Top GPU models

Message boards : Number crunching : Top GPU models
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1147120 - Posted: 30 Aug 2011, 8:01:36 UTC
Last modified: 30 Aug 2011, 8:20:14 UTC

boincadm/DA has done a couple of changesets: [trac]#24069[/trac], [trac]#24071[/trac] and [trac]#24072[/trac] that shows the Top GPU models here at Seti, (the link is beneath Statistics on the front page),

I assume this is with the Stock GPU apps, as Linux, Mac and ATI/AMD are unfilled.

Claggy
ID: 1147120 · Report as offensive
Profile AI4FR
Avatar

Send message
Joined: 13 Apr 11
Posts: 57
Credit: 23,590,991
RAC: 0
United States
Message 1147126 - Posted: 30 Aug 2011, 8:27:16 UTC - in response to Message 1147120.  

Interesting, thanks for the link. Some thing doesn't look right in the line-up though.



http://www.AI4FR.com
ID: 1147126 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1147132 - Posted: 30 Aug 2011, 8:44:47 UTC - in response to Message 1147120.  

We also need to work with David to help him refine the list into something more useful.

The stated thinking behind the list is:

I added a page that shows the most productive GPU types,
total and by platform:

This will at least provide suggestions for which type
of GPU board to buy.

-- David

It looks as if the data is being pulled from recently returned results, so it should be up-to-date. But so far (and it's very early days yet) the list would be positively misleading as a buyer's guide: for a single card, no way can 9600GT>9800GT>9800GTX+ be right.

I presume the scores must be strongly influenced by the number of each type of card in use here: the 9800GT was pretty much 'most bang for the buck' three years ago, and there'll be lots of them around, but that doesn't make it a 'best buy' in 2011.

Do we have anyone reading the board today who could analyse gpu_list.php, and work out exactly what's being shown in the lists? My guess is that it's the raw total credits for results in the database for each card type - and I'm wondering if it might be for people running stock apps only.

My first suggestion, given space constraints, would be to add the count of cards in brackets after the model name, and to sort the list by average (total/count). But I'm sure between us we can help make it even better than that. Any ideas?

ID: 1147132 · Report as offensive
Blake Bonkofsky
Volunteer tester
Avatar

Send message
Joined: 29 Dec 99
Posts: 617
Credit: 46,383,149
RAC: 0
United States
Message 1147133 - Posted: 30 Aug 2011, 8:45:19 UTC - in response to Message 1147126.  

Is this based on computation potential or purely the number of cards that are present on the project? Surely it can't be based on processing power, listing a 9600GT over a GTX580...

Perhaps it is just total contribution based on card type?
ID: 1147133 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1147138 - Posted: 30 Aug 2011, 9:03:55 UTC - in response to Message 1147132.  
Last modified: 30 Aug 2011, 9:13:52 UTC

...Do we have anyone reading the board today who could analyse gpu_list.php, and work out exactly what's being shown in the lists? My guess is that it's the raw total credits for results in the database for each card type - and I'm wondering if it might be for people running stock apps only...


A quick look suggest it's counting number of results in the last day for each model. I'll look again to check when my eyes decide to cooperate better with the PhP code.

[Edit:] I would suggest, if that's so, that it reflects a compound value of raw performance x number in circulation of each model, rather than performance alone. They would need to divide the figures by some complicated figure based on active hosts with the given model. Unfortunately Boinc doesn't identify mixed card setups properly, but if a count of the cards per machine is available it might provide something 'near enough'

Jason
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1147138 · Report as offensive
Highlander
Avatar

Send message
Joined: 5 Oct 99
Posts: 167
Credit: 37,987,668
RAC: 16
Germany
Message 1147143 - Posted: 30 Aug 2011, 9:31:52 UTC - in response to Message 1147138.  

A quick look suggest it's counting number of results in the last day for each model. I'll look again to check when my eyes decide to cooperate better with the PhP code.

[Edit:] I would suggest, if that's so, that it reflects a compound value of raw performance x number in circulation of each model, rather than performance alone. They would need to divide the figures by some complicated figure based on active hosts with the given model. Unfortunately Boinc doesn't identify mixed card setups properly, but if a count of the cards per machine is available it might provide something 'near enough'
Jason


and i vote for an overview of incoming results of the last 30 days, but like jason said, thats near enough
- Performance is not a simple linear function of the number of CPUs you throw at the problem. -
ID: 1147143 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1147269 - Posted: 30 Aug 2011, 21:42:44 UTC

David has changed the sort order - does that look better, everyone?
ID: 1147269 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1147270 - Posted: 30 Aug 2011, 21:48:01 UTC - in response to Message 1147269.  
Last modified: 30 Aug 2011, 21:50:04 UTC

David has changed the sort order - does that look better, everyone?

Yep, that looks a bit more reasonable, some figures would be nice, and what happened to the 8400 GS? has it dropped off the bottom?

Claggy
ID: 1147270 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1147277 - Posted: 30 Aug 2011, 22:14:48 UTC - in response to Message 1147270.  

David has changed the sort order - does that look better, everyone?

Yep, that looks a bit more reasonable, some figures would be nice, and what happened to the 8400 GS? has it dropped off the bottom?

Claggy

The code in [trac]changeset:24085[/trac] seems to consider the first 5,000 results it encounters in a list of tasks created in the last 30 days. With SETI turning over 50,000 tasks an hour, that's a pretty sparse sample - my guess is that the 8400 GS just hasn't reported in the last 5,000.

It's still not right - although David (brownie point for this one) has skipped tasks running for less than 100 seconds, basically the -9s - he's done nothing to account for shorty storms. I'd prefer to see a page like list of recently connected client types, with some explanations and cautions. And like that list, I'd like to see the query run periodically, and the results cached - the time it takes to refresh the 'Top GPU models' page suggests the code hits the servers pretty hard. Caching the page would help, though hopefully with a refresh interval of less than 10 months. Remind me to try and find a way of celebrating the anniversary of the current list ;-)
ID: 1147277 · Report as offensive
Profile Sutaru Tsureku
Volunteer tester

Send message
Joined: 6 Apr 07
Posts: 7105
Credit: 147,663,825
RAC: 5
Germany
Message 1147300 - Posted: 30 Aug 2011, 23:54:53 UTC
Last modified: 30 Aug 2011, 23:58:55 UTC

What about double chip grafic cards: 9800GX2, GTX295, GTX590?

As I looked GTX295 #13, GTX260 #11.

Maybe x2 or /2 (depend of how the place is calculated), so the double chip grafic cards are higher placed.


1. GeForce GTX 570
2. GeForce GTX 580
3. GeForce GTX 480
4. GeForce GTX 470
5. GeForce GTX 560 Ti
6. GeForce GTX 465
7. GeForce GTX 460
8. GeForce GTX 280
9. GeForce GTX 275
10. GeForce GTX 285
11. GeForce GTX 260
12. GeForce GTS 450
13. GeForce GTX 295
14. GeForce 8800 GTX
15. GeForce 9800 GTX/9800 GTX+
16. GeForce GTX 560
17. GeForce GTX 460 SE
18. GeForce 9800 GTX+
19. GeForce GTS 250
20. GeForce 8800 GTS 512
21. GeForce 9800 GT
22. GeForce 8800 GT
23. GeForce 8800 GTS
24. GeForce GT 440
25. GeForce GT 240
26. GeForce GT 220
27. GeForce GT 430
28. GeForce GT 330M
29. GeForce 9600 GT
30. GeForce 8600 GT
31. GeForce 9500 GT
32. GeForce 210
33. GeForce 9400 GT



- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC. - SETI@home needs your help. -
ID: 1147300 · Report as offensive
EdwardPF
Volunteer tester

Send message
Joined: 26 Jul 99
Posts: 389
Credit: 236,772,605
RAC: 374
United States
Message 1147338 - Posted: 31 Aug 2011, 3:05:48 UTC - in response to Message 1147300.  

where is my GeForce GTX 550 Ti ??

Ed F
ID: 1147338 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19012
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1147348 - Posted: 31 Aug 2011, 3:46:06 UTC

With so few tasks sampled and so many gpu's, and with the processing of mid range AR's and VHAR's havin at least a 30% variation. This exersize seems pointless and misleading, especially if it takes up a lot of server power.

It is nice to see that -9's etc can be disregarded, now maybe Dr A. can apply it where it is most needed, calculating APR.
ID: 1147348 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1147423 - Posted: 31 Aug 2011, 9:15:38 UTC
Last modified: 31 Aug 2011, 9:16:27 UTC

This is what the Top GPU models list looks like now:

1.GeForce GTX 590
2.GeForce GTX 570
3.GeForce GTX 480
4.GeForce GTX 580
5.GeForce GTX 470
6.GeForce GTX 465
7.GeForce GTX 560 Ti
8.GeForce GTX 460
9.GeForce GTX 285
10.GeForce GTX 275
11.GeForce GTX 260
12.GeForce GTX 295
13.GeForce GTS 450
14.GeForce 8800 GTX
15.GeForce 8800 GTS 512
16.GeForce 8800 GT
17.GeForce GTS 250
18.GeForce 9800 GT
19.GeForce GT 240
20.GeForce GT 430
21.GeForce 8800 GTS
22.GeForce GT 220
23.GeForce 9600 GT
24.GeForce GT 330M
25.Quadro FX 580
26.GeForce 9500 GT
27.GeForce 9400 GT

some GPU have now made the list, while others have vanished like the GTX 9800/GTX 9800+, wouldn't it be better if it was a running average of app times, ie update the list, don't make a new list.

Claggy
ID: 1147423 · Report as offensive
Profile BMH
Avatar

Send message
Joined: 27 May 99
Posts: 419
Credit: 166,294,083
RAC: 125
United Kingdom
Message 1147568 - Posted: 31 Aug 2011, 18:05:09 UTC

A performance graph would be nice ;)

For example I'm using both GTX 460s and 560ti's, but there is quite a price jump, I wonder if the performance increase reflects this - I'm still trying to find out from my own tests.

Useful information though, I'm not sure how it compares to other performance comparisons which I tend to refer to when buying new stuff... also referring to power consumption and price.
Brian.
ID: 1147568 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1147654 - Posted: 31 Aug 2011, 20:59:05 UTC
Last modified: 31 Aug 2011, 20:59:38 UTC

The Top GPU models list now has some numbers, not sure what they mean:

1.(1.000) GeForce GTX 570
2.(0.839) GeForce GTX 480
3.(0.779) GeForce GTX 560 Ti
4.(0.628) GeForce GTX 470
5.(0.569) GeForce GTX 460
6.(0.507) GeForce GTX 275
7.(0.493) GeForce GTX 260
8.(0.475) GeForce GTX 550 Ti
9.(0.432) GeForce GTX 295
10.(0.423) GeForce GTS 450
11.(0.317) GeForce GTX 460M
12.(0.312) GeForce 8800 GTX
13.(0.298) GeForce GTS 240
14.(0.285) GeForce 8800 GT
15.(0.284) GeForce GT 240
16.(0.268) GeForce GTS 250
17.(0.227) GeForce 9800 GT
18.(0.223) Quadro FX 3800
19.(0.220) GeForce GT 330
20.(0.208) GeForce 8800 GTS
21.(0.106) Quadro FX 2700M
22.(0.100) GeForce 9500 GT
23.(0.087) GeForce 8600 GT
24.(0.058) GeForce 9400 GT
25.(0.056) GeForce 210

Claggy
ID: 1147654 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1147663 - Posted: 31 Aug 2011, 21:12:28 UTC - in response to Message 1147654.  

The Top GPU models list now has some numbers, not sure what they mean:

They're relative performance, presumably what fraction of the current 'top dog' each card is achieving.

I added a new web page, html/user/gpu_list.php

This shows a list of the top-performing GPUs for your project,
broken down by vendor and platform. For example:
http://setiathome.berkeley.edu/gpu_list.php

The performance figures shown on this page are not
the manufacturer's peak-FLOPS claims;
they are the actual performance for your projects' applications,
based on job runtime info from your database.

Projects with GPU apps should consider adding this to your site
(to do so, just upgrade to the latest web code).
Your users will be interested,
and it may influence their GPU-buying decisions to your benefit.

-- David

(from the 'BOINC Projects' mailing list, for those who don't monitor it regularly)
ID: 1147663 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1147676 - Posted: 31 Aug 2011, 21:33:17 UTC - in response to Message 1147663.  

Thanks for that, i hadn't noticed the 'Relative speeds are shown in parentheses.' entry,

Claggy
ID: 1147676 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1147677 - Posted: 31 Aug 2011, 21:37:10 UTC - in response to Message 1147676.  

Thanks for that, i hadn't noticed the 'Relative speeds are shown in parentheses.' entry,

It may not have been there at first - I got it from an off-list email. He may still be tuning the code - it hasn't been committed into trac yet.
ID: 1147677 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1147692 - Posted: 31 Aug 2011, 22:10:49 UTC

This should help answer those "best video card bang for the buck" questions.

I wonder how big of a mess it would be to do this for CPUs as well.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1147692 · Report as offensive
Profile Khangollo
Avatar

Send message
Joined: 1 Aug 00
Posts: 245
Credit: 36,410,524
RAC: 0
Slovenia
Message 1147696 - Posted: 31 Aug 2011, 22:15:28 UTC

So where has GTX 285 gone, now?
ID: 1147696 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : Top GPU models


 
©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.