CUDA version

Questions and Answers : GPU applications : CUDA version
Message board moderation

To post messages, you must log in.

AuthorMessage
Joseph Monk

Send message
Joined: 31 Mar 07
Posts: 150
Credit: 1,181,197
RAC: 0
Korea, South
Message 918524 - Posted: 16 Jul 2009, 20:05:02 UTC

I downloaded CUDA version 2.2, but when I start up BOINC it says:

Thu 16 Jul 2009 08:10:05 PM KST CUDA devices: GeForce GTX 260 (driver version 0, CUDA version 1.3, 895MB, est. 104GFLOPS), GeForce GTX 260 (driver version 0, CUDA version 1.3, 896MB, est. 104GFLOPS)

Why does is say version 1.3?

System is Ubuntu 8.10 64 bit, using drivers 180 (the ones in the repos, manual install causes issues but if it's needed and there's a big improvement I'll deal with it).
ID: 918524 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 918527 - Posted: 16 Jul 2009, 20:13:24 UTC - in response to Message 918524.  

Why does is say version 1.3?

Because that is the maximum hardware limit CUDA version that your GPU can do. You can use later drivers, but not all new functions in those drivers will be used by your GPU.
ID: 918527 · Report as offensive
Joseph Monk

Send message
Joined: 31 Mar 07
Posts: 150
Credit: 1,181,197
RAC: 0
Korea, South
Message 918539 - Posted: 16 Jul 2009, 20:48:19 UTC - in response to Message 918527.  

I guess that makes sense... I would think the GTX 260 would do the latest... they're pretty new cards. Oh well. Thanks.
ID: 918539 · Report as offensive
Profile Sutaru Tsureku
Volunteer tester

Send message
Joined: 6 Apr 07
Posts: 7105
Credit: 147,663,825
RAC: 5
Germany
Message 921034 - Posted: 24 Jul 2009, 14:53:16 UTC
Last modified: 24 Jul 2009, 14:58:26 UTC


Hmm.. I don't know, if this is true..

Because AFAIK, current are only '1.3 CUDA GPUs' available (max.).

So I would go with the latest CUDA_Vx.. current 2.3 .

BOINC don't show the latest CUDA_Vx .

Also I would everytime install the latest nVIDIA_driver.. current 190.38 .


I guess BOINC interpret/identify the GPU from this ? :
Device 1 :
GeForce GTX 260
totalGlobalMem = 939196416
sharedMemPerBlock = 16384
regsPerBlock = 16384
warpSize = 32
memPitch = 262144
maxThreadsPerBlock = 512
clockRate = 1458000
totalConstMem = 65536
major = 1
minor = 3

textureAlignment = 256
deviceOverlap = 1
multiProcessorCount = 27

[one of my 4 OCed GPUs]


Ops.. after writing.. I see 'Linux'.. but I guess my 'windows statement' could be for Linux also.

ID: 921034 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 921107 - Posted: 24 Jul 2009, 20:15:35 UTC - in response to Message 921034.  

I guess BOINC interpret/identify the GPU from this ?

No, BOINC will not use information that the Seti application wrote away to determine the CUDA capability of your GPU. Nvidia provided a special API for BOINC to use, which will query the GPU directly and lift te required information from either the VBIOS or the chipset.

BOINC will then write that information to the client_state.xml file where applications such as the Seti CUDA application can read that.

The numbering of CUDA revisions is major.minor, where major is differences in the actual architecture of the GPU, while minor reflects incremental changes to the architecture. (1.0, 1.1, 1.2, 1.3)

The numbering of the CUDA libraries can be off, while they also show major.minor, where major is differences in the capabilities and minor is the incremental changes. (1.1, 2.0, 2.1, 2.2, 2.3)

The two numbers have nothing whatsoever to do with each other. Installing the latest drivers with CUDA 2.3 libraries won't make the hardware capability of your GPU get another number. A VBIOS update might do that, but not a software update.

Compare it to you running a DirectX 9 compliant videocard with DirectX 10. Your card will only use all of the DirectX 9 options within DirectX 10, it won't use all of DirectX 10. You need a DirectX 10 compliant card for that.
ID: 921107 · Report as offensive

Questions and Answers : GPU applications : CUDA version


 
©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.