Recently, i tried to test PGI Visual Fortran Accelerator on my workstation. Compilation of a very simple code achieved very well, but when i ran the program, an un-managed runtime error occured.
NB : my GPU is a nVidia Quadro NVS 290, and i selected “Compute Capability 1.1” in compiler properties.
I known that my card is CUDA - enabled, as template applications of the CUDA Toolkit from nVidia ran very well on it. Was Quadro NVS 290 tested by engineers of Portland Group ???
Thank you very much.
If your device supports CUDA, then we support the device. Hence, I suspect something else is going on. Can you please post the error message your getting?
Note that we do use CUDA version 2.3 hence your NVIDIA CUDA driver also needs to support CUDA 2.3. I have seen failures due to older drivers.
Thank you for your help
I will check driver version tomorrow. I think i actualised it very recently, but i have to be sure of that.
However, the problem is that i can see no error message at all. Error is not managed, and a debugging with visual studio is proposed. I will send you tomorrow error message the debugger display. An access violation if i remember well.
NB : the file nvcudart.dll was missing, according to compiled software. The file cudart.dll was present on my drive : i copied it in the release folder and modified it’s name. At this point, runtime error occured.
The debugger’s error message is (translated from french language) :
Not managed error at 0x00b390fe in gpgpu.exe : 0xC0000005: Acces violation during reading at 0x00000000000000a8.
My NVCUDA.DLL driver version is : 3.0.1
NB : the file nvcudart.dll was missing, according to compiled software.
This error occurs when you mix CUDA Fortran (-Mcuda) with the PGI Accelerator Model (-ta=nvidia). While we will support mixing the two as some point, currently they should not be used together. Selecting one or the other will most likely fix your access violation.
Hope this helps,
Thanks a lot… it worked !!!
However, now, i will test CUDA Fortran…