CUDA Driver version insufficient for CUDA runtime version.

I have installed pgfortran on my GPU-enabled workstation under trial license.
When I run the sample multiplication code using the complier as

pgfortran -Mcuda matmul_drv.F90 matmul1.cuf

And the executable ./a.out is executed:

Starting host calculation.
0: ALLOCATE: 400 bytes requested; status = 35(CUDA driver version is insufficient for CUDA runtime version)

The GPU details are follows
Device 0: “Tesla M1060”
CUDA Driver Version: 3.10
CUDA Runtime Version: 3.10
Total amount of global memory: 4294770688 bytes
Number of multiprocessors: 30
Number of cores: 240
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 16384 bytes

I also add that CUDA Toolkit 3.2 is installed in my pgfortran compiler installation and the compiler command

pgfortran -Mcuda=3.1 a.f90

gives the response

“pgfortran-Error-CUDA version 3.1 is not available in this installation”

And same for -Mcuda=2.3

So why isn’t CUDS version 3.1 is not installed in my installation?

I found this on PGI site:

“If you run a program built with the version 3.0 toolkit on a system with a version 2.3 driver, you’ll get an error code of 35:”

CUDA version is insufficient for CUDART version

This, I think, is what my problem is.
So what should I do to get 3.1 toolkit installed?
Use an older release of pgfortran like pgfortran2010?

I have now installed PGFORTRAN 2010 (version 10.9) but the runtime problem has surfaced in another way.

I compile the matrix multiplication code with

pgfortran -Mcuda matmul1.cuf matmul_drv.F90

and when I execute a.out I get

enter array size and number of iterations to run
10 2
Starting host calculation.
ConfigureCall FAILED:9

Now this error “ConfigureCall FAILED:9” was supposed to have been corrected in the version 10.6 and then why is this cropping up here?

Pls help me with a workable combitnation of -Mcuda flags that would run on my TeslaM1060. i.e what Toolkit I should target and what compute capability?

Hi madhavan73,

We’ve made many improvements to the compilers over the last few years so I don’t generally recommend going backwards in versions. Instead, why not update you CUDA Driver version? It will still work with your card and will give you access to the latest versions of CUDA.

You can find the CUDA 4.0 development drivers at:

  • Mat