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.

1 · 2 · 3 · Next

AuthorMessage
depsini

Send message
Joined: 2 Feb 09
Posts: 5
Credit: 2,166
RAC: 0
Norway
Message 884695 - Posted: 12 Apr 2009, 19:12:19 UTC

This might have been up before and surely there are lists etc but heck i ask anyways:

Is there a big difference in speed, when running CUDA, between for example
Nvidia Geforce
-8600M
-9500
-9600
-9800
-260


Which card would be the best to buy, as to value?
ID: 884695 · 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 884710 - Posted: 12 Apr 2009, 20:19:20 UTC

Speed of the cards varies greatly across the nVidia range. Generally, it is not a good idea to run BOINC on laptops and mobile graphics card because of the heat produced.
The FAQs on GPUGrid have a list of the GFLOPS that each nVidia card gives, the bigger the number the better. However, currently I am unable to access their site. It is pretty easy to work out the speeds tho for the 9000 series, they go up in order so 9400GT, 9500GT, 9600GT etc. The GTS 250 is roughly equal to the 9800GTX+ and of course the more expensive cards, the GTX models are even faster, with the GTX 295 being the fastest.

ID: 884710 · Report as offensive
slozomby

Send message
Joined: 16 Nov 04
Posts: 20
Credit: 242,588
RAC: 0
United States
Message 884783 - Posted: 13 Apr 2009, 2:00:01 UTC - in response to Message 884695.  

my 3 cards average time per seti wu
8400gs 2.5 hours
8800gt 18 minutes
gts250(9800gtx+) 15 minutes

i hear the gtx280's are around 6 minutes.

i'd say the gtx260-216 core is the best bang for the buck in nvidia cards currently.


ID: 884783 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 65709
Credit: 55,293,173
RAC: 49
United States
Message 884872 - Posted: 13 Apr 2009, 6:37:01 UTC

My GTX295 does a Seti WU in about 8.5-9.5 minutes.
The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's
ID: 884872 · Report as offensive
Profile Fred J. Verster
Volunteer tester
Avatar

Send message
Joined: 21 Apr 04
Posts: 3252
Credit: 31,903,643
RAC: 0
Netherlands
Message 885097 - Posted: 13 Apr 2009, 21:00:05 UTC - in response to Message 884872.  
Last modified: 13 Apr 2009, 21:02:41 UTC

Hi, I have a GT8500 card, which I tested on GPUgrid, till I found out, that 1 of their WU's, took 124 hours to complete, just 4 hours after the dead-line, without it running High Priority?

This type of card, is too slow run a similar prog/WU, faster then the average Q6600(OC'ed or Q6700;QX6800 or 9000 series, like Q9450;9550 and Q9650. (my QX9650(@3.73), does 3.8GFLOPs a core)
I'am gonna replace 8500GT with 9800GTX, I, still find it a less pleasant, the VIDEO RESPONCE & QUALITY on my main driver :( , but . . . , IMHO, a 9800GTX is the MINIMUM, for acceptable crunching, (in more then 1 project) and is not to expensive too.

In EUROPE THESE CARDs ARE INCOMPATIBLE EXPENSIVE !!? As a lot of other (electronic) stuff.
Harddisks are about the same price, as are PSU's.
CPU? , what do you pay in the U.S. now a days, for a Q9650 a pay €~300,00 (€1090,00 for a QX9650!)

[A BIT OFF TOPIC :)]
ID: 885097 · Report as offensive
depsini

Send message
Joined: 2 Feb 09
Posts: 5
Credit: 2,166
RAC: 0
Norway
Message 885112 - Posted: 13 Apr 2009, 22:08:16 UTC

Tx a lot!

Nice post Fred :S
ID: 885112 · Report as offensive
slozomby

Send message
Joined: 16 Nov 04
Posts: 20
Credit: 242,588
RAC: 0
United States
Message 885139 - Posted: 13 Apr 2009, 23:00:51 UTC - in response to Message 885097.  

Hi, I have a GT8500 card, which I tested on GPUgrid, till I found out, that 1 of their WU's, took 124 hours to complete, just 4 hours after the dead-line, without it running High Priority?



GPUGrid is pretty open with the fact that an 8500gt is not recommended to use there. the 8500gt has 16 stream processors. there is only 1 cuda capable card with a lower stream processor count. in fact, they state in thier faq that anything under 64 stream processors is NOT recommended.

high priority wont make a the card go any faster. only not using the video card for anything other than gpu crunching will allow that card to complete in under the 5 day deadline.

regardless that card will do about 2.5 hours per non-vlar seti mb wu.

think video cards are expensive in europe. try having to pay for your health insurance in the US.


ID: 885139 · 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 885281 - Posted: 14 Apr 2009, 9:24:27 UTC

Good point!
GPUGrid is back up and running after their network failure, you can look at the graphic card comparison here.
ID: 885281 · Report as offensive
Sitarow
Avatar

Send message
Joined: 1 Oct 99
Posts: 4
Credit: 2,294,290
RAC: 0
Canada
Message 885906 - Posted: 16 Apr 2009, 21:16:12 UTC - in response to Message 884872.  
Last modified: 16 Apr 2009, 21:17:43 UTC

My GTX295 does a Seti WU in about 8.5-9.5 minutes.


I have a EVGA GTX 285 and it does a Seti Cuda WU in about 3.5 - 4mins.

Using BOINC 6.6.20 and Drivers 182.50 On a i7 920 not OC'ed Vista Enterprise 64 SP1
ID: 885906 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 65709
Credit: 55,293,173
RAC: 49
United States
Message 885912 - Posted: 16 Apr 2009, 21:51:44 UTC - in response to Message 885906.  
Last modified: 16 Apr 2009, 21:55:01 UTC

My GTX295 does a Seti WU in about 8.5-9.5 minutes.


I have a EVGA GTX 285 and it does a Seti Cuda WU in about 3.5 - 4mins.

Using BOINC 6.6.20 and Drivers 182.50 On a i7 920 not OC'ed Vista Enterprise 64 SP1

Well that 8.5-9.5 is now 9.5-10.5 and that's per gpu and I have two of them, You have one, Unless You've got another of those hungry beasts laying around there somewhere of course.

295=2x448bit path.
285=1x512bit path.
The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's
ID: 885912 · Report as offensive
Fred W
Volunteer tester

Send message
Joined: 13 Jun 99
Posts: 2524
Credit: 11,954,210
RAC: 0
United Kingdom
Message 885928 - Posted: 16 Apr 2009, 22:56:56 UTC - in response to Message 885906.  

My GTX295 does a Seti WU in about 8.5-9.5 minutes.


I have a EVGA GTX 285 and it does a Seti Cuda WU in about 3.5 - 4mins.

Using BOINC 6.6.20 and Drivers 182.50 On a i7 920 not OC'ed Vista Enterprise 64 SP1

Let's be sure we're comparing apples with apples here; the 295 knocks out 12 - 14 credit WU's in 3.5 - 4 mins.

F.
ID: 885928 · Report as offensive
Profile Codeman05

Send message
Joined: 16 Dec 01
Posts: 33
Credit: 15,457,430
RAC: 0
United States
Message 885971 - Posted: 17 Apr 2009, 0:58:25 UTC - in response to Message 885912.  

My GTX295 does a Seti WU in about 8.5-9.5 minutes.


I have a EVGA GTX 285 and it does a Seti Cuda WU in about 3.5 - 4mins.

Using BOINC 6.6.20 and Drivers 182.50 On a i7 920 not OC'ed Vista Enterprise 64 SP1

Well that 8.5-9.5 is now 9.5-10.5 and that's per gpu and I have two of them, You have one, Unless You've got another of those hungry beasts laying around there somewhere of course.

295=2x448bit path.
285=1x512bit path.


hmm, well I have 3 GTX285's ...also on a i7 rig w/ 6.6.20 and I'm seeing anywhere from 7mins per WU per GPU....to 16min per WU per GPU. I am running the optimizations though the times were roughly the same before.

I was thinking this was due to differences between WU's (angle etc...), but now you guys have me wondering.
ID: 885971 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 65709
Credit: 55,293,173
RAC: 49
United States
Message 886003 - Posted: 17 Apr 2009, 3:26:29 UTC - in response to Message 885971.  
Last modified: 17 Apr 2009, 3:27:30 UTC

My GTX295 does a Seti WU in about 8.5-9.5 minutes.


I have a EVGA GTX 285 and it does a Seti Cuda WU in about 3.5 - 4mins.

Using BOINC 6.6.20 and Drivers 182.50 On a i7 920 not OC'ed Vista Enterprise 64 SP1

Well that 8.5-9.5 is now 9.5-10.5 and that's per gpu and I have two of them, You have one, Unless You've got another of those hungry beasts laying around there somewhere of course.

295=2x448bit path.
285=1x512bit path.


hmm, well I have 3 GTX285's ...also on a i7 rig w/ 6.6.20 and I'm seeing anywhere from 7mins per WU per GPU....to 16min per WU per GPU. I am running the optimizations though the times were roughly the same before.

I was thinking this was due to differences between WU's (angle etc...), but now you guys have me wondering.

Recently I've seen as high as 17:06(min:sec) and as low as 4:55(min:sec) from the 295, So It can vary depending on how fast the card is, how warm or cool, wide a bit path, etc.
The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's
ID: 886003 · Report as offensive
HFB1217
Avatar

Send message
Joined: 25 Dec 05
Posts: 102
Credit: 9,424,572
RAC: 0
United States
Message 888180 - Posted: 25 Apr 2009, 5:46:52 UTC

My EVGA 260/216 cards are overclocked to a Core Speed of 650 and memory speed of 2300. They are finishing a workunit from 6 to 9 minutes depending on the complexity of the WU.
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: 888180 · Report as offensive
(Ryle)

Send message
Joined: 18 Aug 05
Posts: 3
Credit: 1,916,218
RAC: 0
Denmark
Message 888211 - Posted: 25 Apr 2009, 10:45:29 UTC

I just got a GTX275, but so far its 40 minutes in and 30% done.

I suspect it has to do with Vista64 powersaving features, but i've already set it to High Performance.

I have the 182.50 64-bit drivers.

Is it because i have a Asus p5b-deluxe mainboard, that doesnt support pci-e 2.0?

Boinc reports cuda version 1.3, but shouldn't it be 2.0?

Before that I had a 8800GT that also took 1,5 hours per workunit. Is there something i can do? Is it the Vista aero interface?

Later i will upgrade the mainboard, but it could be nice with a fast cruncher until then ;)

Boinc is version 6.6.20.
ID: 888211 · Report as offensive
Profile Gundolf Jahn

Send message
Joined: 19 Sep 00
Posts: 3184
Credit: 446,358
RAC: 0
Germany
Message 888213 - Posted: 25 Apr 2009, 10:50:21 UTC - in response to Message 888211.  
Last modified: 25 Apr 2009, 10:50:36 UTC

And your tasks do run on the GPU (no fallback to CPU)?
ID: 888213 · Report as offensive
(Ryle)

Send message
Joined: 18 Aug 05
Posts: 3
Credit: 1,916,218
RAC: 0
Denmark
Message 888214 - Posted: 25 Apr 2009, 11:04:58 UTC - in response to Message 888213.  

Yes, I believe so, Gundolf.

The CPU is running 2 Astropulse units (one on each core; its a core2duo E6600). The GPU runs a (cuda) unit.
ID: 888214 · Report as offensive
Fred W
Volunteer tester

Send message
Joined: 13 Jun 99
Posts: 2524
Credit: 11,954,210
RAC: 0
United Kingdom
Message 888217 - Posted: 25 Apr 2009, 11:33:29 UTC

You could have been unlucky and just hit a VLAR unit. These take over an hour on my GTX295 where most WU's take 6 - 12 minutes.

F.
ID: 888217 · Report as offensive
Profile Gundolf Jahn

Send message
Joined: 19 Sep 00
Posts: 3184
Credit: 446,358
RAC: 0
Germany
Message 888239 - Posted: 25 Apr 2009, 13:03:52 UTC - in response to Message 888214.  

And you could try to free one core for feeding the GPU. See Getting maximum credits.

Gruß,
Gundolf
ID: 888239 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 65709
Credit: 55,293,173
RAC: 49
United States
Message 888262 - Posted: 25 Apr 2009, 14:50:59 UTC - in response to Message 888217.  

You could have been unlucky and just hit a VLAR unit. These take over an hour on my GTX295 where most WU's take 6 - 12 minutes.

F.

I'll 2nd that one.
The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's
ID: 888262 · Report as offensive
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.