berkeley.edu/result.php?resultid=3880635022
http://setiathome.berkeley.edu/result.php?resultid=3880568416
http://setiathome.berkeley.edu/result.php?resultid=3880809911

So next step will be?
"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -
ID: 1620713 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34258
Credit: 79,922,639
RAC: 80
Germany
Message 1620715 - Posted: 30 Dec 2014, 13:22:16 UTC - in response to Message 1620713.  

Now increase unroll to 10.
But only change unroll again.


So far so good, no freezes.

http://setiathome.berkeley.edu/result.php?resultid=3880635022
http://setiathome.berkeley.edu/result.php?resultid=3880568416
http://setiathome.berkeley.edu/result.php?resultid=3880809911

So next step will be?


Now change to unroll 12.


With each crime and every kindness we birth our future.
ID: 1620715 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 · Next

Message boards : Number crunching : The GTX750(Ti) 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.