Consensus on GTX460 768MB, 2 or 3 WUs = best?

Message boards : Number crunching : Consensus on GTX460 768MB, 2 or 3 WUs = best?
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Profile Gecko
Volunteer tester
Avatar

Send message
Joined: 17 Nov 99
Posts: 454
Credit: 6,946,910
RAC: 47
United States
Message 1067320 - Posted: 16 Jan 2011, 16:08:09 UTC

Have seen comments in various threads re: GTX 460 & number of WUs in parallel for best production.

Any consensus if 2 or 3 is ideal on cards w/ 768MB memory?
ID: 1067320 · Report as offensive
baron_iv
Volunteer tester
Avatar

Send message
Joined: 4 Nov 02
Posts: 109
Credit: 104,905,241
RAC: 0
United States
Message 1067337 - Posted: 16 Jan 2011, 16:39:20 UTC - in response to Message 1067320.  

I run 2 and it works perfectly with both the 768mb and the 1gb model. I'm guessing you could run three, but generally that's only done on the 480/570/580 models. As far as memory goes, you'd likely be fine with three, but it will put a lot of stress on your GPU. Of course each GPU is different as well, even within the same series. You're welcome to try three and monitor the times for a few days (don't go by just 1 or 2 sets of work units, because the sizes of the work units can be pretty radically different). I've run 2 for quite a while now with no hiccups whatsoever.

If you do decide to run 3, please report back with your results because there will be others with the same question in the future.

Best of luck to you and happy crunching.
-baron_iv
Proud member of:
GPU Users Group
ID: 1067337 · Report as offensive
Profile Tim Norton
Volunteer tester
Avatar

Send message
Joined: 2 Jun 99
Posts: 835
Credit: 33,540,164
RAC: 0
United Kingdom
Message 1067338 - Posted: 16 Jan 2011, 16:42:11 UTC - in response to Message 1067320.  

Gecko

With 768 meg it would be two - limitation as less ram as you need circ 300meg per work unit

depending on what you clock speeds are but approx 22-24 mins each when running 2 at once - should get ~90% gpu utilisation say 5 per hour

with 1gig it would be three

I have messed about with the 4 460's i have and if you leave enough cpu threads available to feed the card(s) you will get slightly better performance from 3 wu on a 1gig card

if you use the machine with the gpu in then probably best at two per card especially if you have win7/vista and aero on but 3 wu will crunch ok at approx 95+% utilisation and 28-30 mins each say 6 per hour - i have dual monitors running on one with full aero in win7 and mem load is currently 919meg
Tim

ID: 1067338 · Report as offensive
Profile arkayn
Volunteer tester
Avatar

Send message
Joined: 14 May 99
Posts: 4438
Credit: 55,006,323
RAC: 0
United States
Message 1067461 - Posted: 16 Jan 2011, 22:58:36 UTC

I run 2 at a time on my 460, current RAC is about 12k with 4k coming from the CPU.

ID: 1067461 · Report as offensive
Profile Paul D Harris
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 1122
Credit: 33,600,005
RAC: 0
United States
Message 1067522 - Posted: 17 Jan 2011, 1:46:05 UTC

How would a person edit their app_info file for three wu's?
ID: 1067522 · Report as offensive
-BeNt-
Avatar

Send message
Joined: 17 Oct 99
Posts: 1234
Credit: 10,116,112
RAC: 0
United States
Message 1067538 - Posted: 17 Jan 2011, 3:19:09 UTC - in response to Message 1067522.  
Last modified: 17 Jan 2011, 3:19:44 UTC

How would a person edit their app_info file for three wu's?


Change your count to .33.


<coproc>
        <type>CUDA</type>
        <count>.33</count>
</coproc>


Obviously .50 does 2 and 1 only does 1.
Traveling through space at ~67,000mph!
ID: 1067538 · Report as offensive
Profile Manuel Palacios

Send message
Joined: 2 Nov 99
Posts: 74
Credit: 30,209,980
RAC: 56
Venezuela
Message 1067570 - Posted: 17 Jan 2011, 6:06:18 UTC - in response to Message 1067320.  

Regardless of 768 MB or 1GB, 2 WU's in parallel is best for a GTX 460. Look at the GTX 460 thread where I documented my findings if you need further clarifications :)
ID: 1067570 · Report as offensive
Profile Gecko
Volunteer tester
Avatar

Send message
Joined: 17 Nov 99
Posts: 454
Credit: 6,946,910
RAC: 47
United States
Message 1067665 - Posted: 17 Jan 2011, 16:03:03 UTC - in response to Message 1067570.  

Thanks for the replies.
Responses confirm my perceptions re: 2.

Thanks again!
ID: 1067665 · Report as offensive
JohnDK Crowdfunding Project Donor*Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 28 May 00
Posts: 1222
Credit: 451,243,443
RAC: 1,127
Denmark
Message 1067711 - Posted: 17 Jan 2011, 18:03:11 UTC

On my GTX 460 3 is better. Doing a test on midrange WUs I got 2 at a time = 11:23 per WU, 3 = 10:42.
ID: 1067711 · Report as offensive
Profile Manuel Palacios

Send message
Joined: 2 Nov 99
Posts: 74
Credit: 30,209,980
RAC: 56
Venezuela
Message 1067755 - Posted: 17 Jan 2011, 20:31:57 UTC - in response to Message 1067711.  

On my GTX 460 3 is better. Doing a test on midrange WUs I got 2 at a time = 11:23 per WU, 3 = 10:42.


That's quite interesting, may I ask exactly what your WU throughput is per day and what clock speeds you have your GPU running at?

I'm quite curious as I run my 1 GB GTX 460 24/7 with 2 WU's in parallel and achieve anywhere between 129-140 WU's per day. When I ran it with 3 WU's in parallel I achieved about 10 WUs less per day consistently.

It's important that I mention my clock speeds, as you may have set yours to ones higher than mine which could explain the difference in throughput. Also I was overlooking an important fact, my CPU is tied up 100% with Einstein WU's and so SETI has to compete with Einstein when WU preparation needs to occur. I'm not sure if this is a negligible difference or not though.


ID: 1067755 · Report as offensive
Profile Tim Norton
Volunteer tester
Avatar

Send message
Joined: 2 Jun 99
Posts: 835
Credit: 33,540,164
RAC: 0
United Kingdom
Message 1067779 - Posted: 17 Jan 2011, 21:52:21 UTC - in response to Message 1067755.  

Manuel

My times on both my dual 460 rigs are similar to John's see beginning of the thread

the difference you may see is that you appear to be using all your cpu for Einstein so to improve your gpu throughput allow one thread to be idle to feed your GPU and you will see 3 wu at once to be ~10-15% better per Day on average

its not a huge difference but it is there :)
Tim

ID: 1067779 · Report as offensive
Profile dcappello
Avatar

Send message
Joined: 3 Apr 99
Posts: 261
Credit: 170,969,320
RAC: 0
United States
Message 1067792 - Posted: 17 Jan 2011, 22:37:06 UTC

I just tweeked a config on a machine with a 480 in it to run 3 at a time:
http://setiathome.berkeley.edu/show_host_detail.php?hostid=5624447

That card does have 1.5G onboard. We can keep on eye on it to see how it does.
ID: 1067792 · Report as offensive
Profile Westsail and *Pyxey*
Volunteer tester
Avatar

Send message
Joined: 26 Jul 99
Posts: 338
Credit: 20,544,999
RAC: 0
United States
Message 1067859 - Posted: 18 Jan 2011, 1:18:37 UTC
Last modified: 18 Jan 2011, 1:19:42 UTC

Running GTX460 1G x2 here...Seems to be leveling out around 31k rac with 2 per GPU (4 total)..
Just switched to 3 per card:
First impression, even at 99% load my power meter would fluctuate a bit /w occasion dips. Now, much more solid at ~10-15 watts fluctuation under full load.
With out those dips, i'm betting throughput sees an increase.
So far so good...also gui seems perfectly usable still.
"The most exciting phrase to hear in science, the one that heralds new discoveries, is not Eureka! (I found it!) but rather, 'hmm... that's funny...'" -- Isaac Asimov
ID: 1067859 · Report as offensive
Profile Michael W.F. Miles
Avatar

Send message
Joined: 24 Mar 07
Posts: 268
Credit: 34,410,870
RAC: 0
Canada
Message 1067898 - Posted: 18 Jan 2011, 5:04:54 UTC

I have the Zotac 460 GTX 768 and I find 2 6.10 wu at a time is good for
16000 rac.
I run a Phenom 2 965 @3.7 and the video card at 800 from 675 stock.

Solid temps at 40 - 50 C and am very pleased with this card.
The only drawback was switching back to windows from a Linux 64 os.
I could not get this card to work at all with the Cuda 3 opt apps 6.09 with
or without vlarkill.
So I switched to Win 7 x64 for the time being until the cuda thing gets worked out with Linux.

I so like the security and speed that Linux offers my computer compared to win 7
but I will just have to get by with everyone in my computer compared to waiting at the gate with Linux.
I could not believe the viruses coming back to windows after a year on Linux

Anyway a question about running 3 apps
setting the count to 0.33 is leaving a uneven division.
How does that affect the overall performance sine .01 is not being used
ID: 1067898 · 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 1068014 - Posted: 18 Jan 2011, 16:50:44 UTC - in response to Message 1067538.  

Brand new to CUDA computing on a GTX460. I'm confused just in what part of app_info.xml to modify the "count". I see multiple count statements in many parts of the file. I would assume it would go somewhere in the cuda part of the file, but I see multiple entries for count statements in multiple cuda entries. Do I just modify all the cuda entries? I would think you would leave the count statement for CPU apps alone. Any help in my understanding?

Cheers, Keith
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1068014 · Report as offensive
Profile Tim Norton
Volunteer tester
Avatar

Send message
Joined: 2 Jun 99
Posts: 835
Credit: 33,540,164
RAC: 0
United Kingdom
Message 1068018 - Posted: 18 Jan 2011, 17:12:08 UTC - in response to Message 1068014.  
Last modified: 18 Jan 2011, 17:13:30 UTC

Keith

just change

<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>

to say for two wu at a time

<coproc>
<type>CUDA</type>
<count>0.5</count>
</coproc>

or 0.33 for three depending on the card memeory

it occurs 3 or 6 time depending on your setup (32 or 64bit) change all of them to be safe until you understand difference by plan class etc

Edit: sorry the "type" will also be different for each entry - just change the number!
Tim

ID: 1068018 · Report as offensive
Profile SciManStev Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Jun 99
Posts: 6651
Credit: 121,090,076
RAC: 0
United States
Message 1068030 - Posted: 18 Jan 2011, 21:31:55 UTC - in response to Message 1067898.  

The only drawback was switching back to windows from a Linux 64 os.
I could not get this card to work at all with the Cuda 3 opt apps 6.09 with
or without vlarkill.


Fermi GPU's are not compatible with that version. Go to Lunatics and download the x32f build. Check back often for new releases, as they won't auto update.

http://lunatics.kwsn.net/index.php?module=Downloads;catd=9

Steve
Warning, addicted to SETI crunching!
Crunching as a member of GPU Users Group.
GPUUG Website
ID: 1068030 · Report as offensive
xx
Volunteer tester

Send message
Joined: 23 May 99
Posts: 166
Credit: 3,450,910
RAC: 0
United States
Message 1068031 - Posted: 18 Jan 2011, 21:48:21 UTC

I run three at a time on the GTX 480 OCd to 800/1600/1900 and am getting about a 23,600 RAC out of it. Five CPU threads are processing climateprediction.net and one is free. Hyperthreading off. i7-970 CPU OCd to 3800 MHz (181 X 21).

Chuck
ID: 1068031 · Report as offensive
Profile Michael W.F. Miles
Avatar

Send message
Joined: 24 Mar 07
Posts: 268
Credit: 34,410,870
RAC: 0
Canada
Message 1068037 - Posted: 18 Jan 2011, 22:12:15 UTC - in response to Message 1068030.  

Yes, I have been keeping an eye out for these apps to become available.

if anyone else sees them before I do don't be afraid to let me know.

Thanks in advance

Michael Miles
Amiga5@live.com
ID: 1068037 · Report as offensive
Profile SciManStev Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Jun 99
Posts: 6651
Credit: 121,090,076
RAC: 0
United States
Message 1068041 - Posted: 18 Jan 2011, 22:23:30 UTC - in response to Message 1068037.  
Last modified: 18 Jan 2011, 22:49:31 UTC

Yes, I have been keeping an eye out for these apps to become available.

if anyone else sees them before I do don't be afraid to let me know.

Thanks in advance

Michael Miles
Amiga5@live.com

The x32f build is fully Fermi compatible, and available. The older V12 build was where the Fermi problem was. Now that VLAR's are only scheduled to the CPU, the VLARKILL is no longer needed.

Steve
Warning, addicted to SETI crunching!
Crunching as a member of GPU Users Group.
GPUUG Website
ID: 1068041 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : Consensus on GTX460 768MB, 2 or 3 WUs = best?


 
©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.