I find this very dissapointing, as other FreeBSD users will, especially since the FreeBSD Display Driver – x86 page http://www.nvidia.com/object/freebsd_180.22.html
still says that CUDA 2.1 support has been added, and FreeBSD CUDA support has been mirrored in many third-party announcements.
When I saw the line "Added support for CUDA 2.1.", NVIDIA jumped an order of magnitude on my cred-o-meter, but has since fallen two orders of magnitude now that I realise it was a mistake that is still yet to be rectified. I had planned to make use of this new capability by adding CUDA throughput to CFD analyses I have been conducting on FreeBSD. On a per-processor-clock basis, FreeBSD runs these analyses faster than any Linux flavour, so I won't be abandoning FreeBSD any time soon. But I will have to abandon CUDA for some other solution.
My question to NVIDIA is: Can any niche product with limited mass-market appeal afford to be turning potential customers away ?