CUDA

I understand that FreeBSD supports NVidia graphics cards, via binary distribution from NVidia themselves, but, I was just wondering what the blocker is to a CUDA library on FreeBSD, anyone know? Is it just NVidia chose not to release CUDA for FreeBSD?
 
I understand that FreeBSD supports NVidia graphics cards, via binary distribution from NVidia themselves, but, I was just wondering what the blocker is to a CUDA library on FreeBSD, anyone know? Is it just NVidia chose not to release CUDA for FreeBSD?
Correction. FreeBSD doesn't support any graphics hardware (maybe graphics console). Graphics is supported via a third party application called XOrg which
includes open source drivers but can also utilize vendor binary blob drivers like Nvidia's infamous drivers.

Speaking of CUDA. The blocker is NVidia corporation itself. Unless they release CUDA for FreeBSD it is not going to happen. Maybe the real reason that they are so scared is that FreeBSD people are intelligent enough not to believe their PR BS and actually evaluate how useful is that technology for super computing. Over the past 2 years in our Lab (machine learning/data mining) we have evaluated CUDA three times and every time came up with the same conclusion: "Technology is of experimental quality at best and of questionable use in our research". Now I would be first to admit that our conclusions are valid only for our particular Lab and that people should try to use CUDA by all means (if we don't use we will not find out if it useful).

Unfortunately Scientific Computing is an area where BSDs are non-factor except in very special circumstances. The real show stopper is the lack of proprietary vendor support. In the order of importance for my professional work I would like to see native version of MATLAB (which depends on Oracle Java of course not just for GUI). Last time I checked FreeBSD foundation made agreement and Oracle Java will be officially released for FreeBSD. The next step is talking to MathWorks.
I would like to see serious proprietary computer algebra system working like Mathematica (MATLAB includes MuPAD)). A nice proprietary compiler like Intel or Portland Group compilers are really, really needed. Arguably really specialized research software like GAP System for Computational Discrete Algebra can be ported to anything but the fact that very few people who actually do scientific computing are now familiar with anything except Linux is what makes such porting efforts unlike in the real life.

Sorry to be pessimistic on this one but scientific computing ship might have been already sailed away for BSDs.
 
Correction. FreeBSD doesn't support any graphics hardware (maybe graphics console). Graphics is supported via a third party application called XOrg which
includes open source drivers but can also utilize vendor binary blob drivers like Nvidia's infamous drivers.

Speaking of CUDA. The blocker is NVidia corporation itself. Unless they release CUDA for FreeBSD it is not going to happen. Maybe the real reason that they are so scared is that FreeBSD people are intelligent enough not to believe their PR BS and actually evaluate how useful is that technology for super computing. Over the past 2 years in our Lab (machine learning/data mining) we have evaluated CUDA three times and every time came up with the same conclusion: "Technology is of experimental quality at best and of questionable use in our research". Now I would be first to admit that our conclusions are valid only for our particular Lab and that people should try to use CUDA by all means (if we don't use we will not find out if it useful).

Unfortunately Scientific Computing is an area where BSDs are non-factor except in very special circumstances. The real show stopper is the lack of proprietary vendor support. In the order of importance for my professional work I would like to see native version of MATLAB (which depends on Oracle Java of course not just for GUI). Last time I checked FreeBSD foundation made agreement and Oracle Java will be officially released for FreeBSD. The next step is talking to MathWorks.
I would like to see serious proprietary computer algebra system working like Mathematica (MATLAB includes MuPAD)). A nice proprietary compiler like Intel or Portland Group compilers are really, really needed. Arguably really specialized research software like GAP System for Computational Discrete Algebra can be ported to anything but the fact that very few people who actually do scientific computing are now familiar with anything except Linux is what makes such porting efforts unlike in the real life.

Sorry to be pessimistic on this one but scientific computing ship might have been already sailed away for BSDs.
Is this true in 2020 too?
 
Back
Top