OT: nVidia driver [was: Wish list]

[Bryan J. Smith <b.j.smith@ieee.org>]

From: Sean <seanlkml at sympatico.ca>
> There are clearly cases where the benefit of using a proprietary solution
> outweigh the risks involved.

By "proprietary" do you mean proprietary "source", or "standard" or both?
See how quickly the over-simplification of "open" v. "proprietary" can get?

I don't like the fact that the nVidia "nvidia" driver is proprietary any more
than you do, but there are sound, legal reasons for it (even if I'm not
aware of all the latest developments -- I just remember the ones from
the XFree86 3.3.x nVidia GLX release fiasco, as well as the initial
nvnet drivers, etc...).

At the same time, at least *1* person was unaware that most of the
latest nVidia GeForce 6000 series card (NV40, 42, 43, 45, etc...) work
"out-of-the-box" with Fedora Core 4, and even the 6800 (NV40) did
with Fedora Core 3.  I know, I loaded it on an Athlon64 PCIe GeForce 
6800GT and bam!  It worked solid at 2D.  ATI is _no_better_ than
nVidia, and they have only more recently joined them on the "unified
driver" approach and are 3+ years behind.

> However, there are _way_ too many people making excuses for
> abandoning open source.   Many of whom seem motivated
> by brand loyalty without any concern for system integrity or overall
> viability of open source alternatives.

And I agree with you.  IBM gets far too much credit right now, while
people demonize them.  Heck, even Sun gets demonized compared to
IBM, and people don't stop to think that IBM and Sun aren't much
different when it comes to licenses -- and at least Sun gave us
StarOffice/OpenOffice.org with a dual-license that includes LGPL.

> These bloody zealots for nVidia are every bit as much engaged in
> religion (or not) as the supporters of open source software (ie. both
> are just supporting what they believe in).

No offense, but I _am_ as much of an "apologist" for nVidia as Red
Hat, because there _are_ legal issues involved.  Whether it's NDAs
or trademarks, sometimes the hands of people are tied by 3rd party,
Common Law, etc...  Eventually these details _will_ get commodity.
I already appreciate the fact that more and more chipset-integrated
GPUs are supported out-of-the-box by GLX.

I spent 20 months in the semiconductor industry, and have many
colleagues at both ATI and nVidia.  I was there when the whole portage
of CAM and EDA tools to Linux happened, as nVidia started offering
quality GLX support for Linux.  A lot of vendors already had "lite"
versions running on Win32/GLX (via X11 emulation), and were working
on Win32/DX ports.  But once Linux opened up an "economies-of-scale"
POSIX/GLX platform, those vendors went to Linux instead.

It's not the game market -- it was the CAM/EDA market that put nVidia
where it's at.  Games are chump change in the Linux realm.  Sure, Sony
is going to change that, and GNU is already the development platform
outside of Microsoft for games.  But the reality is that most people are
completely oblivious of how important it was for an viable, performance
capable GLX solution to be available for Linux before the mass ports
started to Win32/DX.

> Anyway, my comments were not about abolishing personal choice or in
> denying the existence of _exceptional_ cases, rather as a counterbalance
> to the unthinking there-is-no-cost-at-all-in-using-binary-modules
> mentality.

I don't think _anyone_ is saying that.

I think people are just tired of ATI being "held up high" or other vendors
when nVidia _does_ put a number of people on the MIT 2D drivers.
I typically find that XFree/Xorg works _better_ out-of-the-box on newer
nVidia cards than ATI -- and the lag between release date is not much
different at all (typically in nVidia's favor).

Yes, sometimes there are new cards of a new nVidia NV3x or NV4x series
that aren't supported in the MIT drivers until the next XFree/Xorg revision.
But same deal for ATI!  Same deal for Matrox!  Same deal for a number
of vendors!  In fact, Matrox isn't open either.

Now from what I understand, the AGPgart and memory logic in nVidia's
kernel driver is becoming more open, because Intel just freed nVidia of
some contractual obligations on some of their more NDA stuff.  I don't
have all the details, because I'm not privy to them, but it explains a
lot.

> That applies both in terms of system integrity and the social
> implications.

Agreed.  But understand not all of us are "ignorant" of them.  In fact,
we are no more "ignorant" of them than those who are "ignorant" of
the benefit nVidia provided by having a GLX option in lieu of virtually
no other viable solution.

And the last time I checked, even the Freedomware BSD/MIT GLX
implementations aren't the best in stability either.  Which is how this
whole thread started -- people complaining its the proprietary model.
The reality is that we're talking about products being released and
obsoleted so fast that the drivers don't have time to mature -- be they
proprietary or open!  ;->



--
Bryan J. Smith   mailto:b.j.smith at ieee.org