Multiple CUDA Processors

Questions and Answers : GPU applications : Multiple CUDA Processors
Message board moderation

To post messages, you must log in.

AuthorMessage
Eric Pohlke

Send message
Joined: 20 Jan 09
Posts: 2
Credit: 261
RAC: 0
Canada
Message 855717 - Posted: 20 Jan 2009, 10:36:26 UTC
Last modified: 20 Jan 2009, 10:54:51 UTC

I have two GTX 295's on my eVGA X58 mobo. I have the two set physical cards set for SLI mode. Is there a way to get the CUDA enabled Boinc to address the second GPU on each card. I found out that Boinc (Cuda enabled) does not support SLI as SLI is not a true parallel interface of the GPUs. With SLI turned off, I get 2 processing threads in Boinc. But the 295's have 2 physical GPUs on each card via a FAT Parallel interface. However, I'm getting the same CUDA processing time from a GTX 295 as I from a GTX 280, meaning only one GPU is being addressed.
ID: 855717 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 855718 - Posted: 20 Jan 2009, 10:53:11 UTC - in response to Message 855717.  

BOINC should be able to find all GPUs, be it whether they are in SLI or not. But this has proven to be somewhat of a hit and miss in the earlier BOINC versions (including 6.4.5) and should be better (I dare not say fixed) in the upcoming 6.6.2
ID: 855718 · Report as offensive
Profile CADuke
Volunteer tester

Send message
Joined: 14 Mar 03
Posts: 5
Credit: 1,224,708
RAC: 0
United States
Message 855859 - Posted: 21 Jan 2009, 1:41:46 UTC - in response to Message 855718.  

Does Cuda scale with sli or not?
ID: 855859 · Report as offensive
Eric Pohlke

Send message
Joined: 20 Jan 09
Posts: 2
Credit: 261
RAC: 0
Canada
Message 855875 - Posted: 21 Jan 2009, 3:09:09 UTC - in response to Message 855718.  
Last modified: 21 Jan 2009, 3:14:15 UTC

Thanks for the response and I will try the newer version when available.

Currently, CUDA enabled Boinc only addresses the primary GPU on each card for me, and thus only addresses the cards seperately when SLI is disabled. With SLI enabled, there is one task listed and it's computations are the same speed as if there was only one 280 core. I was hoping to see all cores work together or atleast the paired cores work in parallel like the Telsa.
I don't mind this though; I wanted to see how benefical this system could be with a BOINC Project that will utilize the CUDA architecture. My other 2 systems have been with World Community Grid as it doesn't utilize CUDA just yet.
Intel i7 940, eVGA X58, 3 GB Corsair XMS3 DDR3-1333 3-channel, 2 x BFG GTX 295 (SLI), 1 x eVGA 280 (PhysX), X-Fi Titanium PCI-E, 4 x 30GB OCZ v2 SSD (BOOT RAID 0), 150GB VelociRaptor, 2 x 22X Plextor DVD-RW/DL,SD Drive, 1200 Watt PC Power & Cooling,
ID: 855875 · Report as offensive
Profile CADuke
Volunteer tester

Send message
Joined: 14 Mar 03
Posts: 5
Credit: 1,224,708
RAC: 0
United States
Message 855882 - Posted: 21 Jan 2009, 3:30:09 UTC - in response to Message 855875.  
Last modified: 21 Jan 2009, 3:58:11 UTC

I had to remove the SLI bridge as well as disable SLI and have PhysX enabled in Nvidia Control Panel for both of my 9600 GT to show up.
ID: 855882 · Report as offensive
Profile mr.kjellen
Volunteer tester
Avatar

Send message
Joined: 4 Jan 01
Posts: 195
Credit: 71,324,196
RAC: 0
Sweden
Message 855945 - Posted: 21 Jan 2009, 10:39:44 UTC - in response to Message 855717.  

I have two GTX 295's on my eVGA X58 mobo. I have the two set physical cards set for SLI mode. Is there a way to get the CUDA enabled Boinc to address the second GPU on each card. I found out that Boinc (Cuda enabled) does not support SLI as SLI is not a true parallel interface of the GPUs. With SLI turned off, I get 2 processing threads in Boinc. But the 295's have 2 physical GPUs on each card via a FAT Parallel interface. However, I'm getting the same CUDA processing time from a GTX 295 as I from a GTX 280, meaning only one GPU is being addressed.


This seems to be NVidias fault, the vista drivers are reported as not working properly...The Foldig and GPUGrid people are complaining as well.
ID: 855945 · Report as offensive
Robert P. Herbst
Volunteer tester
Avatar

Send message
Joined: 10 Jun 03
Posts: 45
Credit: 64,523,408
RAC: 142
United States
Message 855965 - Posted: 21 Jan 2009, 13:42:21 UTC - in response to Message 855718.  

Where does one find this version 6.6.2? I'd like to try some of this new stuff.
Please Visit Mount Perry, Florida
Home to Florida's Only Snow Capped Mountain
www.mountperryfl.com
ID: 855965 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 856008 - Posted: 21 Jan 2009, 16:14:41 UTC

All right, I asked the developer about this.

Since SLI basically combines 2 (or more) matched GPU devices into 1 logical GPU device, the system sees only 1 logical GPU and unfortunately for CUDA this means that it only has visibility to 1 physical device (not 2, 3 or 4). Disabling SLI mode for CUDA is best because it allows SETI to take advantage of each GPU on its own.
ID: 856008 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 856010 - Posted: 21 Jan 2009, 16:21:04 UTC - in response to Message 855965.  

Where does one find this version 6.6.2? I'd like to try some of this new stuff.

32bit version
64bit version
ID: 856010 · Report as offensive
Profile perryjay
Volunteer tester
Avatar

Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 20,676,751
RAC: 0
United States
Message 856067 - Posted: 21 Jan 2009, 18:48:33 UTC - in response to Message 856010.  

Thanks Ageless,


PROUD MEMBER OF Team Starfire World BOINC
ID: 856067 · Report as offensive

Questions and Answers : GPU applications : Multiple CUDA Processors


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.