gpu support for modern systems

[Alexander Best]

On Sat Jun 25 11, "C. Bergström" wrote:
>  On 06/25/11 05:45 PM, Alexander Best wrote:
> >On Thu Jun 23 11, Christopher Bergström wrote:
> >>On Thu, Jun 23, 2011 at 6:43 AM, Alexander Best<arundel at freebsd.org>  
> >>wrote:
> >>>hi there,
> >>>
> >>>modern systems with their suffisticated gpus provide quite a potential 
> >>>for
> >>>moving some of the workload from cpu to gpu. for certain stuff gpus are 
> >>>much
> >>>faster than cpus, like number crunching or encoding/decoding multimedia
> >>>contents.
> >>>
> >>>anybody who is using mplayer(1) in combination with nvidia cards and 
> >>>vdpau has
> >>>probably experienced how much faster and less cpu intensive things can 
> >>>work out
> >>>when decoding HD video stuff e.g.
> >>>
> >>>since opencl/cuda isn't available under freebsd, it doesn't seem 
> >>>possible to
> >>>somehow hook the nvidia gpu into the every day freebsd workload that 
> >>>easily.
> >>PathScale has HMPP and partial CUDA support on FreeBSD now.  Some
> >>caveats to this
> >>
> >>1) Non-free (We could possibly make the tools free for FreeBSD
> >>community, but I'd have to get approval for it)
> >>2) CUDA support isn't complete (Basic core is there and HMPP C/Fortran 
> >>works)
> >>3) Tesla 20xx series only
> >>
> >>We offer real support for FreeBSD and it's not using any linux emulator 
> >>hacks.
> >>
> >>Let me know if anyone is interested.  (If we get enough positive
> >>feedback I'll pursue some resolution to #1)
> >i'd really love to see this stuff being made available to the community. i
> >think this could pose a huge advantage to all kinds of audiences: home 
> >users,
> >servers people, etc.
> >
> >basically every modern computer comes with a gpu that performs much better 
> >at
> >certain tasks than the installed CPU. not making use of the gpu is a huge 
> >waste
> >of ressources. i think microsoft had a video on their channel 9 platform
> >demonstrating the huge advantages that one can get when offloading certain
> >tasks onto the gpu. just think of gimp being able to perform 10 times 
> >faster!
> >
> >..or being able to compile world 10 times faster...or...or... ;)
> You missed the part where I said it's *Tesla* only.  This is really only 
> useful for HPC/life sciences/financial/server market.
> 
> This isn't to say we're not excited about it, but I'm still unsure who 
> if anyone is using FreeBSD in those areas.

...but is the CUDA api for tesla so much different than the CUDA geforce api?