PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise

Advanced search

Message boards : Number crunching : GTX970 / GTX980 - nVidia Maxwell

Author Message
Profile [AF>Amis des Lapins] Phil1966Project donor
Send message
Joined: 20 Jul 13
Posts: 117
ID: 242019
Credit: 442,628,776
RAC: 131,533
Found 3 primes in the 2019 Tour de PrimesFound 4 primes in the 2021 Tour de Primes321 LLR Bronze: Earned 10,000 credits (29,564)Cullen LLR Bronze: Earned 10,000 credits (36,084)ESP LLR Bronze: Earned 10,000 credits (15,196)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (512,981)PPS LLR Turquoise: Earned 5,000,000 credits (5,875,793)PSP LLR Bronze: Earned 10,000 credits (48,561)SoB LLR Bronze: Earned 10,000 credits (62,033)SR5 LLR Gold: Earned 500,000 credits (631,249)SGS LLR Amethyst: Earned 1,000,000 credits (1,459,279)TRP LLR Gold: Earned 500,000 credits (536,929)Woodall LLR Bronze: Earned 10,000 credits (54,727)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,116,301)Generalized Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (3,629,679)PPS Sieve Double Silver: Earned 200,000,000 credits (347,880,458)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Turquoise: Earned 5,000,000 credits (7,661,419)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,495,590)AP 26/27 Sapphire: Earned 20,000,000 credits (30,674,241)WW Ruby: Earned 2,000,000 credits (2,216,000)GFN Sapphire: Earned 20,000,000 credits (29,770,630)PSA Turquoise: Earned 5,000,000 credits (6,923,280)
Message 79911 - Posted: 2 Oct 2014 | 17:25:50 UTC

Hello !

this short message to ask you if the new Maxwell nVidia cards (GTX970 & GTX980) can already work on your projects + PRPNet.

I am asking because none appears in the "fastest GPU" stats.

Thank You

Kind Regards

eXtreme Warhead
Send message
Joined: 19 Nov 12
Posts: 4
ID: 180420
Credit: 157,774,293
RAC: 85
321 LLR Bronze: Earned 10,000 credits (20,021)PPS LLR Bronze: Earned 10,000 credits (55,457)SGS LLR Bronze: Earned 10,000 credits (11,016)Woodall LLR Silver: Earned 100,000 credits (105,960)Generalized Cullen/Woodall Sieve (suspended) Bronze: Earned 10,000 credits (77,003)PPS Sieve Double Bronze: Earned 100,000,000 credits (157,081,858)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Silver: Earned 100,000 credits (106,403)TRP Sieve (suspended) Silver: Earned 100,000 credits (316,574)
Message 79918 - Posted: 2 Oct 2014 | 19:03:52 UTC - in response to Message 79911.
Last modified: 2 Oct 2014 | 19:05:35 UTC

it works, but horrible. because: with a 660ti i had a wu done in about 9min, two parallel in about 16min

with a 970 i get the same project wu done in about 10,5min which is so horrible embarrassing for the project, which should run here minimal 50% better than the 660ti and still is worse...i can only hope for an update...

ps: forgotten the current project which was tested: Proth Prime Search Sieve (PPS-Sieve)

HonzaProject donor
Volunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1902
ID: 352
Credit: 3,599,461,531
RAC: 5,389,503
Discovered 8 mega primesEliminated 4 conjecture "k"sFound 2 primes in the 2018 Tour de PrimesFound 1 prime in the 2018 Tour de Primes Mountain Stage2019 Tour de Primes largest primeFound 4 primes in the 2019 Tour de PrimesFound 1 mega prime in the 2019 Tour de PrimesFound 1 prime in the 2019 Tour de Primes Mountain StageFound 1 prime in the 2020 Tour de PrimesFound 4 primes in the 2021 Tour de PrimesFound 1 mega prime in the 2021 Tour de PrimesFound 1 prime in the 2021 Tour de Primes Mountain Stage321 LLR Double Bronze: Earned 100,000,000 credits (100,425,701)Cullen LLR Emerald: Earned 50,000,000 credits (96,735,787)ESP LLR Emerald: Earned 50,000,000 credits (50,853,190)Generalized Cullen/Woodall LLR Emerald: Earned 50,000,000 credits (50,309,119)PPS LLR Double Bronze: Earned 100,000,000 credits (155,424,135)PSP LLR Emerald: Earned 50,000,000 credits (50,563,867)SoB LLR Double Bronze: Earned 100,000,000 credits (113,220,736)SR5 LLR Double Bronze: Earned 100,000,000 credits (101,883,529)SGS LLR Emerald: Earned 50,000,000 credits (55,573,609)TPS LLR (retired) Bronze: Earned 10,000 credits (43,033)TRP LLR Double Bronze: Earned 100,000,000 credits (100,338,924)Woodall LLR Emerald: Earned 50,000,000 credits (50,956,094)321 Sieve (suspended) Double Bronze: Earned 100,000,000 credits (115,948,450)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,142,109)Generalized Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (50,504,945)PPS Sieve Double Gold: Earned 500,000,000 credits (513,057,580)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,288,222)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,149,354)AP 26/27 Double Silver: Earned 200,000,000 credits (232,724,773)WW Double Silver: Earned 200,000,000 credits (404,064,000)GFN Double Gold: Earned 500,000,000 credits (777,188,770)PSA Double Gold: Earned 500,000,000 credits (535,232,795)
Message 79928 - Posted: 3 Oct 2014 | 6:33:58 UTC - in response to Message 79918.

with a 970 i get the same project wu done in about 10,5min which is so horrible embarrassing for the project, which should run here minimal 50% better than the 660ti and still is worse...i can only hope for an update...

...which is horrible for nVidia.

PPS Sieve needs a bit of double precision computing.
GTX 970 gives embbarrasing 109 GFLOPs (1/32 of single precision), which is about
760Ti that gives 103 GLOPs (1/24 of single precision)

How do you expect software update would solve that?

(HD7970 back from 2011/2012 gives ~950 GFLOPs of double precision)

____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Profile [AF>Amis des Lapins] Phil1966Project donor
Send message
Joined: 20 Jul 13
Posts: 117
ID: 242019
Credit: 442,628,776
RAC: 131,533
Found 3 primes in the 2019 Tour de PrimesFound 4 primes in the 2021 Tour de Primes321 LLR Bronze: Earned 10,000 credits (29,564)Cullen LLR Bronze: Earned 10,000 credits (36,084)ESP LLR Bronze: Earned 10,000 credits (15,196)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (512,981)PPS LLR Turquoise: Earned 5,000,000 credits (5,875,793)PSP LLR Bronze: Earned 10,000 credits (48,561)SoB LLR Bronze: Earned 10,000 credits (62,033)SR5 LLR Gold: Earned 500,000 credits (631,249)SGS LLR Amethyst: Earned 1,000,000 credits (1,459,279)TRP LLR Gold: Earned 500,000 credits (536,929)Woodall LLR Bronze: Earned 10,000 credits (54,727)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,116,301)Generalized Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (3,629,679)PPS Sieve Double Silver: Earned 200,000,000 credits (347,880,458)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Turquoise: Earned 5,000,000 credits (7,661,419)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,495,590)AP 26/27 Sapphire: Earned 20,000,000 credits (30,674,241)WW Ruby: Earned 2,000,000 credits (2,216,000)GFN Sapphire: Earned 20,000,000 credits (29,770,630)PSA Turquoise: Earned 5,000,000 credits (6,923,280)
Message 79929 - Posted: 3 Oct 2014 | 7:25:37 UTC - in response to Message 79928.
Last modified: 3 Oct 2014 | 7:56:40 UTC

Dear Honza,

I fully understand eXtreme Warhead's frustration.

My "old" GTX770 performed MUCH BETTER on PPS Sieve than my actual GTX780Ti.

850 seconds for 2 WU's in parallel, while the 780Ti takes 500 seconds for 1 WU or 1200 for 2 WU's in parallel !

My 660Ti = 610 secs for 1 WU and 1200 secs for 2 WU's = not bad at all.

Bought 2 * GTX970 Gigabyte G1, but if not efficient on PPS Sieve, will run on GPUGRID or Collatz now that these cards do support OpenCL 1.2 (opencl for nVidia + CUDA apps available)

And vice versa.

Bought GTX970 because they need less power than 780Ti, and my crunchbox runs 24/7.


You mention that

PPS Sieve needs a bit of double precision computing.
:

a) I was not aware of this particularity

b) Have run PPS Sieve on HD 7950Boost, and calculation times are pathetic, even with powerful DP computing capability.


Should one understand from your post that PPS Sieve is designed to work "at best" on TITAN / TITAN BLACK only ?


By the way (this is an open question), do you intend to propose an "opencl for nVidia" app for PPS Sieve, in order to benefit (if any benefit ?) from OpenCL 1.2 (GTX970/980) ?


Or do you intend to implement the same kind of system as GPUGRID ? (but without excluding the "old cards" from the game as they do :/ )

ie :


* If you have a GM204 you will get a CUDA65 application, or nothing if your driver is too old

* If you have a GM107 you will get a CUDA60 application, or nothing if your driver is too old

* If you have a Fermi or Kepler you will get a CUDA60 application, or CUDA42 if your driver is too old

* If you have a Fermi card you will get a CUDA60 application if your driver is new enough, or a CUDA42 application if not.


Thank You

Kind regards

Philippe

HonzaProject donor
Volunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1902
ID: 352
Credit: 3,599,461,531
RAC: 5,389,503
Discovered 8 mega primesEliminated 4 conjecture "k"sFound 2 primes in the 2018 Tour de PrimesFound 1 prime in the 2018 Tour de Primes Mountain Stage2019 Tour de Primes largest primeFound 4 primes in the 2019 Tour de PrimesFound 1 mega prime in the 2019 Tour de PrimesFound 1 prime in the 2019 Tour de Primes Mountain StageFound 1 prime in the 2020 Tour de PrimesFound 4 primes in the 2021 Tour de PrimesFound 1 mega prime in the 2021 Tour de PrimesFound 1 prime in the 2021 Tour de Primes Mountain Stage321 LLR Double Bronze: Earned 100,000,000 credits (100,425,701)Cullen LLR Emerald: Earned 50,000,000 credits (96,735,787)ESP LLR Emerald: Earned 50,000,000 credits (50,853,190)Generalized Cullen/Woodall LLR Emerald: Earned 50,000,000 credits (50,309,119)PPS LLR Double Bronze: Earned 100,000,000 credits (155,424,135)PSP LLR Emerald: Earned 50,000,000 credits (50,563,867)SoB LLR Double Bronze: Earned 100,000,000 credits (113,220,736)SR5 LLR Double Bronze: Earned 100,000,000 credits (101,883,529)SGS LLR Emerald: Earned 50,000,000 credits (55,573,609)TPS LLR (retired) Bronze: Earned 10,000 credits (43,033)TRP LLR Double Bronze: Earned 100,000,000 credits (100,338,924)Woodall LLR Emerald: Earned 50,000,000 credits (50,956,094)321 Sieve (suspended) Double Bronze: Earned 100,000,000 credits (115,948,450)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,142,109)Generalized Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (50,504,945)PPS Sieve Double Gold: Earned 500,000,000 credits (513,057,580)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,288,222)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,149,354)AP 26/27 Double Silver: Earned 200,000,000 credits (232,724,773)WW Double Silver: Earned 200,000,000 credits (404,064,000)GFN Double Gold: Earned 500,000,000 credits (777,188,770)PSA Double Gold: Earned 500,000,000 credits (535,232,795)
Message 79931 - Posted: 3 Oct 2014 | 8:22:56 UTC

Philippe, I can't answer your questions about future developement of PPS Sieve app, I'll leave it to those who knows better/best.

From a project point of view, our effort is to run *any* app as effecient as possible.
See latest developement in LLR (64-bit, AVX, FMA3) or Genefer.

It is not anyone's intention to prefer one GPU over another (or CPU in that matter).
For example, AMD has not done the best job when implementing FMA instructions...it's not LLR app fault.

It is possible to fine-tune PPS Sieve app (or Genefer or wwwwCL) for latest GTX 9xx series. Or perhaps there is not much that can be done, it may reveal that 256-bit memory bus width is not helping, or poor DP performance...or both and evem more.
This would need hands on both of developers and testers, it is still very new.

My "old" GTX770 performed MUCH BETTER on PPS Sieve than my actual GTX780Ti.

See Fastest GPUs, 780Ti should do better.

Should one understand from your post that PPS Sieve is designed to work "at best" on TITAN / TITAN BLACK only ?

I would says other way round - TITAN is designed to do best in scientific math, ie. double precision. It is expensive for a reason.

My "old" GTX770 performed MUCH BETTER on PPS Sieve than my actual GTX780Ti.

Yeah, and my old GTX580 is also doing very well on Genefer WR, on par with 780Ti and leaving all other 7xx and 6xx behind.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Profile [AF>Amis des Lapins] Phil1966Project donor
Send message
Joined: 20 Jul 13
Posts: 117
ID: 242019
Credit: 442,628,776
RAC: 131,533
Found 3 primes in the 2019 Tour de PrimesFound 4 primes in the 2021 Tour de Primes321 LLR Bronze: Earned 10,000 credits (29,564)Cullen LLR Bronze: Earned 10,000 credits (36,084)ESP LLR Bronze: Earned 10,000 credits (15,196)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (512,981)PPS LLR Turquoise: Earned 5,000,000 credits (5,875,793)PSP LLR Bronze: Earned 10,000 credits (48,561)SoB LLR Bronze: Earned 10,000 credits (62,033)SR5 LLR Gold: Earned 500,000 credits (631,249)SGS LLR Amethyst: Earned 1,000,000 credits (1,459,279)TRP LLR Gold: Earned 500,000 credits (536,929)Woodall LLR Bronze: Earned 10,000 credits (54,727)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,116,301)Generalized Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (3,629,679)PPS Sieve Double Silver: Earned 200,000,000 credits (347,880,458)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Turquoise: Earned 5,000,000 credits (7,661,419)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,495,590)AP 26/27 Sapphire: Earned 20,000,000 credits (30,674,241)WW Ruby: Earned 2,000,000 credits (2,216,000)GFN Sapphire: Earned 20,000,000 credits (29,770,630)PSA Turquoise: Earned 5,000,000 credits (6,923,280)
Message 79932 - Posted: 3 Oct 2014 | 8:59:44 UTC - in response to Message 79931.
Last modified: 3 Oct 2014 | 9:00:45 UTC

Dear Honza,

The "fastest GPU list" is INACCURATE, as it is not only taking into consideration the "time to run WU's", but also the relative numbers of WU's processed by type of cards !

Otherwhise, the TITAN would not be ranked #17 !


(1.000) GeForce GTX TITAN Black
(0.951) GeForce GTX 780 Ti
(0.943) Quadro K6000
(0.757) GeForce GTX 670
(0.731) GeForce GTX 770
(0.727) Tesla K20m
(0.705) GeForce GTX 680
(0.682) GeForce GTX 880M
(0.676) GeForce GTX 870M
(0.651) GeForce GTX 690
(0.648) GeForce GTX 660 Ti
(0.596) GeForce GTX 760
(0.587) GeForce GTX 760 (192-bit)
(0.572) GeForce GTX 780
(0.548) GRID K520
(0.546) GeForce GTX 860M
(0.546) GeForce GTX TITAN



But OK, don't want to insist, as I like this project very much.

Looking forward to receive my GTX970 G1 and test them on PPS Sieve,

Best Regards,

Philippe

HonzaProject donor
Volunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1902
ID: 352
Credit: 3,599,461,531
RAC: 5,389,503
Discovered 8 mega primesEliminated 4 conjecture "k"sFound 2 primes in the 2018 Tour de PrimesFound 1 prime in the 2018 Tour de Primes Mountain Stage2019 Tour de Primes largest primeFound 4 primes in the 2019 Tour de PrimesFound 1 mega prime in the 2019 Tour de PrimesFound 1 prime in the 2019 Tour de Primes Mountain StageFound 1 prime in the 2020 Tour de PrimesFound 4 primes in the 2021 Tour de PrimesFound 1 mega prime in the 2021 Tour de PrimesFound 1 prime in the 2021 Tour de Primes Mountain Stage321 LLR Double Bronze: Earned 100,000,000 credits (100,425,701)Cullen LLR Emerald: Earned 50,000,000 credits (96,735,787)ESP LLR Emerald: Earned 50,000,000 credits (50,853,190)Generalized Cullen/Woodall LLR Emerald: Earned 50,000,000 credits (50,309,119)PPS LLR Double Bronze: Earned 100,000,000 credits (155,424,135)PSP LLR Emerald: Earned 50,000,000 credits (50,563,867)SoB LLR Double Bronze: Earned 100,000,000 credits (113,220,736)SR5 LLR Double Bronze: Earned 100,000,000 credits (101,883,529)SGS LLR Emerald: Earned 50,000,000 credits (55,573,609)TPS LLR (retired) Bronze: Earned 10,000 credits (43,033)TRP LLR Double Bronze: Earned 100,000,000 credits (100,338,924)Woodall LLR Emerald: Earned 50,000,000 credits (50,956,094)321 Sieve (suspended) Double Bronze: Earned 100,000,000 credits (115,948,450)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,142,109)Generalized Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (50,504,945)PPS Sieve Double Gold: Earned 500,000,000 credits (513,057,580)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,288,222)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,149,354)AP 26/27 Double Silver: Earned 200,000,000 credits (232,724,773)WW Double Silver: Earned 200,000,000 credits (404,064,000)GFN Double Gold: Earned 500,000,000 credits (777,188,770)PSA Double Gold: Earned 500,000,000 credits (535,232,795)
Message 79933 - Posted: 3 Oct 2014 | 9:04:28 UTC - in response to Message 79932.

Looking forward to receive my GTX970 G1 and test them on PPS Sieve,



Philippe,
would be great if you can do some benchmarks on Genefer and WWWWCL as well. I'm sure there are more participants curious about it's performance.

Yes, could be inaccurate - especialy where only a few pieces (like Titans) goes into stats.

____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Profile [AF>Amis des Lapins] Phil1966Project donor
Send message
Joined: 20 Jul 13
Posts: 117
ID: 242019
Credit: 442,628,776
RAC: 131,533
Found 3 primes in the 2019 Tour de PrimesFound 4 primes in the 2021 Tour de Primes321 LLR Bronze: Earned 10,000 credits (29,564)Cullen LLR Bronze: Earned 10,000 credits (36,084)ESP LLR Bronze: Earned 10,000 credits (15,196)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (512,981)PPS LLR Turquoise: Earned 5,000,000 credits (5,875,793)PSP LLR Bronze: Earned 10,000 credits (48,561)SoB LLR Bronze: Earned 10,000 credits (62,033)SR5 LLR Gold: Earned 500,000 credits (631,249)SGS LLR Amethyst: Earned 1,000,000 credits (1,459,279)TRP LLR Gold: Earned 500,000 credits (536,929)Woodall LLR Bronze: Earned 10,000 credits (54,727)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,116,301)Generalized Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (3,629,679)PPS Sieve Double Silver: Earned 200,000,000 credits (347,880,458)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Turquoise: Earned 5,000,000 credits (7,661,419)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,495,590)AP 26/27 Sapphire: Earned 20,000,000 credits (30,674,241)WW Ruby: Earned 2,000,000 credits (2,216,000)GFN Sapphire: Earned 20,000,000 credits (29,770,630)PSA Turquoise: Earned 5,000,000 credits (6,923,280)
Message 79934 - Posted: 3 Oct 2014 | 9:18:42 UTC - in response to Message 79933.

Hello Honza,

Concerning Genefer .... have some doubt as the DP is very very low on Maxwells ... ?

Does OpenCL 1.2 capability help ?

On PRPNet I will try to run some WSS during the next challenge.

Need to find the accurate settings for WSS, as last time I ran WSS / WFS, it was with HD7950.

Will see.

Hope to get them tomorrow or on Monday.

Scott BrownProject donor
Volunteer moderator
Project administrator
Volunteer tester
Project scientist
Avatar
Send message
Joined: 17 Oct 05
Posts: 2248
ID: 1178
Credit: 9,810,783,138
RAC: 13,705,804
Discovered the World's First base 116 Generalized Cullen prime!!!Discovered 28 mega primesEliminated 7 conjecture "k"sDiscovered 1 Sophie Germain pairDiscovered 2 Fermat divisors2012 Tour de Primes highest prime count2012 Tour de Primes most Mountain Stage primes2015 Tour de Primes highest prime count2016 Tour de Primes highest prime countFound 23 primes in the 2018 Tour de PrimesFound 1 mega prime in the 2018 Tour de PrimesFound 2 primes in the 2018 Tour de Primes Mountain Stage2019 Tour de Primes highest prime countFound 22 primes in the 2019 Tour de Primes2020 Tour de Primes highest prime scoreFound 21 primes in the 2020 Tour de PrimesFound 4 mega primes in the 2020 Tour de PrimesFound 8 primes in the 2021 Tour de PrimesFound 1 mega prime in the 2021 Tour de Primes321 LLR Double Silver: Earned 200,000,000 credits (201,403,028)Cullen LLR Double Silver: Earned 200,000,000 credits (201,409,561)ESP LLR Double Silver: Earned 200,000,000 credits (203,249,784)Generalized Cullen/Woodall LLR Double Bronze: Earned 100,000,000 credits (109,580,172)PPS LLR Double Gold: Earned 500,000,000 credits (728,284,255)PSP LLR Double Bronze: Earned 100,000,000 credits (126,982,721)SoB LLR Double Bronze: Earned 100,000,000 credits (189,060,638)SR5 LLR Double Silver: Earned 200,000,000 credits (214,437,634)SGS LLR Double Silver: Earned 200,000,000 credits (202,538,622)TPS LLR (retired) Silver: Earned 100,000 credits (235,439)TRP LLR Double Silver: Earned 200,000,000 credits (201,215,056)Woodall LLR Double Bronze: Earned 100,000,000 credits (104,130,013)321 Sieve (suspended) Double Silver: Earned 200,000,000 credits (235,451,253)Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (83,794,448)Generalized Cullen/Woodall Sieve (suspended) Double Silver: Earned 200,000,000 credits (285,139,652)PPS Sieve Double Ruby: Earned 2,000,000,000 credits (2,728,939,994)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Double Silver: Earned 200,000,000 credits (203,523,358)TRP Sieve (suspended) Double Silver: Earned 200,000,000 credits (201,489,157)AP 26/27 Double Silver: Earned 200,000,000 credits (385,720,907)WW Double Gold: Earned 500,000,000 credits (896,876,000)GFN Double Ruby: Earned 2,000,000,000 credits (2,048,310,105)PSA Double Silver: Earned 200,000,000 credits (259,058,048)
Message 79935 - Posted: 3 Oct 2014 | 10:15:29 UTC - in response to Message 79934.


Concerning Genefer .... have some doubt as the DP is very very low on Maxwells ... ?



This has nothing to do with Genefer app. The DP performance of different families of NVidia GPUs is published as a ratio of that families single precision performance as follows (excluding Quadro, Tesla, and Titan cards that have different ratio with higher DP levels):

Fermi - 1/8
Kepler - 1/24
Maxwell - 1/32

OpenCL implementations have helped close the gap between Kepler and Fermi. My guess is that this will be ultimately similar with Maxwell, but that the drivers for these brand new cards are not yet mature enough (as was the case with many Kepler cards). I'd give it a few months before anything definitive can really be said about the performance comparisons of GTX 9xx cards.

As for the GTX 780 Ti issue, this may be a similar driver problem given that it uses a different Kepler chip than the GTX 770 and below cards that has a couple of newer features...but there should have been time to have these issues addressed by NVidia (and I believe that at least some of these cards do actually also perform less well on other BOINC projects).

eXaPower
Send message
Joined: 30 Sep 13
Posts: 122
ID: 259902
Credit: 1,636,922,447
RAC: 260,510
Discovered 2 mega primesFound 1 prime in the 2018 Tour de PrimesFound 1 mega prime in the 2018 Tour de PrimesFound 1 prime in the 2020 Tour de Primes321 LLR Amethyst: Earned 1,000,000 credits (1,031,618)Cullen LLR Gold: Earned 500,000 credits (620,013)ESP LLR Amethyst: Earned 1,000,000 credits (1,043,069)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (585,120)PPS LLR Jade: Earned 10,000,000 credits (10,589,974)PSP LLR Gold: Earned 500,000 credits (506,876)SoB LLR Silver: Earned 100,000 credits (379,643)SR5 LLR Turquoise: Earned 5,000,000 credits (5,022,053)SGS LLR Amethyst: Earned 1,000,000 credits (1,006,021)TRP LLR Ruby: Earned 2,000,000 credits (2,403,013)Woodall LLR Gold: Earned 500,000 credits (509,918)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,035,186)Generalized Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,122,369)PPS Sieve Double Gold: Earned 500,000,000 credits (758,913,230)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Ruby: Earned 2,000,000 credits (2,058,245)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (3,185,685)AP 26/27 Jade: Earned 10,000,000 credits (10,192,403)WW Double Bronze: Earned 100,000,000 credits (173,424,000)GFN Double Gold: Earned 500,000,000 credits (658,294,252)
Message 79938 - Posted: 3 Oct 2014 | 10:35:55 UTC

Aren't Sieve Apps: integer? In any case- Maxwell C.C 5.0/5.2 SMM has 64 integer core compared to 32 a SMX for C.C 3.0 and 3.5 (GeForce 780[ti]) Titan's are supposed to be 64 integer a SMX, like C.C 3.5 Tesla and Quadro. (check CUDA programming Guide for Throughput of Native Arithmetic Instructions)

A Gtx 780ti-- total of 480 integer cores. A GTX980-- 1024 total integer cores.

I've seen GTX 750ti (320 integer) PPS sieve times near Kelper GTX 650ti with 128 integer cores. Maxwell cores are non-superscalar. Kelper cores are Superscalar, maybe this accounting for PPS sieves time differences.

An update to new CUDA 6.5 may straighten this out ?

Profile [AF>Amis des Lapins] Phil1966Project donor
Send message
Joined: 20 Jul 13
Posts: 117
ID: 242019
Credit: 442,628,776
RAC: 131,533
Found 3 primes in the 2019 Tour de PrimesFound 4 primes in the 2021 Tour de Primes321 LLR Bronze: Earned 10,000 credits (29,564)Cullen LLR Bronze: Earned 10,000 credits (36,084)ESP LLR Bronze: Earned 10,000 credits (15,196)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (512,981)PPS LLR Turquoise: Earned 5,000,000 credits (5,875,793)PSP LLR Bronze: Earned 10,000 credits (48,561)SoB LLR Bronze: Earned 10,000 credits (62,033)SR5 LLR Gold: Earned 500,000 credits (631,249)SGS LLR Amethyst: Earned 1,000,000 credits (1,459,279)TRP LLR Gold: Earned 500,000 credits (536,929)Woodall LLR Bronze: Earned 10,000 credits (54,727)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,116,301)Generalized Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (3,629,679)PPS Sieve Double Silver: Earned 200,000,000 credits (347,880,458)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Turquoise: Earned 5,000,000 credits (7,661,419)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,495,590)AP 26/27 Sapphire: Earned 20,000,000 credits (30,674,241)WW Ruby: Earned 2,000,000 credits (2,216,000)GFN Sapphire: Earned 20,000,000 credits (29,770,630)PSA Turquoise: Earned 5,000,000 credits (6,923,280)
Message 79940 - Posted: 3 Oct 2014 | 11:21:41 UTC - in response to Message 79935.
Last modified: 3 Oct 2014 | 11:41:05 UTC


Concerning Genefer .... have some doubt as the DP is very very low on Maxwells ... ?



This has nothing to do with Genefer app. The DP performance of different families of NVidia GPUs is published as a ratio of that families single precision performance as follows (excluding Quadro, Tesla, and Titan cards that have different ratio with higher DP levels):

Fermi - 1/8
Kepler - 1/24
Maxwell - 1/32



Exactly what I wanted to point out : given the poor DP computation performance of nVidia's, except TITAN + Prof. cards, I am not sure I want to try to run Genefer on a GTX970.

In the beginning (when I sold the 770s) I wanted to purchase 2 * TITAN BLACK, in order to have cards that can run > 1500/1700 GFLOPS in DP. But finaly decided to go for 2 * 970. Simply because it's cheaper to buy + cheaper to run + only 2 or 3 subprojects require DP capabilities.

+ I did not want ATI, although this brand still proposes high DP computing power.

Concerning all the technicals / performances aspects (software, hardware, apps, SP, DP, CUDA, OpenCL, ...), my opinion is only (poorly) built through my BOINC / end user experiments.

Have absolutely no IT knowledge / background.

=> I post here also to learn from you and the community.

=> If I was an IT engineer, I would probably understand why a system that is supposed to be stronger than another is finally performing poorly, without having to post here ;)

Scott BrownProject donor
Volunteer moderator
Project administrator
Volunteer tester
Project scientist
Avatar
Send message
Joined: 17 Oct 05
Posts: 2248
ID: 1178
Credit: 9,810,783,138
RAC: 13,705,804
Discovered the World's First base 116 Generalized Cullen prime!!!Discovered 28 mega primesEliminated 7 conjecture "k"sDiscovered 1 Sophie Germain pairDiscovered 2 Fermat divisors2012 Tour de Primes highest prime count2012 Tour de Primes most Mountain Stage primes2015 Tour de Primes highest prime count2016 Tour de Primes highest prime countFound 23 primes in the 2018 Tour de PrimesFound 1 mega prime in the 2018 Tour de PrimesFound 2 primes in the 2018 Tour de Primes Mountain Stage2019 Tour de Primes highest prime countFound 22 primes in the 2019 Tour de Primes2020 Tour de Primes highest prime scoreFound 21 primes in the 2020 Tour de PrimesFound 4 mega primes in the 2020 Tour de PrimesFound 8 primes in the 2021 Tour de PrimesFound 1 mega prime in the 2021 Tour de Primes321 LLR Double Silver: Earned 200,000,000 credits (201,403,028)Cullen LLR Double Silver: Earned 200,000,000 credits (201,409,561)ESP LLR Double Silver: Earned 200,000,000 credits (203,249,784)Generalized Cullen/Woodall LLR Double Bronze: Earned 100,000,000 credits (109,580,172)PPS LLR Double Gold: Earned 500,000,000 credits (728,284,255)PSP LLR Double Bronze: Earned 100,000,000 credits (126,982,721)SoB LLR Double Bronze: Earned 100,000,000 credits (189,060,638)SR5 LLR Double Silver: Earned 200,000,000 credits (214,437,634)SGS LLR Double Silver: Earned 200,000,000 credits (202,538,622)TPS LLR (retired) Silver: Earned 100,000 credits (235,439)TRP LLR Double Silver: Earned 200,000,000 credits (201,215,056)Woodall LLR Double Bronze: Earned 100,000,000 credits (104,130,013)321 Sieve (suspended) Double Silver: Earned 200,000,000 credits (235,451,253)Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (83,794,448)Generalized Cullen/Woodall Sieve (suspended) Double Silver: Earned 200,000,000 credits (285,139,652)PPS Sieve Double Ruby: Earned 2,000,000,000 credits (2,728,939,994)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Double Silver: Earned 200,000,000 credits (203,523,358)TRP Sieve (suspended) Double Silver: Earned 200,000,000 credits (201,489,157)AP 26/27 Double Silver: Earned 200,000,000 credits (385,720,907)WW Double Gold: Earned 500,000,000 credits (896,876,000)GFN Double Ruby: Earned 2,000,000,000 credits (2,048,310,105)PSA Double Silver: Earned 200,000,000 credits (259,058,048)
Message 79941 - Posted: 3 Oct 2014 | 12:42:01 UTC - in response to Message 79938.

Aren't Sieve Apps: integer?


Integer is extremely important for sieves, but other aspects of the GPU (and the system it is installed in) are also important to varying degrees. For example, memory is often important in these kinds of applications. Indeed, I suspect that less than optimal driver development with the advent of different memory structure and management on the newer GPU models--e.g., dynamic parallelism--may account for some of the under performance of CC 3.5, 5.0, & 5.2 cards (this would include all Maxwells and the GTX 780s and Titans).


A Gtx 780ti-- total of 480 integer cores. A GTX980-- 1024 total integer cores.


I am not sure where these numbers come from. A GTX 780Ti has 2880 shaders/cores and a GTX 980 has 2048 shaders/cores. Additionally, NVidia claims that 128 Maxwell cores have the equivalent performance of 90% of 192 Kepler cores. That would mean that the GTX 780Ti has the equivalent of 2592 Maxwell cores. If NVidia's claimed conversion is true, then this would mean that the 980/970 cards should perform very similarly to the 780/770 cards (and worse than the 780Ti or Titans) on the PPS sieve application.


I've seen GTX 750ti (320 integer) PPS sieve times near Kelper GTX 650ti with 128 integer cores. Maxwell cores are non-superscalar. Kelper cores are Superscalar, maybe this accounting for PPS sieves time differences.


Again, I don't know where these core numbers come from. The 650Ti has 768 shaders and the 750Ti has 640. Using the same Kepler to Maxwell conversion, that means the he 650Ti has the equivalent of about 690 Maxwell shaders, which matches the similar performance times you note.


An update to new CUDA 6.5 may straighten this out ?


Maybe, but I think more mature driver support will also be needed. It is also important to note that the 980/970 are second generation Maxwell cores compared to the 750/750Ti. As noted briefly on Wikipedia, "Second Generation Maxwell introduced a several new technologies: Dynamic Super Resolution, Third Generation Delta Color Compression, Multi-Pixel Programming Sampling, Nvidia VXGI (Real-Time-Voxel-Global Illumination), VR Direct, Multi-Projection Acceleration, and Multi-Frame Sampled Anti-Aliasing(MFAA), however, support for CSAA was removed. HDMI 2.0 support was also added." Any/Each of these and how they interact with current drivers may result in less than optimal performance for the 980/970 cards on any given application.

eXaPower
Send message
Joined: 30 Sep 13
Posts: 122
ID: 259902
Credit: 1,636,922,447
RAC: 260,510
Discovered 2 mega primesFound 1 prime in the 2018 Tour de PrimesFound 1 mega prime in the 2018 Tour de PrimesFound 1 prime in the 2020 Tour de Primes321 LLR Amethyst: Earned 1,000,000 credits (1,031,618)Cullen LLR Gold: Earned 500,000 credits (620,013)ESP LLR Amethyst: Earned 1,000,000 credits (1,043,069)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (585,120)PPS LLR Jade: Earned 10,000,000 credits (10,589,974)PSP LLR Gold: Earned 500,000 credits (506,876)SoB LLR Silver: Earned 100,000 credits (379,643)SR5 LLR Turquoise: Earned 5,000,000 credits (5,022,053)SGS LLR Amethyst: Earned 1,000,000 credits (1,006,021)TRP LLR Ruby: Earned 2,000,000 credits (2,403,013)Woodall LLR Gold: Earned 500,000 credits (509,918)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,035,186)Generalized Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,122,369)PPS Sieve Double Gold: Earned 500,000,000 credits (758,913,230)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Ruby: Earned 2,000,000 credits (2,058,245)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (3,185,685)AP 26/27 Jade: Earned 10,000,000 credits (10,192,403)WW Double Bronze: Earned 100,000,000 credits (173,424,000)GFN Double Gold: Earned 500,000,000 credits (658,294,252)
Message 79942 - Posted: 3 Oct 2014 | 12:58:26 UTC
Last modified: 3 Oct 2014 | 13:14:17 UTC

To clarify: The core numbers are from CUDA toolkit 6.5 documents ---Table 2 in the current CUDA programming guide--- the throughputs of the arithmetic instructions natively supported in hardware for devices of various compute capabilities. All throughputs are for one multiprocessor (SM,SMX,SMM)

Note: "In this section, throughputs are given in number of operations per clock cycle per multiprocessor. For a warp size of 32, one instruction corresponds to 32 operations, so if N is the number of operations per clock cycle, the instruction throughput is N/32 instructions per clock cycle." and "All throughputs are for one multiprocessor. They must be multiplied by the number of multiprocessors in the device to get throughput for the whole device."

Kelly DaviesProject donor
Send message
Joined: 4 Apr 10
Posts: 107
ID: 58144
Credit: 6,305,744,229
RAC: 0
321 LLR Ruby: Earned 2,000,000 credits (2,050,518)Cullen LLR Ruby: Earned 2,000,000 credits (2,282,176)ESP LLR Ruby: Earned 2,000,000 credits (3,106,610)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,142,891)PPS LLR Turquoise: Earned 5,000,000 credits (5,014,094)PSP LLR Ruby: Earned 2,000,000 credits (3,475,366)SoB LLR Double Bronze: Earned 100,000,000 credits (180,081,042)SR5 LLR Ruby: Earned 2,000,000 credits (3,027,344)SGS LLR Turquoise: Earned 5,000,000 credits (5,042,519)TRP LLR Ruby: Earned 2,000,000 credits (2,010,534)Woodall LLR Ruby: Earned 2,000,000 credits (2,077,196)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,483,203)Generalized Cullen/Woodall Sieve (suspended) Sapphire: Earned 20,000,000 credits (28,991,973)PPS Sieve Double Turquoise: Earned 5,000,000,000 credits (6,003,339,738)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Sapphire: Earned 20,000,000 credits (24,236,704)TRP Sieve (suspended) Jade: Earned 10,000,000 credits (10,007,588)AP 26/27 Jade: Earned 10,000,000 credits (10,058,984)GFN Jade: Earned 10,000,000 credits (15,782,029)PSA Gold: Earned 500,000 credits (525,865)
Message 79943 - Posted: 3 Oct 2014 | 14:55:25 UTC - in response to Message 79942.

Fastest GPU list is not reliable.
I have a machine with two 780s and two 650Tis but BOINC sees it as four 780s, which would bring the 780 average down if that is how it is reporting.
similar with my other machines with mixed cards.
I like the 650TIs. good bang for the buck if you get on sale.
780s and my 770 do sieves in about 8-10minutes. 650TIs in about 20 minutes, and is pretty consistent across computers up to 7 years old.
____________
My Lucky Number is 1893*2^1283297+1

Scott BrownProject donor
Volunteer moderator
Project administrator
Volunteer tester
Project scientist
Avatar
Send message
Joined: 17 Oct 05
Posts: 2248
ID: 1178
Credit: 9,810,783,138
RAC: 13,705,804
Discovered the World's First base 116 Generalized Cullen prime!!!Discovered 28 mega primesEliminated 7 conjecture "k"sDiscovered 1 Sophie Germain pairDiscovered 2 Fermat divisors2012 Tour de Primes highest prime count2012 Tour de Primes most Mountain Stage primes2015 Tour de Primes highest prime count2016 Tour de Primes highest prime countFound 23 primes in the 2018 Tour de PrimesFound 1 mega prime in the 2018 Tour de PrimesFound 2 primes in the 2018 Tour de Primes Mountain Stage2019 Tour de Primes highest prime countFound 22 primes in the 2019 Tour de Primes2020 Tour de Primes highest prime scoreFound 21 primes in the 2020 Tour de PrimesFound 4 mega primes in the 2020 Tour de PrimesFound 8 primes in the 2021 Tour de PrimesFound 1 mega prime in the 2021 Tour de Primes321 LLR Double Silver: Earned 200,000,000 credits (201,403,028)Cullen LLR Double Silver: Earned 200,000,000 credits (201,409,561)ESP LLR Double Silver: Earned 200,000,000 credits (203,249,784)Generalized Cullen/Woodall LLR Double Bronze: Earned 100,000,000 credits (109,580,172)PPS LLR Double Gold: Earned 500,000,000 credits (728,284,255)PSP LLR Double Bronze: Earned 100,000,000 credits (126,982,721)SoB LLR Double Bronze: Earned 100,000,000 credits (189,060,638)SR5 LLR Double Silver: Earned 200,000,000 credits (214,437,634)SGS LLR Double Silver: Earned 200,000,000 credits (202,538,622)TPS LLR (retired) Silver: Earned 100,000 credits (235,439)TRP LLR Double Silver: Earned 200,000,000 credits (201,215,056)Woodall LLR Double Bronze: Earned 100,000,000 credits (104,130,013)321 Sieve (suspended) Double Silver: Earned 200,000,000 credits (235,451,253)Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (83,794,448)Generalized Cullen/Woodall Sieve (suspended) Double Silver: Earned 200,000,000 credits (285,139,652)PPS Sieve Double Ruby: Earned 2,000,000,000 credits (2,728,939,994)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Double Silver: Earned 200,000,000 credits (203,523,358)TRP Sieve (suspended) Double Silver: Earned 200,000,000 credits (201,489,157)AP 26/27 Double Silver: Earned 200,000,000 credits (385,720,907)WW Double Gold: Earned 500,000,000 credits (896,876,000)GFN Double Ruby: Earned 2,000,000,000 credits (2,048,310,105)PSA Double Silver: Earned 200,000,000 credits (259,058,048)
Message 79944 - Posted: 3 Oct 2014 | 14:57:38 UTC - in response to Message 79942.

Okay, I see what you are looking at and think I understand what is going on here. The SMM of Maxwell is absolutely a more efficient and better structure than the SMX of Kepler, and that might lead one to believe that Maxwell GPUs should outperform Kepler GPUs as a matter of fact.

However, the key aspect to regard is the whole device throughput. With that in mind, NVidia hasn't really done much with the 980/970 vs. 780/770 cards. Essentially, Maxwell's superior design allows the 980/970 cards to do roughly the same amount as the 780/770, but do so using much less power. In other words, a GTX 980 with 2048 shaders (16 SMX with 128 shaders each) should perform about the same as a GTX 780 with 2304 shaders (12 SMM with 192 shaders each) since these yield very similar total theoretical single-precision GFLOPs.

Bottom line, I would not expect much in the way of performance increase for the 9xx cards over their 7xx counterparts for applications like PPS sieve. That said, they do use substantially less power, and thus, will produce much more work completed per unit of electrical usage.

eXaPower
Send message
Joined: 30 Sep 13
Posts: 122
ID: 259902
Credit: 1,636,922,447
RAC: 260,510
Discovered 2 mega primesFound 1 prime in the 2018 Tour de PrimesFound 1 mega prime in the 2018 Tour de PrimesFound 1 prime in the 2020 Tour de Primes321 LLR Amethyst: Earned 1,000,000 credits (1,031,618)Cullen LLR Gold: Earned 500,000 credits (620,013)ESP LLR Amethyst: Earned 1,000,000 credits (1,043,069)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (585,120)PPS LLR Jade: Earned 10,000,000 credits (10,589,974)PSP LLR Gold: Earned 500,000 credits (506,876)SoB LLR Silver: Earned 100,000 credits (379,643)SR5 LLR Turquoise: Earned 5,000,000 credits (5,022,053)SGS LLR Amethyst: Earned 1,000,000 credits (1,006,021)TRP LLR Ruby: Earned 2,000,000 credits (2,403,013)Woodall LLR Gold: Earned 500,000 credits (509,918)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,035,186)Generalized Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,122,369)PPS Sieve Double Gold: Earned 500,000,000 credits (758,913,230)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Ruby: Earned 2,000,000 credits (2,058,245)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (3,185,685)AP 26/27 Jade: Earned 10,000,000 credits (10,192,403)WW Double Bronze: Earned 100,000,000 credits (173,424,000)GFN Double Gold: Earned 500,000,000 credits (658,294,252)
Message 79947 - Posted: 3 Oct 2014 | 15:49:08 UTC
Last modified: 3 Oct 2014 | 15:50:28 UTC

A significant change from Kelper can be attributed to a more efficient architecture -- Maxwell Crossbar, dispatch, issue change from Kelper's design.

(192cores/32load-store/32SFU per SMX) Kelper shared all cores, Ld./ST unit, SFU with four warp schedulers (a flat design with one large crossbar---> issue sharing 32cores with 8 dispatches all going through to crossbar. Single issue for 32ld/st and single issue for 32SFU in a SMX)

Maxwell's [4] 32core block with one warp scheduler for each 32core subset. (1 warp scheduler per 32core subset/ [4] warp for [4] 32 core blocks for 128totalcore/32load-store/32SFU per SMM). Maxwell's now two dispatch---> separate crossbar-->issue per 32cores subset with separate issue for 8 Ld/st units and separate issue 8SFU.

Also, amount of Maximum number of resident blocks per multiprocessor changed from 16 for C.C 3.0/3.5 to 32 for C.C 5.0/5.2. This will speed up processing for thread block sizes 64 or smaller. Maximum number of 32-bit registers per thread for C.C 3.0 is 63, while C.C 3.5/5.0/5.2 allows 255. Dynamic Parallelism is not a C.C 3.0 feature, only C.C 3.5/5.0/5.2 can.



http://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/2

http://images.anandtech.com/doci/7764/SMX_575px.png

http://images.anandtech.com/doci/7764/SMMrecolored_575px.png

Profile [AF>Amis des Lapins] Phil1966Project donor
Send message
Joined: 20 Jul 13
Posts: 117
ID: 242019
Credit: 442,628,776
RAC: 131,533
Found 3 primes in the 2019 Tour de PrimesFound 4 primes in the 2021 Tour de Primes321 LLR Bronze: Earned 10,000 credits (29,564)Cullen LLR Bronze: Earned 10,000 credits (36,084)ESP LLR Bronze: Earned 10,000 credits (15,196)Generalized Cullen/Woodall LLR Gold: Earned 500,000 credits (512,981)PPS LLR Turquoise: Earned 5,000,000 credits (5,875,793)PSP LLR Bronze: Earned 10,000 credits (48,561)SoB LLR Bronze: Earned 10,000 credits (62,033)SR5 LLR Gold: Earned 500,000 credits (631,249)SGS LLR Amethyst: Earned 1,000,000 credits (1,459,279)TRP LLR Gold: Earned 500,000 credits (536,929)Woodall LLR Bronze: Earned 10,000 credits (54,727)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,116,301)Generalized Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (3,629,679)PPS Sieve Double Silver: Earned 200,000,000 credits (347,880,458)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Turquoise: Earned 5,000,000 credits (7,661,419)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,495,590)AP 26/27 Sapphire: Earned 20,000,000 credits (30,674,241)WW Ruby: Earned 2,000,000 credits (2,216,000)GFN Sapphire: Earned 20,000,000 credits (29,770,630)PSA Turquoise: Earned 5,000,000 credits (6,923,280)
Message 79956 - Posted: 4 Oct 2014 | 9:24:31 UTC
Last modified: 4 Oct 2014 | 9:50:41 UTC

Hello,

A short message to tell you I started running 2 * GTX970 Gaming G1 on (ie) PPS Sieve, on my son's PrimeGrid's account.

The first results are not bad at all !

http://www.primegrid.com/result.php?resultid=576262205

http://www.primegrid.com/result.php?resultid=576262198

Given the huge decrease of power use, these cards seem fantastic. (have them since less than 2 hours ;) )

Running PPS Sieve on both, the Corsair RM1000 PSU's fan is even not running !

Temp° are about 10° lower than the 780Ti.

Same improvement on other projetcs I quickly tested.

However, when I choosed these cards, I did not realize these are the OC version of the 970.

Gigabyte GTX970 Gaming G1 (they run at 1400 Mhz without manual settings)

@Scott Brown : Thank you for your detailled explanation.

eXtreme Warhead
Send message
Joined: 19 Nov 12
Posts: 4
ID: 180420
Credit: 157,774,293
RAC: 85
321 LLR Bronze: Earned 10,000 credits (20,021)PPS LLR Bronze: Earned 10,000 credits (55,457)SGS LLR Bronze: Earned 10,000 credits (11,016)Woodall LLR Silver: Earned 100,000 credits (105,960)Generalized Cullen/Woodall Sieve (suspended) Bronze: Earned 10,000 credits (77,003)PPS Sieve Double Bronze: Earned 100,000,000 credits (157,081,858)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Silver: Earned 100,000 credits (106,403)TRP Sieve (suspended) Silver: Earned 100,000 credits (316,574)
Message 80051 - Posted: 7 Oct 2014 | 13:45:56 UTC

not bad at all from your sight. as said above, those performance can be done by a 660ti at stock, perhaps even better and that is the worst case happened for those project, because things like folding@home can use the much more speed from the 970 and i think there will be much more projects which benefit from the newer cards as well. so at the moment only prime looks very "old fashioned" and unreasonable for me wioth the 970.

the 660ti uses about the same power as the 970 +- , so why should i make my 970 to another 660ti if there were other projects where my 970 is a 970...

and for me it's completely irrelevant, if thats all about dp-speed or not. i'm only the person which wants to support some science, but only then if my ferrari can be driven out and is not completely broken down to a stupid prius

Scott BrownProject donor
Volunteer moderator
Project administrator
Volunteer tester
Project scientist
Avatar
Send message
Joined: 17 Oct 05
Posts: 2248
ID: 1178
Credit: 9,810,783,138
RAC: 13,705,804
Discovered the World's First base 116 Generalized Cullen prime!!!Discovered 28 mega primesEliminated 7 conjecture "k"sDiscovered 1 Sophie Germain pairDiscovered 2 Fermat divisors2012 Tour de Primes highest prime count2012 Tour de Primes most Mountain Stage primes2015 Tour de Primes highest prime count2016 Tour de Primes highest prime countFound 23 primes in the 2018 Tour de PrimesFound 1 mega prime in the 2018 Tour de PrimesFound 2 primes in the 2018 Tour de Primes Mountain Stage2019 Tour de Primes highest prime countFound 22 primes in the 2019 Tour de Primes2020 Tour de Primes highest prime scoreFound 21 primes in the 2020 Tour de PrimesFound 4 mega primes in the 2020 Tour de PrimesFound 8 primes in the 2021 Tour de PrimesFound 1 mega prime in the 2021 Tour de Primes321 LLR Double Silver: Earned 200,000,000 credits (201,403,028)Cullen LLR Double Silver: Earned 200,000,000 credits (201,409,561)ESP LLR Double Silver: Earned 200,000,000 credits (203,249,784)Generalized Cullen/Woodall LLR Double Bronze: Earned 100,000,000 credits (109,580,172)PPS LLR Double Gold: Earned 500,000,000 credits (728,284,255)PSP LLR Double Bronze: Earned 100,000,000 credits (126,982,721)SoB LLR Double Bronze: Earned 100,000,000 credits (189,060,638)SR5 LLR Double Silver: Earned 200,000,000 credits (214,437,634)SGS LLR Double Silver: Earned 200,000,000 credits (202,538,622)TPS LLR (retired) Silver: Earned 100,000 credits (235,439)TRP LLR Double Silver: Earned 200,000,000 credits (201,215,056)Woodall LLR Double Bronze: Earned 100,000,000 credits (104,130,013)321 Sieve (suspended) Double Silver: Earned 200,000,000 credits (235,451,253)Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (83,794,448)Generalized Cullen/Woodall Sieve (suspended) Double Silver: Earned 200,000,000 credits (285,139,652)PPS Sieve Double Ruby: Earned 2,000,000,000 credits (2,728,939,994)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Double Silver: Earned 200,000,000 credits (203,523,358)TRP Sieve (suspended) Double Silver: Earned 200,000,000 credits (201,489,157)AP 26/27 Double Silver: Earned 200,000,000 credits (385,720,907)WW Double Gold: Earned 500,000,000 credits (896,876,000)GFN Double Ruby: Earned 2,000,000,000 credits (2,048,310,105)PSA Double Silver: Earned 200,000,000 credits (259,058,048)
Message 80056 - Posted: 7 Oct 2014 | 15:05:27 UTC - in response to Message 80051.

not bad at all from your sight. as said above, those performance can be done by a 660ti at stock, perhaps even better and that is the worst case happened for those project, because things like folding@home can use the much more speed from the 970 and i think there will be much more projects which benefit from the newer cards as well. so at the moment only prime looks very "old fashioned" and unreasonable for me wioth the 970.

the 660ti uses about the same power as the 970 +- , so why should i make my 970 to another 660ti if there were other projects where my 970 is a 970...

and for me it's completely irrelevant, if thats all about dp-speed or not. i'm only the person which wants to support some science, but only then if my ferrari can be driven out and is not completely broken down to a stupid prius


I am baffled by this comment given Phill1966's times for his 970s, which are running around 480 secs or so per work unit. Among the top 20 computers at PG, only the one with a Titan Black is faster. His times are better than the GTX 770s, 780s, and 780Ti in that group (by 25 seconds ore more), and his times are running 100 seconds or more faster than the 660Ti typically does (I know, because I have access to two of these).

He also just posted a 27 second time on the WSS work on PRPnet in another thread (the fastest times I have seen prior to that from any card were around 36 seconds). I'd hardly call that prius-level performance.


Profile rebirther
Avatar
Send message
Joined: 10 Aug 05
Posts: 783
ID: 85
Credit: 175,774,608
RAC: 0
Eliminated 3 conjecture "k"s321 LLR Silver: Earned 100,000 credits (186,594)Cullen LLR Silver: Earned 100,000 credits (106,665)ESP LLR Gold: Earned 500,000 credits (502,416)PPS LLR Gold: Earned 500,000 credits (504,111)PSP LLR Gold: Earned 500,000 credits (513,785)SoB LLR Gold: Earned 500,000 credits (564,944)SR5 LLR Ruby: Earned 2,000,000 credits (2,790,118)SGS LLR Gold: Earned 500,000 credits (501,099)TPS LLR (retired) Bronze: Earned 10,000 credits (46,235)TRP LLR Gold: Earned 500,000 credits (708,706)Woodall LLR Silver: Earned 100,000 credits (133,626)321 Sieve (suspended) Bronze: Earned 10,000 credits (21,527)Cullen/Woodall Sieve (suspended) Jade: Earned 10,000,000 credits (14,729,132)PPS Sieve Double Bronze: Earned 100,000,000 credits (132,786,707)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Gold: Earned 500,000 credits (510,317)TRP Sieve (suspended) Gold: Earned 500,000 credits (620,991)AP 26/27 Silver: Earned 100,000 credits (418,876)GFN Amethyst: Earned 1,000,000 credits (1,795,613)PSA Jade: Earned 10,000,000 credits (18,329,123)
Message 80057 - Posted: 7 Oct 2014 | 15:22:12 UTC - in response to Message 80056.

not bad at all from your sight. as said above, those performance can be done by a 660ti at stock, perhaps even better and that is the worst case happened for those project, because things like folding@home can use the much more speed from the 970 and i think there will be much more projects which benefit from the newer cards as well. so at the moment only prime looks very "old fashioned" and unreasonable for me wioth the 970.

the 660ti uses about the same power as the 970 +- , so why should i make my 970 to another 660ti if there were other projects where my 970 is a 970...

and for me it's completely irrelevant, if thats all about dp-speed or not. i'm only the person which wants to support some science, but only then if my ferrari can be driven out and is not completely broken down to a stupid prius


I am baffled by this comment given Phill1966's times for his 970s, which are running around 480 secs or so per work unit. Among the top 20 computers at PG, only the one with a Titan Black is faster. His times are better than the GTX 770s, 780s, and 780Ti in that group (by 25 seconds ore more), and his times are running 100 seconds or more faster than the 660Ti typically does (I know, because I have access to two of these).

He also just posted a 27 second time on the WSS work on PRPnet in another thread (the fastest times I have seen prior to that from any card were around 36 seconds). I'd hardly call that prius-level performance.



Yeah, this is amazing but how are the times of the GTX980?

Usucapio LibertatisProject donor
Avatar
Send message
Joined: 21 Apr 10
Posts: 742
ID: 59072
Credit: 644,954,295
RAC: 634,178
Discovered 1 mega prime2012 Tour de Primes highest prime scoreFound 2 primes in the 2018 Tour de PrimesFound 1 prime in the 2020 Tour de PrimesFound 1 prime in the 2021 Tour de Primes321 LLR Jade: Earned 10,000,000 credits (14,136,498)Cullen LLR Ruby: Earned 2,000,000 credits (2,665,899)ESP LLR Ruby: Earned 2,000,000 credits (3,461,313)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,005,831)PPS LLR Jade: Earned 10,000,000 credits (13,444,750)PSP LLR Ruby: Earned 2,000,000 credits (2,414,623)SoB LLR Ruby: Earned 2,000,000 credits (4,391,412)SR5 LLR Ruby: Earned 2,000,000 credits (2,716,765)SGS LLR Ruby: Earned 2,000,000 credits (2,389,989)TRP LLR Ruby: Earned 2,000,000 credits (4,112,658)Woodall LLR Amethyst: Earned 1,000,000 credits (1,664,441)321 Sieve (suspended) Ruby: Earned 2,000,000 credits (2,106,124)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,449,019)Generalized Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (2,388,253)PPS Sieve Double Silver: Earned 200,000,000 credits (392,359,856)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Ruby: Earned 2,000,000 credits (2,072,601)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,441,257)AP 26/27 Sapphire: Earned 20,000,000 credits (21,294,481)WW Sapphire: Earned 20,000,000 credits (28,012,000)GFN Emerald: Earned 50,000,000 credits (56,477,276)PSA Emerald: Earned 50,000,000 credits (79,952,619)
Message 80059 - Posted: 7 Oct 2014 | 16:58:05 UTC - in response to Message 80056.



I am baffled by this comment given Phill1966's times for his 970s, which are running around 480 secs or so per work unit. Among the top 20 computers at PG, only the one with a Titan Black is faster. His times are better than the GTX 770s, 780s, and 780Ti in that group (by 25 seconds ore more), and his times are running 100 seconds or more faster than the 660Ti typically does (I know, because I have access to two of these).



If I'm not mistaken, the Stderr of the 480 seconds WUs on his host (on the host linked on Phill1966's post) says: "Detected GPU 1: GeForce GTX 780 Ti". The GTX970 WUs report ~570 seconds.
____________
676754^262144+1 is prime

eXtreme Warhead
Send message
Joined: 19 Nov 12
Posts: 4
ID: 180420
Credit: 157,774,293
RAC: 85
321 LLR Bronze: Earned 10,000 credits (20,021)PPS LLR Bronze: Earned 10,000 credits (55,457)SGS LLR Bronze: Earned 10,000 credits (11,016)Woodall LLR Silver: Earned 100,000 credits (105,960)Generalized Cullen/Woodall Sieve (suspended) Bronze: Earned 10,000 credits (77,003)PPS Sieve Double Bronze: Earned 100,000,000 credits (157,081,858)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Silver: Earned 100,000 credits (106,403)TRP Sieve (suspended) Silver: Earned 100,000 credits (316,574)
Message 80061 - Posted: 7 Oct 2014 | 17:52:53 UTC
Last modified: 7 Oct 2014 | 17:53:24 UTC

for the problem with the double precision performance. if tthe 970gtx should be at the same, but practically even worse level like a 660ti, i ask me, why does the folding@home performance under single precision even with the 970 roll over everything else and still does at double precision about 33% better than a 670gtx which is around the 660ti...: http://www.anandtech.com/show/8568/the-geforce-gtx-970-review-feat-evga/14


If I'm not mistaken, the Stderr of the 480 seconds WUs on his host (on the host linked on Phill1966's post) says: "Detected GPU 1: GeForce GTX 780 Ti". The GTX970 WUs report ~570 seconds.

thx...about 500s was the time i hoped for before buying the 970...

i tested the same project as said above and the results are not on those levels where they should be, fact.

Profile Wingless WonderProject donor
Avatar
Send message
Joined: 25 Dec 12
Posts: 174
ID: 186381
Credit: 396,958,672
RAC: 83,364
Discovered 1 mega primeFound 3 primes in the 2018 Tour de Primes321 LLR Amethyst: Earned 1,000,000 credits (1,374,650)Cullen LLR Amethyst: Earned 1,000,000 credits (1,661,696)ESP LLR Gold: Earned 500,000 credits (710,747)Generalized Cullen/Woodall LLR Amethyst: Earned 1,000,000 credits (1,205,397)PPS LLR Ruby: Earned 2,000,000 credits (2,679,100)PSP LLR Ruby: Earned 2,000,000 credits (2,113,325)SoB LLR Ruby: Earned 2,000,000 credits (3,082,298)SR5 LLR Amethyst: Earned 1,000,000 credits (1,284,829)SGS LLR Gold: Earned 500,000 credits (632,629)TRP LLR Turquoise: Earned 5,000,000 credits (5,321,051)Woodall LLR Silver: Earned 100,000 credits (127,395)321 Sieve (suspended) Bronze: Earned 10,000 credits (88,433)Generalized Cullen/Woodall Sieve (suspended) Amethyst: Earned 1,000,000 credits (1,626,395)PPS Sieve Double Bronze: Earned 100,000,000 credits (135,099,567)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Gold: Earned 500,000 credits (555,042)TRP Sieve (suspended) Ruby: Earned 2,000,000 credits (2,089,296)AP 26/27 Turquoise: Earned 5,000,000 credits (6,347,510)WW Amethyst: Earned 1,000,000 credits (1,284,000)GFN Double Silver: Earned 200,000,000 credits (229,675,312)
Message 80063 - Posted: 7 Oct 2014 | 18:15:53 UTC - in response to Message 80061.

i tested the same project as said above and the results are not on those levels where they should be, fact.

Am I understanding correctly that you were expecting better performance from your GTX 970? If so, are you using EVGA Precision or other software to alter the card fan speed profile when under load to prevent downclocking as the card heats up? Are you monitoring the 970's core temperature when under load?

eXtreme Warhead
Send message
Joined: 19 Nov 12
Posts: 4
ID: 180420
Credit: 157,774,293
RAC: 85
321 LLR Bronze: Earned 10,000 credits (20,021)PPS LLR Bronze: Earned 10,000 credits (55,457)SGS LLR Bronze: Earned 10,000 credits (11,016)Woodall LLR Silver: Earned 100,000 credits (105,960)Generalized Cullen/Woodall Sieve (suspended) Bronze: Earned 10,000 credits (77,003)PPS Sieve Double Bronze: Earned 100,000,000 credits (157,081,858)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Silver: Earned 100,000 credits (106,403)TRP Sieve (suspended) Silver: Earned 100,000 credits (316,574)
Message 80072 - Posted: 8 Oct 2014 | 4:11:41 UTC

core temp is everytime under 40°C because of fullcover watercooling.

and no other software is used to manipulate anything of the card. the 970 runs @stock with constant 1266mhz boost clock which is used about 90% without app_info

it's only the program which cannot use all possibilities the card does bring

Profile Microcruncher*Project donor
Volunteer tester
Avatar
Send message
Joined: 28 Jun 09
Posts: 391
ID: 42625
Credit: 45,226,534
RAC: 0
321 LLR Silver: Earned 100,000 credits (131,556)Cullen LLR Bronze: Earned 10,000 credits (20,170)PPS LLR Silver: Earned 100,000 credits (293,229)PSP LLR Bronze: Earned 10,000 credits (31,786)SoB LLR Bronze: Earned 10,000 credits (20,916)SR5 LLR Bronze: Earned 10,000 credits (11,249)SGS LLR Silver: Earned 100,000 credits (101,042)TRP LLR Silver: Earned 100,000 credits (120,654)Woodall LLR Bronze: Earned 10,000 credits (30,702)321 Sieve (suspended) Silver: Earned 100,000 credits (108,873)Cullen/Woodall Sieve (suspended) Ruby: Earned 2,000,000 credits (4,002,190)PPS Sieve Sapphire: Earned 20,000,000 credits (34,201,696)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Bronze: Earned 10,000 credits (41,068)TRP Sieve (suspended) Silver: Earned 100,000 credits (131,415)AP 26/27 Gold: Earned 500,000 credits (660,671)GFN Bronze: Earned 10,000 credits (86,071)PSA Turquoise: Earned 5,000,000 credits (5,233,138)
Message 80501 - Posted: 23 Oct 2014 | 10:23:13 UTC - in response to Message 79938.
Last modified: 23 Oct 2014 | 11:22:21 UTC

An update to new CUDA 6.5 may straighten this out ?

No. Been there, done that...

BTW: Take a look at the operations per clock cycle per multiprocessor:

32-bit integer multiply, multiply-add, extended-precision multiply-add:

16 for CC 2.x.
32 for CC 3.x

"Multiple Instructions" for CC 1.x and now again for CC 5.x

Source: CUDA C Programming Guide (CUDA 6.5 SDK)

Message boards : Number crunching : GTX970 / GTX980 - nVidia Maxwell

[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 3.18, 3.20, 3.25
Generated 19 Apr 2021 | 3:23:27 UTC