OT: nVidia driver [was: Wish list]

[Paul A. Houle]

On Thu, 9 Jun 2005 07:58:45 -0400 (EDT), Sean <seanlkml at sympatico.ca>  
wrote:

>
> Actually you are tied to nVidia.  Every time a significant kernel change
> comes out you have to rely on them to produce a new driver for you.   If
> they tire of that exercise you are SOL or you'll have to go buy a
> different piece of hardware.   Why not just start out by buying a
> different piece of hardware that _is_ supported by open source and reduce
> the risk?

	(i) I've suffered through a lot of graphics cards that were "supported"  
by open source drivers that crashed all the time.  My nVidia card is the  
first one I've had in about six years that doesn't crash when I'm using  
it.  (Sometimes it gets crashed by screensavers,  but that's another job  
for "rpm --erase")

	(ii) There isn't any 3-d card with open source driver support that's  
within an order of magntitue of current nvidia and ATI cards in performance

	(iii) I like doing stream programming with the GPU

       In a lot of ways,  propreitary hardware/software combos from vendors  
like Apple and Sun are starting to look good to me.  Linux has a lot of  
quality problems because much of the hardware it supports is junk and it  
has bad drivers even for good hardware:  for instance,  Apache disables  
the sendfile() system call on Linux because some network cards supported  
by Linux are total crap and can corrupt data when using sendfile() on an  
NFS-mounted file.

	What's terrible is that there isn't any reliable way to know what's junk  
and what isn't.  I'll ask around online and it's like calling your average  
software vendor for support:  "Yeah,  there's a driver for that card,   
it's supported,  it's fine."  A year later I finally find out other people  
are having horrible performance and crashes too -- cold comfort.

>
> No, that's a false analogy.  There are _real_ risks when running binary
> only modules in kernel-mode.   Those same risks don't come into play with
> binary only user applications.   That's a big difference.   Not to  
> mention
> there is a very real risk of you losing support for your beloved  
> hardware.
>  Again, a problem that doesn't exist in your analogy.

	Yeah sure,  but there are risks everywhere.  You can get hit by a bus  
crossing the street.  Tainted kernel or not,  I've never seen a Linux 2.4  
system running non-scientific workloads on an SMP machines that didn't  
have strange concurrency problems.   There are lots of open source drivers  
that suck -- I'd rather trade a propreitary driver that actually works for  
an open source driver that crashes my machine.

	It might not be fair that good graphic cards are propreitary and that you  
can't make free drivers for 802.11g but the real choice is between being  
pure and being relevant:  you ought to be glad that I'm choosing to run  
Linux with modern graphics cards and modern wireless networking rather  
than choosing to foresake Linux so I can support modern hardware.

>
> Supporting open source as a preferred platform is no more ideological  
> than
> the arguments you're putting forth.  Again, there are perfectly viable
> fully open source solutions today for the vast majority of uses.  There  
> is
> nothing wrong with people promoting them over binary only solutions.
> Period.
>

	In some areas that's true.  Name a specific graphics card I should be  
using,  and show me some evidence that it can make it more than two hours  
without a crash and I might believe you.