Optimize your GPU. Find the value the easy way.


log in

Advanced search

Message boards : Number crunching : Optimize your GPU. Find the value the easy way.

1 · 2 · 3 · 4 . . . 13 · Next
Author Message
Profile S@NL - eFMer - efmer.com/boincProject donor
Volunteer tester
Avatar
Send message
Joined: 7 Jun 99
Posts: 512
Credit: 130,277,243
RAC: 34,318
United States
Message 1275307 - Posted: 25 Aug 2012, 10:25:29 UTC
Last modified: 25 Aug 2012, 10:46:23 UTC

NVIDIA GPU's of the 4xx, 5xx, 6xx series normally run better when more than one instance runs / device.
But this is difficult to do manually.

This tool does this automatically and you get a result within 15 minutes.

http://www.efmer.eu/forum_tt/index.php?topic=974.0

Warning, this tool is tested on a limited number of machines. (V 1.0)

It uses the Lunatics optimized exe as a reference.
But you can drop in any version you like as long as you update the config.xml file.


My results:

GTX 295: 260 sec / device = 130 sec / board
GTX 480: 98 sec / device = 98 sec / board
GTX 590: 104 sec / device = 52 sec / board

Interesting to see what a 690 does.
____________
TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking.

Kerwin
Avatar
Send message
Joined: 7 Dec 99
Posts: 18
Credit: 10,046,844
RAC: 1,421
United States
Message 1275348 - Posted: 25 Aug 2012, 12:47:44 UTC - in response to Message 1275307.

Nice tool! Worked fine on my system. :-)

Thanks!
____________

Kamu
Send message
Joined: 19 Jan 02
Posts: 56
Credit: 9,810,425
RAC: 0
Finland
Message 1275369 - Posted: 25 Aug 2012, 13:37:31 UTC
Last modified: 25 Aug 2012, 13:40:49 UTC

Nice little app! Automatic test's result was: count 0.5 fastest.

GTX 690, Devices 2, Count 0.50

x41g

Results: Device: 0, device count: 2, average time / count: 232, average time on device: 116 Seconds (1 Minutes, 56 Seconds) Device: 1, device count: 2, average time / count: 234, average time on device: 117 Seconds (1 Minutes, 57 Seconds)


x41z
Results: Device: 0, device count: 2, average time / count: 174, average time on device: 87 Seconds (1 Minutes, 27 Seconds) Device: 1, device count: 2, average time / count: 178, average time on device: 89 Seconds (1 Minutes, 29 Seconds)


-Kimmo-
____________
Computers: obelix

Profile Tazz
Volunteer tester
Avatar
Send message
Joined: 5 Oct 99
Posts: 137
Credit: 34,342,390
RAC: 0
Canada
Message 1275370 - Posted: 25 Aug 2012, 13:38:28 UTC - in response to Message 1275307.
Last modified: 25 Aug 2012, 13:40:06 UTC

Great app!!
EVGA GTX 560:
count: 1, average time / count: 217, average time on device: 217 Seconds (3 Minutes, 37 Seconds)

count: 2, average time / count: 311, average time on device: 155 Seconds (2 Minutes, 35 Seconds)

count: 3, average time / count: 450, average time on device: 150 Seconds (2 Minutes, 30 Seconds)

Running one at a time the GPU load was in the low 60s, two at a time was low 90s, three at a time was high 90s. Four at a time was low 50s with screen lag. Is this normal (the low 51-54 GPU%)? I ended up cancelling the four at a time test because it was still running after 35 minutes.


____________
</Tazz>

Profile S@NL - eFMer - efmer.com/boincProject donor
Volunteer tester
Avatar
Send message
Joined: 7 Jun 99
Posts: 512
Credit: 130,277,243
RAC: 34,318
United States
Message 1275375 - Posted: 25 Aug 2012, 13:46:05 UTC - in response to Message 1275370.

EVGA GTX 560:
Running one at a time the GPU load was in the low 60s, two at a time was low 90s, three at a time was high 90s. Four at a time was low 50s with screen lag. Is this normal (the low 51-54 GPU%)? I ended up cancelling the four at a time test because it was still running after 35 minutes.

The program stops at 4 programs / device.
But cuda tasks may run into problems and get stuck. Probably a memory problem in the card.
But in your case the CPU may not be able to keep up.

The cancel button, kills the running cuda tasks and clears up the disk.

4 is too much for some cards and the overhead of the CPU increases, but that depends on the CPU.

As you can see the gain from 2 -> 3 is already low.

____________
TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking.

Profile S@NL - eFMer - efmer.com/boincProject donor
Volunteer tester
Avatar
Send message
Joined: 7 Jun 99
Posts: 512
Credit: 130,277,243
RAC: 34,318
United States
Message 1275380 - Posted: 25 Aug 2012, 13:52:54 UTC - in response to Message 1275369.

Nice little app! Automatic test's result was: count 0.5 fastest.

GTX 690, Devices 2, Count 0.50


With the 41g the 690 is actually a bit slower than the 590.
The 41z is quite a bit better.
____________
TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking.

Profile jason_gee
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 5057
Credit: 73,920,738
RAC: 12,036
Australia
Message 1275417 - Posted: 25 Aug 2012, 16:21:28 UTC - in response to Message 1275380.
Last modified: 25 Aug 2012, 16:21:58 UTC

Hi Fred,

Cool tool :) can you make it cope with either command line parameters for the exe, or not die if a mbcuda.cfg (x41z) configuration file is present ? (it dies when trying to copy into the slot for some reason, but it does put the file there). It would be handy for finding good settings, for which x41z has a few that the best setting could depend on the count.

Jason
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

Profile Fred E.Project donor
Volunteer tester
Send message
Joined: 22 Jul 99
Posts: 768
Credit: 24,139,004
RAC: 0
United States
Message 1275418 - Posted: 25 Aug 2012, 16:23:10 UTC

Results with gtx670, driver 304.79, XP 32 bit:

Device: 0, device count: 1, average time / count: 139, average time on device: 139 Seconds (2 Minutes, 19 Seconds)

Device: 0, device count: 2, average time / count: 261, average time on device: 130 Seconds (2 Minutes, 10 Seconds)

Device: 0, device count: 3, average time / count: 389, average time on device: 129 Seconds (2 Minutes, 9 Seconds)

The best average time found: 130 Seconds (2 Minutes, 10 Seconds), with count: 0.50 (2)

1) Why is 130 seconds better than 129? :=)

2) Using the automatic option and a count of .25, it did not test for 4 at a time. I ran that test separately and got:

Device: 0, device count: 4, average time / count: 485, average time on device: 121 Seconds (2 Minutes, 1 Seconds)

So I guess I'll stay with 4 at a time.
____________
Another Fred
Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop.

Profile jason_gee
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 5057
Credit: 73,920,738
RAC: 12,036
Australia
Message 1275426 - Posted: 25 Aug 2012, 16:50:59 UTC

GTX 680, win7x64sp1, driver 304.79, lunatics_x41z_win32_cuda50preview.exe,
at default settings (no mbcuda.cfg or command line switches) :


Device: 0, device count: 1, average time / count: 88, average time on device: 88 Seconds (1 Minutes, 28 Seconds)

Device: 0, device count: 2, average time / count: 155, average time on device: 77 Seconds (1 Minutes, 17 Seconds)

Device: 0, device count: 3, average time / count: 229, average time on device: 76 Seconds (1 Minutes, 16 Seconds)

Device: 0, device count: 4, average time / count: 302, average time on device: 75 Seconds (1 Minutes, 15 Seconds)


Any chance of adding 5 & 6 ?
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

Profile arkaynProject donor
Volunteer tester
Avatar
Send message
Joined: 14 May 99
Posts: 3695
Credit: 48,765,402
RAC: 6,137
United States
Message 1275428 - Posted: 25 Aug 2012, 16:56:11 UTC - in response to Message 1275417.

Hi Fred,

Cool tool :) can you make it cope with either command line parameters for the exe, or not die if a mbcuda.cfg (x41z) configuration file is present ? (it dies when trying to copy into the slot for some reason, but it does put the file there). It would be handy for finding good settings, for which x41z has a few that the best setting could depend on the count.

Jason


Or more than 1 app or multiple versions of the dll files.
____________

Profile S@NL - eFMer - efmer.com/boincProject donor
Volunteer tester
Avatar
Send message
Joined: 7 Jun 99
Posts: 512
Credit: 130,277,243
RAC: 34,318
United States
Message 1275430 - Posted: 25 Aug 2012, 17:00:14 UTC - in response to Message 1275417.

Hi Fred,

Cool tool :) can you make it cope with either command line parameters for the exe, or not die if a mbcuda.cfg (x41z) configuration file is present ? (it dies when trying to copy into the slot for some reason, but it does put the file there). It would be handy for finding good settings, for which x41z has a few that the best setting could depend on the count.

Jason

Easy, the parameter is now --device x to set the right device.
I can make a field where you can put additional.

Like -- devices x additional

PS can you give me the x41z for testing?

____________
TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking.

Profile S@NL - eFMer - efmer.com/boincProject donor
Volunteer tester
Avatar
Send message
Joined: 7 Jun 99
Posts: 512
Credit: 130,277,243
RAC: 34,318
United States
Message 1275433 - Posted: 25 Aug 2012, 17:01:42 UTC - in response to Message 1275426.
Last modified: 25 Aug 2012, 17:02:10 UTC


Any chance of adding 5 & 6 ?

You can do that manually. (manual button)
On auto mode the program tens to hang after 4 for sure. Probably out of memory, but doesn't exit.

I could add a max in the config.xml.
____________
TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking.

Profile S@NL - eFMer - efmer.com/boincProject donor
Volunteer tester
Avatar
Send message
Joined: 7 Jun 99
Posts: 512
Credit: 130,277,243
RAC: 34,318
United States
Message 1275434 - Posted: 25 Aug 2012, 17:05:16 UTC - in response to Message 1275418.


1) Why is 130 seconds better than 129? :=)

2) Using the automatic option and a count of .25, it did not test for 4 at a time. I ran that test separately and got:

Device: 0, device count: 4, average time / count: 485, average time on device: 121 Seconds (2 Minutes, 1 Seconds)

So I guess I'll stay with 4 at a time.

1) Because the extra instance causes more overhead, I subtract 1 second. Not worth the extra CPU.

2) The auto option doesn't use count, only the devices.
The value is used when you press the button below it.
____________
TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking.

Profile jason_gee
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 5057
Credit: 73,920,738
RAC: 12,036
Australia
Message 1275435 - Posted: 25 Aug 2012, 17:05:17 UTC - in response to Message 1275430.

Hi Fred,

Cool tool :) can you make it cope with either command line parameters for the exe, or not die if a mbcuda.cfg (x41z) configuration file is present ? (it dies when trying to copy into the slot for some reason, but it does put the file there). It would be handy for finding good settings, for which x41z has a few that the best setting could depend on the count.

Jason

Easy, the parameter is now --device x to set the right device.
I can make a field where you can put additional.

Like -- devices x additional

PS can you give me the x41z for testing?


Sure I can authorise you for testing x41z ;) , please PM address (sure I had it, but not sure where)
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

Profile S@NL - eFMer - efmer.com/boincProject donor
Volunteer tester
Avatar
Send message
Joined: 7 Jun 99
Posts: 512
Credit: 130,277,243
RAC: 34,318
United States
Message 1275437 - Posted: 25 Aug 2012, 17:08:23 UTC - in response to Message 1275428.

Or more than 1 app or multiple versions of the dll files.

I could add a dropdown box to select different tests to copy.

Adding e.g. 3 entries into the config.xml

Like
x40g
x41
x42

And maps data, data1, data2 .....
____________
TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking.

Joe JanuzziProject donor
Avatar
Send message
Joined: 13 Apr 03
Posts: 3
Credit: 117,587,603
RAC: 92,780
United States
Message 1275446 - Posted: 25 Aug 2012, 17:37:28 UTC

+1
Fred,
Thanks for giving us another great app.
Joe

GTX680 (2) EVGA SC+ w/Backplate Part# 02G-P4-2684-KR. 1215 MHz, 6208 MHz Memory, Volts 1.175 SABERTOOTH Z77/ i7-3770K (4.4 GHz), win7x64sp1,
driver 301.42

Count: 1, Temp 52c, Memory 314 MB, Usage 78%, Power 87%
Count: 2, Temp 54c, Memory 566 MB, Usage 98%, Power 80%
Count: 3, Temp 55c, Memory 818 MB, Usage 99%, Power 80%
Count: 4, Temp 55c, Memory 1071 MB, Usage 99%, Power 81%
Ambient Temp 27c

Device: 0, device count: 1, average time / count: 119, average time on device: 119 Seconds (1 Minutes, 59 Seconds)
Device: 0, device count: 2, average time / count: 216, average time on device: 108 Seconds (1 Minutes, 48 Seconds)
Device: 0, device count: 3, average time / count: 323, average time on device: 107 Seconds (1 Minutes, 47 Seconds)
Device: 0, device count: 4, average time / count: 429, average time on device: 107 Seconds (1 Minutes, 47 Seconds)

So I'm going to try 4 wu's each on my GPU's (no HT, 2 core's)
____________
Try to learn something new everyday.

Profile S@NL - eFMer - efmer.com/boincProject donor
Volunteer tester
Avatar
Send message
Joined: 7 Jun 99
Posts: 512
Credit: 130,277,243
RAC: 34,318
United States
Message 1275494 - Posted: 25 Aug 2012, 19:51:27 UTC - in response to Message 1275446.


So I'm going to try 4 wu's each on my GPU's (no HT, 2 core's)

0.5 should be the best.
____________
TThrottle Control your temperatures. BoincTasks The best way to view BOINC. Anza Borrego Desert hiking.

Joe JanuzziProject donor
Avatar
Send message
Joined: 13 Apr 03
Posts: 3
Credit: 117,587,603
RAC: 92,780
United States
Message 1275519 - Posted: 25 Aug 2012, 21:13:12 UTC

Thanks
Changed GPU to 0.5 (CPU same)
____________
Try to learn something new everyday.

juan BFBProject donor
Volunteer tester
Avatar
Send message
Joined: 16 Mar 07
Posts: 5421
Credit: 308,618,747
RAC: 357,860
Brazil
Message 1275565 - Posted: 25 Aug 2012, 23:45:46 UTC - in response to Message 1275494.
Last modified: 26 Aug 2012, 0:16:09 UTC


So I'm going to try 4 wu's each on my GPU's (no HT, 2 core's)

0.5 should be the best.


Hi

I´m a x41z user, your build works ok in some hosts but when i try to use in this host: 6690764 (with just one 590 EVGA) its give me 100 sec in all measures (1, 2 or 3 WU), and shows just 1 device (the 2 GPUs on the 590 works OK with Boinc processing 2x2WU at a time XP32 and no OC). Do you know anythink that could cause this?
____________

w1hueProject donor
Volunteer tester
Send message
Joined: 4 Aug 00
Posts: 48
Credit: 1,811,386
RAC: 964
United States
Message 1275613 - Posted: 26 Aug 2012, 4:09:05 UTC

Well, I can't get it to work. I have one NVIDIA GT 520 installed. When I run the program with the default "Devices 1", "Count 0.50", I get this:

Starting test: (lunatics_x41g_win32_cuda32_perf.exe)
25 August 2012 - 20:56:50 Start, devices: 1, device count: 2
25 August 2012 - 20:57:07 Program failed to start on slot: 0
25 August 2012 - 20:57:07 Program failed to start on slot: 0
25 August 2012 - 20:57:08 Program failed to start on slot: 0
25 August 2012 - 20:57:08 Program failed to start on slot: 0

So ... what am I not doing correctly? (Yes, I stopped BOINC Manager before executing the performance utility).

____________

1 · 2 · 3 · 4 . . . 13 · Next

Message boards : Number crunching : Optimize your GPU. Find the value the easy way.

Copyright © 2014 University of California