I recently reported a problem where NVCUVID (H.264 video decode) works fine on a file on a GTX 670 but fails on a 8500 GT.
I am wondering if others have run into similar issues where CUDA code, libraries, etc. work on certain hardware but fails on lower or higher end hardware. I guess this also encompasses the issues where everything works fine on the big desktop machine and fails on a NVIDIA-equipped notebook.
Lots of our customers are running Windows on MacBook Pros and the new ones have NVIDIA GPUs. Do we need to separately test this hardware? Or has everyone found hardware incompatibilities to be rare?
As an addition to this thread, I just got a PNY GeForce GT 610 card and it exhibits the same characteristics as the GT 670 - it works! The GT 610 is 48 cores with 1GB of memory, compared to the 8500 GT which is 8 cores and 512MB of memory.
So, what ever issues there are, they are probably restricted to very, very low-end (old?) cards.
I am guessing from the silence on this that nobody has encountered compatibility problems, which is very, very reassuring.
In addition to the memory requirement. The 8500 GT is the first generation to support CUDA. Since than there were many features added like double precision and atomic functions and many others, more registers, more shared memory. Any library has should have mentioned the architectures on which it can run without problems. My first codes were done on a 8800m gtx in single precision, but it was just for learning purposes and I stopped soon when I changed my laptop and had access to newer cards. I would not bother to keep in my pc such an old card :D.