Welcome to MilkyWay@home

Updated GPU Requirements


Advanced search

Message boards : Number crunching : Updated GPU Requirements
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 8 · Next

AuthorMessage
[TA]Assimilator1
Avatar

Send message
Joined: 22 Jan 11
Posts: 360
Credit: 42,274,270
RAC: 1,095
30 million credit badge8 year member badgeextraordinary contributions badge
Message 61284 - Posted: 1 Mar 2014, 16:19:02 UTC - in response to Message 61281.  
Last modified: 1 Mar 2014, 16:19:19 UTC

Yea, nbody & sep mod fit are Milkyway too, just different simulations.

Swiftm
Nbody is CPU only atm isn't it?
Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, Einstein@H, DHEP.

Main rig - i7 4930k @4.1 GHz, RX 580 8 GB, 16 GB DDR3 1866, Win 7 64bit
2nd rig - Q9550 @3.6 GHz, HD 7870 XT 3GB(DS), 8 GB DDR2 1066, Win 7 64bit
ID: 61284 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
swiftmallard
Avatar

Send message
Joined: 18 Jul 09
Posts: 297
Credit: 303,334,071
RAC: 7,849
300 million credit badge10 year member badgeextraordinary contributions badge
Message 61286 - Posted: 1 Mar 2014, 16:24:42 UTC - in response to Message 61284.  

Swiftm
Nbody is CPU only atm isn't it?

Correct
ID: 61286 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilemikey
Avatar

Send message
Joined: 8 May 09
Posts: 2247
Credit: 321,742,210
RAC: 868,706
300 million credit badge10 year member badgeextraordinary contributions badge
Message 61302 - Posted: 2 Mar 2014, 12:25:48 UTC - in response to Message 60726.  

From Matthew Arsenault:
R7xx is sketchy, since support was dropped from the driver (and I just removed support for it from the backend), so you need to use an old version of it. Anything newer than that should be fine.


Err, is the model number a typo? As the R700 series includes the HD 48xx & 47xx which do have DP & you list as supported (bar the 4730 which I guess is just an omission?), or are you referring to OpenCL support?


No I think he means the new R7 series.


No, he really means the 48xx & 47xx series, OpenCL support is very limited for that series.

So how does that affect MW? As my 4870 ran it just fine.
I know for Folding@H the poor openCL support for 4800s meant it was dropped from the project.

Also what did he mean by 'backend' & older version of what?


"Backend" usually means Server/Project side.
ID: 61302 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[TA]Assimilator1
Avatar

Send message
Joined: 22 Jan 11
Posts: 360
Credit: 42,274,270
RAC: 1,095
30 million credit badge8 year member badgeextraordinary contributions badge
Message 61306 - Posted: 2 Mar 2014, 22:35:00 UTC - in response to Message 61302.  

Ok thx mate, at least someone answered ;)
I still don't know what he's on about though as 4800s run MW just fine!

And what was he talking about when he said an 'older version'?
Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, Einstein@H, DHEP.

Main rig - i7 4930k @4.1 GHz, RX 580 8 GB, 16 GB DDR3 1866, Win 7 64bit
2nd rig - Q9550 @3.6 GHz, HD 7870 XT 3GB(DS), 8 GB DDR2 1066, Win 7 64bit
ID: 61306 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Atanas Kumbarov

Send message
Joined: 9 Mar 14
Posts: 3
Credit: 294,601
RAC: 0
100 thousand credit badge5 year member badge
Message 61415 - Posted: 16 Mar 2014, 14:56:38 UTC
Last modified: 16 Mar 2014, 14:57:10 UTC

I have a Nvidia GeForce GT 530 and I can't get it to work. Is this expected? I am getting a Nvidia Quadro 2000 soon. Is it supported? I can only see Quadro 4000 and above as supported.
ID: 61415 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilemikey
Avatar

Send message
Joined: 8 May 09
Posts: 2247
Credit: 321,742,210
RAC: 868,706
300 million credit badge10 year member badgeextraordinary contributions badge
Message 61416 - Posted: 17 Mar 2014, 10:53:53 UTC - in response to Message 61415.  

I have a Nvidia GeForce GT 530 and I can't get it to work. Is this expected? I am getting a Nvidia Quadro 2000 soon. Is it supported? I can only see Quadro 4000 and above as supported.


If I look at your pc's I only see one that has an AMD 5400 gpu attached to one of them, did you remember to run the Nvidia software driver software? If you do not do that Boinc cannot see the gpu as the standard Windows drivers are for standard display stuff only, they do not support crunching. If for instance I look at one of the gpu's from one of my pc's I see this---NVIDIA GeForce GTX 760 (2048MB) driver: 327.23 OpenCL: 1.01. I do NOT see anything in the gpu section of your pc's except the single AMD card.
ID: 61416 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Atanas Kumbarov

Send message
Joined: 9 Mar 14
Posts: 3
Credit: 294,601
RAC: 0
100 thousand credit badge5 year member badge
Message 61418 - Posted: 17 Mar 2014, 11:31:50 UTC - in response to Message 61416.  
Last modified: 17 Mar 2014, 11:33:07 UTC

Yes, I have the latest Nvidia driver and all of the Nvidia additional software.
ID: 61418 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Atanas Kumbarov

Send message
Joined: 9 Mar 14
Posts: 3
Credit: 294,601
RAC: 0
100 thousand credit badge5 year member badge
Message 61420 - Posted: 17 Mar 2014, 18:55:28 UTC - in response to Message 61418.  

So I removed the Nvidia GeForce GT 530 card and inserted a Nvidia Quadro 2000. Not it works. I guess the former card is not supported. Now I get CUDA jobs for other projects but not for this one.
ID: 61420 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[TA]Assimilator1
Avatar

Send message
Joined: 22 Jan 11
Posts: 360
Credit: 42,274,270
RAC: 1,095
30 million credit badge8 year member badgeextraordinary contributions badge
Message 61421 - Posted: 18 Mar 2014, 0:01:25 UTC - in response to Message 61420.  

Does the GT 530 even have double precision??
Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, Einstein@H, DHEP.

Main rig - i7 4930k @4.1 GHz, RX 580 8 GB, 16 GB DDR3 1866, Win 7 64bit
2nd rig - Q9550 @3.6 GHz, HD 7870 XT 3GB(DS), 8 GB DDR2 1066, Win 7 64bit
ID: 61421 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilemikey
Avatar

Send message
Joined: 8 May 09
Posts: 2247
Credit: 321,742,210
RAC: 868,706
300 million credit badge10 year member badgeextraordinary contributions badge
Message 61423 - Posted: 18 Mar 2014, 11:20:23 UTC - in response to Message 61420.  
Last modified: 18 Mar 2014, 11:28:04 UTC

So I removed the Nvidia GeForce GT 530 card and inserted a Nvidia Quadro 2000. Not it works. I guess the former card is not supported. Now I get CUDA jobs for other projects but not for this one.


What does your 'Event Log' in Boinc say about 'compute capability'? Open the Boinc Manager, down by the clock, and then click Advanced and then Event Log and look for any MilkyWay messages.
ID: 61423 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Miklos M

Send message
Joined: 29 Dec 11
Posts: 24
Credit: 9,111,619
RAC: 86,740
5 million credit badge7 year member badge
Message 61669 - Posted: 30 Apr 2014, 15:43:44 UTC

I am wondering, as I returned to Milkyway, with my two GTX 780 ti cards, why am I able to run only one of them on Milkyway at a time? On all the other projects it runs both. Perhaps I did not check off a box?
Appreciate any help.
ID: 61669 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bfromcolo

Send message
Joined: 27 Apr 13
Posts: 1
Credit: 12,350,372
RAC: 639
10 million credit badge6 year member badge
Message 61688 - Posted: 4 May 2014, 18:39:34 UTC

I tried this on my GTX-750ti in Ubuntu 14.04 and get the following. I am running 334.21 which as far as I know is the only driver for this card in Ubuntu at this time. The card runs Einstein, DiRT and Collatz fine. But it does not run Folding@Home currently. Is this just a matter of waiting for the developers to provide work units that support Maxwell, or is there something I need to do?

Sun 04 May 2014 10:22:23 AM MDT | | CUDA: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version unknown, CUDA version 6.0, compute capability 5.0, 2047MB, 1856MB available, 2409 GFLOPS peak)
Sun 04 May 2014 10:22:23 AM MDT | | App version needs OpenCL but GPU doesn't support it
Sun 04 May 2014 10:22:23 AM MDT | Milkyway@Home | Application uses missing NVIDIA GPU
Sun 04 May 2014 10:22:23 AM MDT | | App version needs OpenCL but GPU doesn't support it
Sun 04 May 2014 10:22:23 AM MDT | Milkyway@Home | Application uses missing NVIDIA GPU


Sun 04 May 2014 12:30:05 PM MDT | Milkyway@Home | [error] App version uses non-existent NVIDIA GPU
Sun 04 May 2014 12:30:05 PM MDT | Milkyway@Home | [error] Missing coprocessor for task de_separation_86_DR8_Rev_7_4_002_1398336302_3976752_1; aborting

ID: 61688 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
David

Send message
Joined: 8 Mar 14
Posts: 8
Credit: 10,100,868
RAC: 0
10 million credit badge5 year member badge
Message 61690 - Posted: 5 May 2014, 11:14:52 UTC

so i replaced my EVGA GeForce 460 with a GTX 760, uploaded the latest driver, and my average credit actually dropped??? This finding has me puzzled.
ID: 61690 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[TA]Assimilator1
Avatar

Send message
Joined: 22 Jan 11
Posts: 360
Credit: 42,274,270
RAC: 1,095
30 million credit badge8 year member badgeextraordinary contributions badge
Message 61691 - Posted: 5 May 2014, 11:22:50 UTC - in response to Message 61690.  

How does the double precision performance compare?

Many newer cards DP performance is worse than older ones!
Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, Einstein@H, DHEP.

Main rig - i7 4930k @4.1 GHz, RX 580 8 GB, 16 GB DDR3 1866, Win 7 64bit
2nd rig - Q9550 @3.6 GHz, HD 7870 XT 3GB(DS), 8 GB DDR2 1066, Win 7 64bit
ID: 61691 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
David

Send message
Joined: 8 Mar 14
Posts: 8
Credit: 10,100,868
RAC: 0
10 million credit badge5 year member badge
Message 61692 - Posted: 5 May 2014, 13:17:11 UTC - in response to Message 61691.  
Last modified: 5 May 2014, 13:53:56 UTC

i would assume that both were running double precision?

I did find this table, ranking gpu's:

(1.000) GeForce GTX 780 Ti
(0.909) GeForce GTX 570
(0.901) GeForce GTX 580
(0.871) GeForce GTX TITAN
(0.811) GeForce GTX 780
(0.737) GeForce GTX 470
(0.648) GeForce GTX 460
(0.604) GeForce GTX 680
(0.517) GeForce GTX 295
(0.512) GeForce GTX 770
(0.501) GeForce GTX 560 Ti
(0.474) Quadro 4000
(0.465) GeForce GTX 560
(0.462) GeForce GTX 670
(0.401) GeForce GTX 750 Ti
(0.387) GeForce GTX 690
(0.378) GeForce GTX 660 Ti
(0.368) Quadro FX 3800
(0.354) GeForce GTX 760

that surprises me, too! (note the relative position of the 460 vs. the 760)
ID: 61692 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[TA]Assimilator1
Avatar

Send message
Joined: 22 Jan 11
Posts: 360
Credit: 42,274,270
RAC: 1,095
30 million credit badge8 year member badgeextraordinary contributions badge
Message 61693 - Posted: 5 May 2014, 17:26:19 UTC - in response to Message 61692.  

Yea it would have to have DP otherwise it wouldn't run MW ;).

And yea that's why your 760 is so much slower in MW! :(
Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, Einstein@H, DHEP.

Main rig - i7 4930k @4.1 GHz, RX 580 8 GB, 16 GB DDR3 1866, Win 7 64bit
2nd rig - Q9550 @3.6 GHz, HD 7870 XT 3GB(DS), 8 GB DDR2 1066, Win 7 64bit
ID: 61693 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilemikey
Avatar

Send message
Joined: 8 May 09
Posts: 2247
Credit: 321,742,210
RAC: 868,706
300 million credit badge10 year member badgeextraordinary contributions badge
Message 61701 - Posted: 6 May 2014, 11:08:40 UTC - in response to Message 61692.  
Last modified: 6 May 2014, 11:18:33 UTC

i would assume that both were running double precision?

I did find this table, ranking gpu's:

(1.000) GeForce GTX 780 Ti
(0.909) GeForce GTX 570
(0.901) GeForce GTX 580
(0.871) GeForce GTX TITAN
(0.811) GeForce GTX 780
(0.737) GeForce GTX 470
(0.648) GeForce GTX 460
(0.604) GeForce GTX 680
(0.517) GeForce GTX 295
(0.512) GeForce GTX 770
(0.501) GeForce GTX 560 Ti
(0.474) Quadro 4000
(0.465) GeForce GTX 560
(0.462) GeForce GTX 670
(0.401) GeForce GTX 750 Ti
(0.387) GeForce GTX 690
(0.378) GeForce GTX 660 Ti
(0.368) Quadro FX 3800
(0.354) GeForce GTX 760

that surprises me, too! (note the relative position of the 460 vs. the 760)


I was looking and saw this chart:
Tesla series: 1/2 (full performance)

Quadro 4000-6000: 1/2 (full performance)

Quadro 600-2000: 1/12

GTX 5xx: 1/8

I know that doesn't compare your 760 and 460, but you get the idea, Nvidia is doing it on purpose.

I think it would be helpful if someone had the time and did the research to compile a list of each dual precision card and it's capabilities. That way we users could buy the best card for our money when choosing to crunch here. I know dual precision speed is not the only decision, but it is one of them.
ID: 61701 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[TA]Assimilator1
Avatar

Send message
Joined: 22 Jan 11
Posts: 360
Credit: 42,274,270
RAC: 1,095
30 million credit badge8 year member badgeextraordinary contributions badge
Message 61702 - Posted: 6 May 2014, 17:45:37 UTC - in response to Message 61701.  
Last modified: 6 May 2014, 17:48:54 UTC

Wikipedia lists the DP power of AMD/ATI cards, don't know why it doesn't for Nvidia :confused:

HD 3830*/3850*, 3870, 3850 X2, 3870 X2
DP .... 85.4 ....... 99.2 .. 171 ....... 211

HD ... 4730, 4770, 4830, 4860, 4850, 4870, 4890, 4850 X2, 4870 X2
DP 179/192 .. 192 ..147 ..179...200... 240 ... 272 ... 400 ...... 480

HD 5830, 5850, 5870, 5970
DP..358..417... 544 .... 928

HD 6930, 6950, 6970, 6990
DP..480...563...675..1277

*depending on version they have different memory bandwidth
DP = Double Precision power in GigaFLOPS

Lots of the HD 7000 series can do DP (although the 7730 & 7750s DP is pretty pathetic) but I've run out of time for today, feel free to add to that table :)

Let's see if the forum screws up the formatting...... [edit] yep! - fixed
Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, Einstein@H, DHEP.

Main rig - i7 4930k @4.1 GHz, RX 580 8 GB, 16 GB DDR3 1866, Win 7 64bit
2nd rig - Q9550 @3.6 GHz, HD 7870 XT 3GB(DS), 8 GB DDR2 1066, Win 7 64bit
ID: 61702 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
David

Send message
Joined: 8 Mar 14
Posts: 8
Credit: 10,100,868
RAC: 0
10 million credit badge5 year member badge
Message 61748 - Posted: 21 May 2014, 19:51:12 UTC - in response to Message 61702.  

OK so I put the GTX 460 in a separate somewhat lesser box, and started to work on MWAH - the problem now is that, using openhardwaremonitor, i am seeing only single digit (0-3-4-7%) usage values for GPU core, whereas before the GPU core was running at 90% or higher all the time.

I realize that there are several differences in the machines - different OS, different CPU, different MB, etc. but shouldn't the servers be able to direct better use of the 460???

am i missing something???
ID: 61748 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilemikey
Avatar

Send message
Joined: 8 May 09
Posts: 2247
Credit: 321,742,210
RAC: 868,706
300 million credit badge10 year member badgeextraordinary contributions badge
Message 61752 - Posted: 22 May 2014, 9:57:16 UTC - in response to Message 61748.  

OK so I put the GTX 460 in a separate somewhat lesser box, and started to work on MWAH - the problem now is that, using openhardwaremonitor, i am seeing only single digit (0-3-4-7%) usage values for GPU core, whereas before the GPU core was running at 90% or higher all the time.

I realize that there are several differences in the machines - different OS, different CPU, different MB, etc. but shouldn't the servers be able to direct better use of the 460???

am i missing something???


Are you leaving a cpu core free to feed the gpu?
ID: 61752 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 8 · Next

Message boards : Number crunching : Updated GPU Requirements

©2019 Astroinformatics Group