CUDA and CUDART versions?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello all,

whilst running some very basic CUDA code, I get the error “CUDA version is insufficient for CUDART version” and my program dies.

This suggests to me that I have two different versions for CUDA and CUDART, but how can I verify this? How can I find out the version for each?

Thank you very much!

2

if you run nvcc --version, that should give you the toolkit version number:

avidday@cuda:~$ nvcc --version

nvcc: NVIDIA (R) Cuda compiler driver

Copyright (c) 2005-2009 NVIDIA Corporation

Built on Fri_Feb_19_19:12:59_PST_2010

Cuda compilation tools, release 3.0, V0.2.1221

if you look at the version number of the driver, you should be able to work it out from the following:

195/196/197 -> driver version 3.0

190 -> driver version 2.3

185 -> driver version 2.2

180/181 -> driver version 2.1

177/178 -> driver version 2.0

Version 2.0 is about as far back as my memory goes…

3

There are some subtleties in the driver numbering scheme. 195.62 on Windows, for example, runs the 3.0 beta but not 3.0 final–you’ll need to upgrade to 197.10 or later (or something like that). On Linux, you’ll need to be running 195.36.15 or later to use 3.0 final.

4

how may I fix this problem? I am not able to launch any kernel through my cuda code.

how can I find out the driver installed on my machine?

I got following info after nvcc --version

[codebox]Copyright © 2005-2009 NVIDIA Corporation

Built on Mon_Oct_26_09:40:14_PDT_2009

Cuda compilation tools, release 3.0, V0.2.1221[/codebox]

Thanks

Shadab

5

You can try using the API calls cudaDriverGetVersion() and cudaRuntimeGetVersion(). If the runtime version is less than the driver version, you have a mismatch (and need to update your driver).