| Author |
Message |
Scott Brown Volunteer moderator Project administrator Volunteer tester Project scientist
 Send message
Joined: 17 Oct 05 Posts: 2535 ID: 1178 Credit: 29,423,992,796 RAC: 9,014,121
                                                               
|
|
I thought it would be informative to have a fairly comprehensive updated measure of how GPUs perform at PG. The below table was created by examining the PPS Sieve times in October 2011 for machines from the first 2000 of the top computers list and supplemented with various GPUs that are mine/I have access to. For those GPUs from the first 2000 of the top computers list, the times were taken from the highest ranked computer with a single GPU install that had completed PPS Sieve units. Thus, it is likely that many (if not most cards) have factory or user overclocking. Thus, these times in many cases represent the best possible times for these cards. Obviously, your times may vary somewhat based on 1) your shader clock/overclock, 2) other loads on the GPU and CPU, 3) whether the GPU is a single or multiple GPU install and the corresponding matched CPU, and 4) other things not herein mentioned. Where needed/possible, I have placed appropriate notes in the table (e.g., when it is one of my GPUs and I know the overclock). Anyone with a GPU not listed is welcome to add their data...Indeed, individuals with dual-GPU NVidia devices (GTX 590, GTX 295, 9800 GX2) are especially encouraged to add data as such GPUs are more difficult to verify in the top computers list since they are always listed as being multiple GPU systems.
GPU....................................................................................Minutes
NVIDIA GeForce GTX 580 (1471MB) driver: 27533...............................................9
NVIDIA GeForce GTX 570 (1248MB) driver: 26677..............................................10
NVIDIA GeForce GTX 560 Ti (961MB) driver: 27550............................................15
NVIDIA GeForce GTX 560 (1023MB) driver: 28026..............................................20
NVIDIA GeForce GTX 550 Ti (992MB) driver: 28026............................................30
NVIDIA GeForce GT 545 (1471MB) driver: 27533...............................................53
NVIDIA GeForce GT 530 (993MB) driver: 26716................................................81
NVIDIA GeForce GT 520 (993MB) driver: 26742...............................................140
NVIDIA GeForce GTX 480 (1535MB) driver: 27061..............................................10
NVIDIA GeForce GTX 470 (1216MB) driver: 27533..............................................15
NVIDIA GeForce GTX 465 (993MB) driver: 28026...............................................18
NVIDIA GeForce GTX 460 (768MB) driver: 26658...............................................21
NVIDIA GeForce GTX 460 SE (993MB) driver: 28026............................................29
NVIDIA GeForce GTS 450 (1023MB) driver: 26658..............................................30
NVIDIA GeForce GT 440-OEM (1503MB) driver: 28026...........................................59
NVIDIA GeForce GT 440 (993MB) driver: 28026................................................71
NVIDIA GeForce GT 430 (961MB) driver: 27533................................................81
NVIDIA GeForce GT 420 (2014MB) driver: 28026..............................................165
**NOTE: GTX 460 factory OC 1600 shader clock**
**NOTE: GTS 450 factory OC 1900 shader clock**
**NOTE: GT 440-OEM has 144 shaders with 1300 clock OC**
NVIDIA GeForce GT 330 (947MB) driver: 27533................................................95
NVIDIA GeForce GTX 275 (873MB) driver: 28026...............................................34
NVIDIA GeForce GTX 285 (966MB) driver: 27533...............................................35
NVIDIA GeForce GTX 260-216 (881MB) driver: 26658...........................................38
NVIDIA GeForce GTX 260 (848MB) driver: 27533...............................................48
NVIDIA GeForce GTS 250 (1023MB)............................................................58
NVIDIA GeForce GTS 240 (997MB) driver: 26658...............................................66
NVIDIA GeForce GT 240 (512MB) driver: 19107................................................95
NVIDIA GeForce GT 230 (1508MB) driver: 25936..............................................102
NVIDIA GeForce GT 220 (947MB) driver: 27533...............................................188
NVIDIA GeForce GT 120 (511MB).............................................................270
NVIDIA GeForce 9800 GTX (495MB) driver: 25896..............................................55
NVIDIA GeForce 9800 GT (1005MB) driver: 28026..............................................58
NVIDIA GeForce 9600 GSO (367MB) driver: 28026..............................................73
NVIDIA GeForce 9600 GT (997MB) driver: 26658..............................................116
NVIDIA GeForce 9500 GT (489MB) driver: 28026..............................................212
**NOTE: 9600 GSO & 9500GT with factory OC 1750 shader clock**
NVIDIA GeForce 8800 GTS 512 (511MB) driver: 26658..........................................58
NVIDIA GeForce 8800 Ultra (731MB) driver: 25896............................................59
NVIDIA GeForce 8800 GTX (700MB) driver: 27061..............................................69
NVIDIA GeForce 8800 GT (511MB) driver: 27061...............................................70
NVIDIA GeForce 8800 GS (362MB) driver: 25896...............................................91
NVIDIA GeForce 8800 GTS (615MB) driver: 19745.............................................107
NVIDIA GeForce 8400 GS (255MB) driver: 25896..............................................799
**NOTE: 8800GS with 1500 shader clock OC**
NVIDIA Tesla C2050 (2635MB) driver: 27565..................................................12
NVIDIA Tesla C2075 (4095MB) driver: 27589..................................................16
NVIDIA Tesla M2050 (2687MB)................................................................17
NVIDIA Tesla C1060 (4058MB) driver: 25896..................................................39
NVIDIA Quadro 6000 (4095MB) driver: 27589..................................................16
NVIDIA Quadro 5000 (2204MB) driver: 26687..................................................22
NVIDIA Quadro 4000 (1982MB) driver: 27071..................................................33
NVIDIA Quadro 2000 (1023MB) driver: 27589..................................................45
NVIDIA Quadro FX 4800 (1477MB) driver: 27589...............................................49
NVIDIA Quadro FX 3800 (1024MB) driver: 19100...............................................55
NVIDIA Quadro FX 5600 (1475MB) driver: 27071...............................................69
NVIDIA Quadro FX 3700 (511MB) driver: 26635................................................85
NVIDIA Quadro 600 (961MB) driver: 27533....................................................88
NVIDIA Quadro FX 4600 (700MB) driver: 27589...............................................114
NVIDIA Quadro FX 1800 (745MB) driver: 25957...............................................132
NVIDIA GeForce GTX 460M (1503MB) driver: 25912.............................................41
NVIDIA GeForce GTX 285M (997MB) driver: 28026..............................................62
NVIDIA GeForce GT 540M (993MB).............................................................72
NVIDIA GeForce 9800M GTX (1007MB) driver: 26099............................................76
NVIDIA GeForce GT 550M (2014MB)............................................................79
NVIDIA GeForce GT 435M (2014MB) driver: 25951..............................................84
NVIDIA GeForce GT 525M (993MB).............................................................96
NVIDIA GeForce GT 425M (993MB) driver: 25942...............................................96
NVIDIA GeForce GT 520M (993MB)............................................................155
NVIDIA Quadro FX 2700M (511MB) driver: 19805..............................................191
NVIDIA Quadro FX 880M (986MB) driver: 25922...............................................210
NVIDIA GeForce 8400M GS (243MB) driver: 25896.............................................923
CAL ATI Radeon HD 5800 series (Cypress) (1024MB) driver: 1.4.1546..........................27
CAL AMD Radeon HD 6900 series (Cayman) (2048MB) driver: 1.4.1457...........................37
CAL AMD Radeon HD 6800 series (Barts) (1024MB) driver: 1.4.1546............................41
CAL ATI Radeon HD 5700 series (Juniper) (1024MB) driver: 1.4.1353..........................51
CAL ATI Radeon HD 4700/4800 (RV740/RV770) (512MB) driver: 1.4.1457.........................76
CAL ATI Radeon HD 5x00 series (Redwood) (1024MB) driver: 1.4.1523..........................99
CAL ATI Radeon HD 4600 series (R730) (512MB) driver: 1.4.1546.............................106
CAL AMD Radeon HD 6x00 series (Turks) (1024MB) driver: 1.4.1523...........................120
CAL AMD Radeon HD 6300 series (Caicos) (1024MB) driver: 1.4.1546..........................378
CAL ATI Radeon HD 4550 (R710) (512MB) driver: 1.4.1523....................................728
From this list, the advantage of fermi-based cards should be clear. Likewise, the ATI cards are clearly hampered by the need to use OpenCL rather than native CAL instructions. Sorry that the list for ATI could not be more specific, but BOINC does not report anything more specific than the ATI families. You can probably safely assume that the times are for the fastest cards in those families.
One final note to remember regarding this list and card performance at PG. Currently there are two GPU applications in standard use at PG: PPS Sieve and CW Sieve. While these are similar in that they are Sieves and are based on the same algorithm (at least as I recall what Ken said about writing them), performance rankings on these may/will not be identical. I have observed some subtle differences in the cards I own in how they rank against one another in the CW sieve vs. what is in the table above. This is likely due to the drastically different ranges being sieved at the moment (I believe that PPS Sieve is in full on 32-bit mode while CW Sieve may still be at a 24-bit level).
Additionally, LLR Cuda is being tested. This new application requires the use of double-precision calculations. That means that many of the cards above will simply not work with that application. Also, the current advantage of the best GTX 5xx and 4xx series over Tesla and Quadro cards (which is due entirely to much higher stock shader clocks) will likely disappear and be reversed since the Tesla and (I believe newer) Quadro cards have better DP capabilities (i.e., fermi-based GeForce cards have 1/4 DP power compared to their single-precision capabilities while DP and SP power on Tesla cards is the same).
____________
141941*2^4299438-1 is prime!
|
|
|
|
|
|
EVGA GTX460FTW 1GB/factory 850MHz/260.99/18:05
EVGA GTX460SC 768MB/factory 763MHz - now 835MHz/260.99/19:05
These are the fastest times, sometimes it will take more than 30 secs more.
A older driver will produce more PPD but in that case I can't use Afterburner to lower my voltages : FTW from 1.075 > .975 and SC from .950 > .887.
Crunchers without the need to adjust their voltages could try 258.96
____________
|
|
|
|
|
|
I have a nVidia 9500 GS on this client.
It takes 264 minutes to complete a PPS sieve workunit.
____________
My Top 5000 Primes:
110059!+1 is prime! (FPS)
14493618614235*2^666666-1 is prime! (SGS) |
|
|
|
|
|
Pallit GTX 460 Sonic Platinum ( Factory OC @ 800 MHz ) takes 20:36, and I'm running a pair in an X58 based system.
HTH
dunx |
|
|
|
|
|
Hi,
With my NVIDIA Quadro FX 580 (511MB) driver: 27533... it's 329 minutes.
Good crunch @ all.
____________
MySpace al@ON =8?()> |
|
|
|
|
|
Hi All,
HD5x00 (Cedar) 512Mb, driver 1.4.1546, takes an average of about 33000s or 550 min, or just over 9 hours.
____________
Member team AUSTRALIA
My lucky number is 9291*2^1085585+1 |
|
|
mackerel Volunteer tester
 Send message
Joined: 2 Oct 08 Posts: 2948 ID: 29980 Credit: 773,786,746 RAC: 206,139
                                      
|
|
HD6850 45m 775/1000 11.11
GTS450 32m 888/1000/1776 285.58
9500GT 260m 550/400/1400 285.58
Clocks are what GPU-Z reports. I think it doubles one of the values compared to those listed elsewhere but I can't remember which one.
Edit: just for comparison, I put one unit on my CPU (i7-2600k stock) and the estimated time was 15 hours! That was manually calculated not BOINC estimates. No surprise I aborted it... |
|
|
mackerel Volunteer tester
 Send message
Joined: 2 Oct 08 Posts: 2948 ID: 29980 Credit: 773,786,746 RAC: 206,139
                                      
|
|
GTX560Ti 16m 900/1002/1800 285.62 |
|
|
|
|
|
Hmmm, is this normal ?
My HD 6970 takes around 45 minutes for the new units but also up to 85% of an CPU-core... ???
Other machine with HD 5770 takes around 110 minutes and not more than 8.5% of an CPU-core.
Is this a Boinc-problem, a Driver-problem of a HD 6970 problem ??
____________
Greetz
Chris
561*2^1423021+1
|
|
|
Scott Brown Volunteer moderator Project administrator Volunteer tester Project scientist
 Send message
Joined: 17 Oct 05 Posts: 2535 ID: 1178 Credit: 29,423,992,796 RAC: 9,014,121
                                                               
|
Hmmm, is this normal ?
My HD 6970 takes around 45 minutes for the new units but also up to 85% of an CPU-core... ???
Other machine with HD 5770 takes around 110 minutes and not more than 8.5% of an CPU-core.
Is this a Boinc-problem, a Driver-problem of a HD 6970 problem ??
It is probably a driver issue as this behavior has been reported under newer driver versions. Since your 5770 appears to be running the older version, I would revert the 6970 back to that version and see if it solves the issue.
____________
141941*2^4299438-1 is prime!
|
|
|
|
|
|
My HD 6970 is taking 48 minutes and using .86 cpu as well. Stock speeds using 11.12. Let us know if you see any difference with other driver versions. |
|
|
DoES Volunteer tester
 Send message
Joined: 11 Oct 08 Posts: 784 ID: 30382 Credit: 75,064,140 RAC: 0
             
|
|
4 x 560GTX reporting 32min to 26min -- each using 0.16 CPU
1 x 9800GTX+ reporting 1hr 15min -- using 0.16 CPU
____________
Member of AtP
Shown here is an Australian native rat (Ratus Kickarsus) |
|
|
|
|
|
GTX570 980s with factory settings and 880s with an overclock. I expect it'd overclock considerably more.
____________
Oh Bondage? Up Yours.
http://www.youtube.com/watch?v=ogypBUCb7DA
|
|
|
|
|
|
GTX560Ti 448 core ... 1082s @ factory clocks (797,1950,1594)
____________
There's someone in our head but it's not us. |
|
|
|
|
Hmmm, is this normal ?
My HD 6970 takes around 45 minutes for the new units but also up to 85% of an CPU-core... ???
Other machine with HD 5770 takes around 110 minutes and not more than 8.5% of an CPU-core.
Is this a Boinc-problem, a Driver-problem of a HD 6970 problem ??
It's a known issue - at the very end of
this pdf about the new 2.6 APP SDK, it mentions that :
When using Catalyst 11.7 drivers, some OpenCL applications use 100% of the CPU time.
AMD aims to fix this issue in a future release of the Catalyst driver. Please upgrade your
Catalyst driver as new releases become available.
I'm using Catalyst 11.12 (driver 8.920.0.0) on Win7 32-bit, and I've got an overclocked 6670, and everything seems fine. There are newer/experimental drivers out there, so trying those is a possibile alternative to downgrading. I had a quick look at this page, but I've got no idea about the reliability of the site, and you know what pre-release or "experimental stuff can be like...
Just some info and ideas. |
|
|
|
|
|
Any chance of updating this list with the NVidia 6xx series and AMD 7xxx series GPUs? |
|
|
Scott Brown Volunteer moderator Project administrator Volunteer tester Project scientist
 Send message
Joined: 17 Oct 05 Posts: 2535 ID: 1178 Credit: 29,423,992,796 RAC: 9,014,121
                                                               
|
Any chance of updating this list with the NVidia 6xx series and AMD 7xxx series GPUs?
I'll see what I can do over the next few days. I only have direct access to a couple of the NVidia 6xx cards, so it will take me a bit of time to go through the leader board where I can see times for various cards.
____________
141941*2^4299438-1 is prime!
|
|
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3358 ID: 50683 Credit: 213,711,583 RAC: 1,519,383
                                
|
|
GT 640 clocked at 1.1GHZ ~ 45 minutes per WU ( PPS sieve)
____________
2*836^798431+1 CRUS PRIME
92*10^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
2022202116^131072+1 GFN
Proud member of team Aggie The Pew. Go Aggie! |
|
|
|
|
|
GTX 560 TiX 17:59 @ stock clocks.
EDIT: Current w/u completed moments ago.
____________
There's someone in our head but it's not us. |
|
|
|
|
|
Is it possible that these are the runtimes with the shorter 6G WUs instead of the 9G WUs we are currently crunching? Since my GTX470 can never do a WU in 15 min, even with OC. But with 1.5x the runtimes they are pretty accurate.
Here my times and some from our (DPC) forum members:
GTX470 @608 MHz (stock) 21:40
GTX470 @700 MHz (OC) 18:25
GTX480 @700 MHz (stock) 17:12
GTX580 (stock) 14:45 - (Arjant2)
GTX570 (stock) 16:39 - (Amberjr)
GTX460 EVGA FTW (Factory OC) 32:21 - (MAB The Frisian)
7970 @1100 MHz (OC) ~28min - (Booster)
Edit: They must be with the shorter WUs, the GTX460 FTW from MAB The Frisian is the same as he listed in the second message in this topic. |
|
|
Crun-chi Volunteer tester
 Send message
Joined: 25 Nov 09 Posts: 3358 ID: 50683 Credit: 213,711,583 RAC: 1,519,383
                                
|
|
My runtime is one day old, so I think it is time for 9G WU.
____________
2*836^798431+1 CRUS PRIME
92*10^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
2022202116^131072+1 GFN
Proud member of team Aggie The Pew. Go Aggie! |
|
|
|
|
|
Around the PPS sieve challenge in 2011 they changed the length of the sieves and raised the credits accordingly. The WUs we are crunching are definitely the longer ones, but the times in the opening post are probably for the shorter ones from before the PPS sieve challenge in 2011. |
|
|
ReZSend message
Joined: 10 Jan 11 Posts: 13 ID: 80789 Credit: 29,259,432 RAC: 0
         
|
|
Could this be updated?
as the PPS sieves take much more time now and there are a few cards added ^^
Titan anyone?
(or am I just living under a rock and totally missed the new list? ) |
|
|
|
|
(or am I just living under a rock and totally missed the new list? )
you probably are but then so am I.
They're all overclocked but I can't remember the exact speeds and can't be bothered to check right now other than 560 is at 900MHz.
Times are rounded off a bit:
560: 1,500s
570: 900s
580: 830s |
|
|
Davina   Send message
Joined: 13 Feb 12 Posts: 3453 ID: 130544 Credit: 2,827,486,251 RAC: 386,371
                                
|
|
Stock 580 c.870 secs (860 @ 783MHz). I've seen someone doing <770 secs. Seen a Titan at 530 secs. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 14644 ID: 53948 Credit: 1,017,870,200 RAC: 639,317
                                      
|
Stock 580 c.870 secs (860 @ 783MHz). I've seen someone doing <770 secs. Seen a Titan at 530 secs.
Just be REALLY careful when looking at OTHER people's results. Lots more people have multiple GPUS today than two years ago, and BOINC doesn't care about switching GPUs in the middle of a run when it restarts a task. If you're looking at timings, you have to read the whole stderr.txt on the result to make sure the entire task ran on one GPU. Otherwise, the time could be a combination of, say, a TITAN and a GT 8600, which isn't going to give you a useful number.
Even if the host display says it has only 1 GPU, that's what it has NOW, and not necessarily what it had when the result was run.
____________
My lucky number is 75898524288+1 |
|
|
ReZSend message
Joined: 10 Jan 11 Posts: 13 ID: 80789 Credit: 29,259,432 RAC: 0
         
|
|
just from the numbers above a titan seems about a third faster than a 580....
2 generations later and a price difference of 2-3x the 580 when it was new retail.. and just 30% faster... :/
4x the cuda cores, higher clock... am I expecting too much? |
|
|
ReZSend message
Joined: 10 Jan 11 Posts: 13 ID: 80789 Credit: 29,259,432 RAC: 0
         
|
|
just for the heck of it: pps sieve @ under 8 minutes... |
|
|
Davina   Send message
Joined: 13 Feb 12 Posts: 3453 ID: 130544 Credit: 2,827,486,251 RAC: 386,371
                                
|
just for the heck of it: pps sieve @ under 8 minutes...
In the blink of a proverbial you'll be Jade then! Race you to double-bronze before Xmas... |
|
|