Video Card for FreeBSD 9.0 (RC2) AMD64

[Matt Dawson]

On Monday 12 Dec 2011 19:22:55 Dieter BSD wrote:
> Full support requires full documentation or full
> reverse-engineering. nVidia is openly hostile towards FLOSS, I
> don't expect any documentation from them in the forseeable future.
>  AMD/ATI is working on documenting their chips, but seems to be
> concentrating on features for games, and never getting around to
> UVD (video decode) or GPGPU, etc.  The only fully documented card
> I know of is the Open Graphics Project's OGP-D1, which is a PCI-X
> FPGA development card. It has linux drivers, but I suspect it
> doesn't have support in BSD.

Just a quick comment on this: One of the reasons I choose FreeBSD over 
$OTHER_OS is that a lot of us are remarkably free of political 
encumbrance when we're trying to get things working, ergo the nVidia 
binary blob and the support from nVidia on the forums is generally 
accepted as a best-effort endeavour. 

When you're sitting in the living room setting up and HTPC with SWMBO 
looking on and making clicking noises and muttering things like "the 
XBox can already do this stuff" because of your dislike of decisions 
being imposed by large corporations, the ability to cd 
/usr/ports/x11/nvidia-driver && make config && make && make install to 
get something that Just Works [TM] rather than trying to get around 
other people's political views can be the difference between violent 
rejection and passive acceptance.

There are reasons why nVidia cannot release specifications, 
particularly on their PureVideo technology, which happen to be the 
same reasons AMD can't release theirs: They don't fully own those 
technologies. As it stands, I'm resigned to trading off full freedom 
of code for functionality. The important part, the interface between 
the kernel and the blob, is fully open in that you can see what passes 
between the two and ensure there's nothing freedom and privacy 
threatening going on.

Idealism is all well and good, but general acceptance in the real 
world requires a certain amount of compromise. There's another example 
right in our kernel: The Highpoint RocketRAID (hptrr(4)) driver has a 
closed binary component. It's right there in the man page for all to 
see.
-- 
Matt Dawson
MTD15-RIPE
matt at chronos.org.uk