How fast are different cards?

Questions and Answers : GPU applications : How fast are different cards?
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · Next

AuthorMessage
HFB1217
Avatar

Send message
Joined: 25 Dec 05
Posts: 102
Credit: 9,424,572
RAC: 0
United States
Message 888301 - Posted: 25 Apr 2009, 18:04:57 UTC - in response to Message 888211.  

First go into your Bios and disable EIST that is a power saving function that keeps the CPU from running at full speed all the time. Secondly is your CUDA is not optimized then download Raistmer's optimizer

http://lunatics.kwsn.net/gpu-crunching/ak-v8-cuda-mb-team-work-mod.0.html

This will error out those VLAR workunits.

Hope this helps out.
Come and Visit Us at
BBR TeamStarFire


****My 9th year of Seti****A Founding Member of the Original Seti Team Starfire at Broadband Reports.com ****
ID: 888301 · Report as offensive
Profile perryjay
Volunteer tester
Avatar

Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 20,676,751
RAC: 0
United States
Message 888308 - Posted: 25 Apr 2009, 19:08:18 UTC - in response to Message 888301.  
Last modified: 25 Apr 2009, 19:10:12 UTC

Hank, you are a bit out of date. If he is running 6.6.20 he doesn't need the team mod anymore. Check around for one of Raistmer's stand alone mods and the correct app_info for your setup.

This thread http://setiathome.berkeley.edu/forum_thread.php?id=52589 will give you a good idea how to set it up.


PROUD MEMBER OF Team Starfire World BOINC
ID: 888308 · Report as offensive
(Ryle)

Send message
Joined: 18 Aug 05
Posts: 3
Credit: 1,916,218
RAC: 0
Denmark
Message 888432 - Posted: 26 Apr 2009, 8:56:45 UTC

I think you may all be right :)

I freed up a cpu-core as Gundolf suggested, and now it produces some of them in under 10 minutes. But I still get many of those VLAR's. I've been trying to keep an eye on the manager for the last day, and mostly notice the VLARs. That may be because the other types go so fast that they disappear before i notice them.

With regards to Raistmers mod, I'm not too keen on editing any boinc files. Not yet at least. So I'll keep that one in mind for now.

ID: 888432 · Report as offensive
Frederic Salve
Avatar

Send message
Joined: 13 Nov 00
Posts: 2
Credit: 2,528,859
RAC: 0
Vietnam
Message 897245 - Posted: 20 May 2009, 14:19:49 UTC
Last modified: 20 May 2009, 14:20:51 UTC

I just installed a Gigabyte GTX 260 (216 cores) OC @630MHz. It is crushing Seti Cuda WU in 10 minutes and 10 seconds average.
ID: 897245 · Report as offensive
Andy Williams
Volunteer tester
Avatar

Send message
Joined: 11 May 01
Posts: 187
Credit: 112,464,820
RAC: 0
United States
Message 897279 - Posted: 20 May 2009, 16:06:04 UTC

MSI GTX 260 factory OC @655 MHz 8 minutes 20 seconds.
--
Classic 82353 WU / 400979 h
ID: 897279 · Report as offensive
Profile TCP JESUS
Avatar

Send message
Joined: 19 Jan 03
Posts: 205
Credit: 1,248,845
RAC: 0
Canada
Message 898153 - Posted: 22 May 2009, 8:51:45 UTC

I have played around a bit with only a few cards, but have no real way of testing other than what BOINC spits out in the 'MESSAGE' section when it detects the CUDA devices upon startup. I am interested in more info as well, if some of the guys with GTX 200 series cards wouldn't mind sharing.....

For example:

CUDA devices: GeForce 9600 GSO (driver version 18585, CUDA version 1.1, 384MB, est. 62GFLOPS), GeForce 9600 GSO (driver version 18585, CUDA version 1.1, 768MB, est. 46GFLOPS)

In the machine above, I am running 2 x 9600 GSO (unmatched cards) in a Crossfire Motherboard. 1 card is stock, the other is overclocked quite a bit.

Here are a few cards that I have made note of when starting BIONC and what it estimates their performance as:
-----------------------------------------------------------------------------
- Quadro NVS 290, 256MB PCIe, 16 Cores - 5 GFLOPS
- Geforce 8500 GT, 512MB PCIe, 16 Cores - 6 GFLOPS

- Geforce 9600 GSO, 768MB (DDR2) PCIe, 96 Cores - 46 GFLOPS <---CURRENT SECONDARY CARD
- (same card above, overclocked by 5% +) ...... - 48 GFLOPS

- Geforce 9600 GSO, 384MB (DDR3) PCIe, 96 Cores - 48 GFLOPS
- (same card above, overclocked by 30% +) ..... - 62 GFLOPS <---CURRENT PRIMARY CARD
------------------------------------------------------------------------------


Anyone else care to share their BOINC message/estimation at startup ?

Thanks.
Allan
ID: 898153 · Report as offensive
Fred W
Volunteer tester

Send message
Joined: 13 Jun 99
Posts: 2524
Credit: 11,954,210
RAC: 0
United Kingdom
Message 898266 - Posted: 22 May 2009, 16:17:57 UTC - in response to Message 898153.  


...

Anyone else care to share their BOINC message/estimation at startup ?

Thanks.
Allan

CUDA device: GeForce GTX 295 (driver version 18585, compute capability 1.3, 896MB, est. 106GFLOPS)
CUDA device: GeForce GTX 295 (driver version 18585, compute capability 1.3, 896MB, est. 106GFLOPS)


F.

ID: 898266 · Report as offensive
Profile Hammeh
Volunteer tester
Avatar

Send message
Joined: 21 May 01
Posts: 135
Credit: 1,143,316
RAC: 0
United Kingdom
Message 898291 - Posted: 22 May 2009, 17:36:58 UTC

My 8500GT:
22/05/2009 16:13:02 CUDA device: GeForce 8500 GT (driver version 18585, compute capability 1.1, 512MB, est. 5GFLOPS)


Quite shockingly low!! I don't even bother running CUDA on it anyone, not worth the hassle of it overheating and causing system slowdowns.
ID: 898291 · Report as offensive
Profile KW2E
Avatar

Send message
Joined: 18 May 99
Posts: 346
Credit: 104,396,190
RAC: 34
United States
Message 898309 - Posted: 22 May 2009, 18:45:55 UTC

Here's some of the ones I have. I have not posted any dupes, although I have many and all vary in GFLOPS ratings for the same card on various systems. (OS?)

CUDA device: GeForce GTX 295 (driver version 18585, compute capability 1.3, 896MB, est. 106GFLOPS)

CUDA device: GeForce GTX 275 (driver version 18585, compute capability 1.3, 896MB, est. 123GFLOPS)

CUDA device: GeForce GTX 280 (driver version 18585, compute capability 1.3, 1024MB, est. 130GFLOPS)

CUDA device: GeForce GTX 260 (driver version 18585, compute capability 1.3, 896MB, est. 100GFLOPS)

CUDA device: GeForce 8400 GS (driver version 18585, compute capability 1.1, 256MB, est. 5GFLOPS) <-PCIe

CUDA device: GeForce 8400 GS (driver version 18585, compute capability 1.1, 256MB, est. 4GFLOPS) <-PCI

CUDA device: GeForce 8600 GT (driver version 18585, compute capability 1.1, 256MB, est. 14GFLOPS)

CUDA device: GeForce 8600M GS (driver version 18585, compute capability 1.1, 512MB, est. 6GFLOPS)

CUDA device: GeForce 9800 GT (driver version 18585, compute capability 1.1, 1024MB, est. 60GFLOPS)

CUDA device: GeForce 9800 GTX+ (driver version 18585, compute capability 1.1, 512MB, est. 84GFLOPS)

CUDA device: GeForce 9600 GT (driver version 18585, compute capability 1.1, 512MB, est. 34GFLOPS)

Rob




ID: 898309 · Report as offensive
Profile perryjay
Volunteer tester
Avatar

Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 20,676,751
RAC: 0
United States
Message 898311 - Posted: 22 May 2009, 18:46:36 UTC - in response to Message 898291.  

5/21/2009 11:49:35 AM CUDA device: GeForce 8500 GT (driver version 18585, compute capability 1.1, 512MB, est. 5GFLOPS)

I wasn't going to post in this until I saw Hammeh's post. I'm running a Celeron E1400 dual @ 2.0G. My 8500GT is giving me about the equivalent of another core and a half. Way slower than the big guns but a lot better than without.


PROUD MEMBER OF Team Starfire World BOINC
ID: 898311 · Report as offensive
Profile TCP JESUS
Avatar

Send message
Joined: 19 Jan 03
Posts: 205
Credit: 1,248,845
RAC: 0
Canada
Message 898382 - Posted: 22 May 2009, 21:24:11 UTC
Last modified: 22 May 2009, 21:32:20 UTC

After doing a little Linpack benching on a couple Q6600 Systems I have, it's easy to see why GPU crunchers reign supreme....even over 16 CPU Machines.

My quick findings....(most likely common knowledge already as well)

System #1 - Q6600 Quad @ 3.2Ghz - 10 GFLOPS/core (40 GFLOPS Total)

System #2 - Q6600 Quad @ STOCK/2.4Ghz - 6.5 GFLOPS/core (26 GFLOPS Total)

This leads me to this great find:
SETI GPU Cruncher's Dream Motherboard

($60 MSI 945GCM478-L) + (Socket.478 P4 $5 CPU) + ($10 worth of DDR2 Memory) + GTX 295 = MONSTER CRUNCHER !

...that is, unless Tri or Quad SLI is your preference and you already have a board capable ;)

Just a thought, perhaps not even a very good one...lol
Allan
I am TCP JESUS...The Carpenter Phenom Jesus....and HAMMERING is what I do best!
formerly known as...MC Hammer.
ID: 898382 · Report as offensive
Fred W
Volunteer tester

Send message
Joined: 13 Jun 99
Posts: 2524
Credit: 11,954,210
RAC: 0
United Kingdom
Message 898400 - Posted: 22 May 2009, 22:17:06 UTC - in response to Message 898382.  

After doing a little Linpack benching on a couple Q6600 Systems I have, it's easy to see why GPU crunchers reign supreme....even over 16 CPU Machines.

My quick findings....(most likely common knowledge already as well)

System #1 - Q6600 Quad @ 3.2Ghz - 10 GFLOPS/core (40 GFLOPS Total)

System #2 - Q6600 Quad @ STOCK/2.4Ghz - 6.5 GFLOPS/core (26 GFLOPS Total)

...

Doesn't quite reflect my experience. I ran a Q6600 (@3.2GHz) for quite a while before I replaced it with my current Q9450 which I am able to run at 3.6GHz. The performance of the two at the same clock speed is not massively different. Now my GTX295 is estimated by Boinc at 106GFlops per GPU and each GPU is roughly equivalent to 2 cores of my Q9450 (i.e. adding the GTX doubled my RAC + a little bit). So that puts the Q9450 @ 3.6GHz at about 53GFlops per core.

F.
ID: 898400 · Report as offensive
Profile TCP JESUS
Avatar

Send message
Joined: 19 Jan 03
Posts: 205
Credit: 1,248,845
RAC: 0
Canada
Message 898409 - Posted: 22 May 2009, 22:30:32 UTC - in response to Message 898400.  

....So that puts the Q9450 @ 3.6GHz at about 53GFlops per core.
F.


Have you run IntelBurnTest on that CPU ?

Was I interpreting the data that I received, wrong ?

Was the 26 GFLOPS output using IntelBurnTest a 'PER CORE' value on my Stock Clock/2.4Ghz Q6600 ?
(..should I NOT have devided the results by 4 ?)....

I am just wondering how I can get actual APPLES-TO-APPLES comparison, rather than estimating using times and WU results.

Thanks.
Allan
I am TCP JESUS...The Carpenter Phenom Jesus....and HAMMERING is what I do best!
formerly known as...MC Hammer.
ID: 898409 · Report as offensive
Fred W
Volunteer tester

Send message
Joined: 13 Jun 99
Posts: 2524
Credit: 11,954,210
RAC: 0
United Kingdom
Message 898428 - Posted: 22 May 2009, 23:21:54 UTC - in response to Message 898409.  



Have you run IntelBurnTest on that CPU ?

Was I interpreting the data that I received, wrong ?

Was the 26 GFLOPS output using IntelBurnTest a 'PER CORE' value on my Stock Clock/2.4Ghz Q6600 ?
(..should I NOT have devided the results by 4 ?)....

I am just wondering how I can get actual APPLES-TO-APPLES comparison, rather than estimating using times and WU results.

Thanks.
Allan

I hadn't come across the IntelBurnTest. Just googled it and looked at a couple of forums. It looks like your are interpreting the results right - perhaps it is the Boinc estimate for my GTX that is inaccurate. I'm not sure if I am prepared to toast my CPU to find out - perhaps after I have caught a few zzzzzzz's.

F.
ID: 898428 · Report as offensive
Profile jrusling
Avatar

Send message
Joined: 8 Sep 02
Posts: 37
Credit: 4,764,889
RAC: 0
United States
Message 898468 - Posted: 23 May 2009, 0:31:03 UTC - in response to Message 898153.  

5/22/2009 3:52:25 PM CUDA device: GeForce 9600 GT (driver version 18585, compute capability 1.1, 512MB, est. 37GFLOPS)

http://boincstats.com/signature/-1/user/18390/sig.png
ID: 898468 · Report as offensive
Profile KW2E
Avatar

Send message
Joined: 18 May 99
Posts: 346
Credit: 104,396,190
RAC: 34
United States
Message 898471 - Posted: 23 May 2009, 0:34:26 UTC - in response to Message 898468.  

Want to know GFLOPS for Intel CPUs?

Check this.
ID: 898471 · Report as offensive
OzzFan Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Apr 02
Posts: 15691
Credit: 84,761,841
RAC: 28
United States
Message 898535 - Posted: 23 May 2009, 3:24:10 UTC - in response to Message 898471.  

Want to know GFLOPS for Intel CPUs?

Check this.


They don't list any Xeon CPUs like my E5430s. I suppose it would be closest to the Core 2 Quad Q9450 CPU, but I have two physical chips.
ID: 898535 · Report as offensive
Profile TCP JESUS
Avatar

Send message
Joined: 19 Jan 03
Posts: 205
Credit: 1,248,845
RAC: 0
Canada
Message 898552 - Posted: 23 May 2009, 5:46:19 UTC - in response to Message 898471.  

Want to know GFLOPS for Intel CPUs?

Check this.


I have seen that link before and although it's directly from Intel, it's different than what the Linpack tests spit out for a Q6600 @ stock speeds.

IntelCPUburn/Linpak Q6600 @ 2.4Ghz = 26.xx GFLOPS

Intel Website Specs Q6600 @ 2.4Ghz = 38.xx GFLOPS


I would just really like to know the method used to calculate by Intel, so that a TRUE Video Card vs CPU comparison can be done.

It's too bad that BOINC didn't just spit out a CPU MFLOPS estimate right below the Video card/Cuda device estimate during startup so that we could compare on a level playing field.

Allan
I am TCP JESUS...The Carpenter Phenom Jesus....and HAMMERING is what I do best!
formerly known as...MC Hammer.
ID: 898552 · Report as offensive
Fred W
Volunteer tester

Send message
Joined: 13 Jun 99
Posts: 2524
Credit: 11,954,210
RAC: 0
United Kingdom
Message 898761 - Posted: 23 May 2009, 20:58:26 UTC - in response to Message 898409.  

....So that puts the Q9450 @ 3.6GHz at about 53GFlops per core.
F.


Have you run IntelBurnTest on that CPU ?


OK. I decided to take the plunge and here is what IntelBurn Test says about my Q9450 (stable at both standard and maximum stress levels - phew!!)

----------------------------
IntelBurnTest v2.1 : Standard Stress Level
Created by AgentGOD
----------------------------

Processor: Intel(R) Core(TM)2 Quad  CPU   Q9450  @ 2.66GHz
Clock Speed: 3.60 GHz
Active Physical Cores: 4
Total System Memory: 4094 MB

Testing started on 23/05/2009 21:25:06
Time (s)		Speed (GFlops)		Result
[21:25:31] 18.645	47.9430			3.068952e-002
[21:25:55] 18.548	48.1918			3.068952e-002
[21:26:20] 18.503	48.3093			3.068952e-002
[21:26:44] 18.457	48.4311			3.068952e-002
[21:27:08] 18.543	48.2057			3.068952e-002
Testing ended on 23/05/2009 21:27:08
----------------------------


----------------------------
IntelBurnTest v2.1 : Maximum Stress Level
Created by AgentGOD
----------------------------

Processor: Intel(R) Core(TM)2 Quad  CPU   Q9450  @ 2.66GHz
Clock Speed: 3.60 GHz
Active Physical Cores: 4
Total System Memory: 4094 MB

Testing started on 23/05/2009 21:28:55
Time (s)		Speed (GFlops)		Result
[21:30:22] 72.133	47.0287			3.134522e-002
[21:31:50] 74.157	45.7451			3.134522e-002
[21:33:17] 74.070	45.7985			3.134522e-002
[21:34:44] 72.996	46.4721			3.134522e-002
[21:36:10] 72.025	47.0988			3.134522e-002
Testing ended on 23/05/2009 21:36:10
----------------------------


And the maximum core temp was 73C for a moment; mainly sat around 67C - 70C. It normally sits around 60C when crunching.

So now I wonder about Boinc's estimates for the GFlops of the GPUs. I can only surmise that the wonderful work that the Lunatics crew has done optimising the code for the CPU makes the difference and that if the GPU code were optimised to the same degree my GPU could do two and a half times as much work as my CPU (on each GPU core?) giving a RAC of about 28K (or 48k if the 106 GFlops for the GTX really is per GPU core).

F.
ID: 898761 · Report as offensive
Uffe F

Send message
Joined: 7 Sep 04
Posts: 2
Credit: 135,337
RAC: 0
Denmark
Message 906208 - Posted: 11 Jun 2009, 15:41:24 UTC - in response to Message 898468.  

5/22/2009 3:52:25 PM CUDA device: GeForce 9600 GT (driver version 18585, compute capability 1.1, 512MB, est. 37GFLOPS)


How do you get it to 37 GFLOPS? any tweaking?

mine says:
11-06-2009 15:27:09 CUDA device: GeForce 9600M GT (driver version 18585, compute capability 1.1, 512MB, est. 14GFLOPS)
ID: 906208 · Report as offensive
Previous · 1 · 2 · 3 · Next

Questions and Answers : GPU applications : How fast are different cards?


 
©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.