Im getting error mesages from my CUDA card that Im finding it all but impossible to solve. After 4 weeks I am a failure. The card is a Quadro FX 1600M that im using to develop with. Im aware its a cc11 and thye second message is related to it being less thatn cc13 but I dont know what the error relates to and how to get round it (if indeed I can). Is it simply that I cant call kernels on this card?
Im simply trying to run the example mmul from the manual.
I compile with
pgfortran.exe -V -ta=nvidia:cuda4.0,nofma,time,wait -Mconcur -Mvect -Mcuda=cc11,ptxinfo -mp -Mchkfpstk -Mipa=fast,inline -Mpfi -Minfo -o mgeno2_MC.exe mgeno2_MC.f90
the error message I get is
42
limit is not supported on this architecture
when i execute this line
Cdev = Csub(1:N,1:L)
and
8
invalid device function
Error! Kernel failed!
when calling the kernel
call mmul_kernel<<>>( Adev, Bdev, Cdev, N, M, L )
The properties of the card are:
Cuda REset device:
no error
number of cuda devices: 1
got cuda device: 0
Cuda set device:
no error
cuda_prop name:
Quadro FX 1600M
cuda_prop major: 100
cuda_prop minor: 10
cuda_prop MP count: 4
cuda_prop GlobalMem: 268107776
cuda_prop ConstMem: 65536
cuda_prop MemPerBlock: 16384
cuda_prop WarpSize: 32
cuda_prop maxThreadsPerBlock: 512
cuda_prop maxThreadsDim(1): 512
cuda_prop maxThreadsDim(2): 512
cuda_prop maxThreadsDim(3): 64
cuda_prop maxGridSize(1): 65535
cuda_prop maxGridSize(2): 65535
cuda_prop maxGridSize(3): 1
cuda_prop asyncEngineCount: 1
cuda_prop integrated: 0
cuda_prop canMapHostMemory: 1
cuda_prop concurrentKernels: 0
cuda_prop ECCEnabled: 0
cudaLimitStackSize: 0
cudaLimitMallocHeapSize: 0
Cuda Driver Version: 4000
Cuda RunTime Version: 4000
I note the stacksize is zero = not sure why.
Any help would be very gratefully recieved since Im now going mad
Mike