[CentOS] Re: Generating xorg.config -- varying "nv" driversupportbetween versions

[Bryan J. Smith]

"David Evennou (Data Masters)" <de at data-masters.com> wrote:
> Thanks Bryan!
> I have a few identical machines, to keep installs, etc.
> easier. I plan to use one or two of the machines for CAD
> work, so I will want to get better video cards.
> What video card(s) do you recommend for high performance?

Depends on what you define as "high performance?"
This will differ between consumer and engineering
applications.  (i.e., what applications?)

For consumer, general and _some_ engineering, nVidia's
Standardware (Open Standard, Proprietary Source) "nvidia" GLX
drivers are the most stable and highest performing.
nVidia's Quadro series are geared more towards engineering,
typically with additional memory and other options.
But even the GeForce 6800GT or latest 7800GTX does fairly
well.  And nVidia is the best when it comes to broad support
for their cards under Linux.

ATI is quickly improving their Standardware driver line for
newer cards, but they still seem to lag in broad support and
stability versus nVidia.  The limited Freedomware (Open
Source, Open Standard) drivers developed by Precision Insight
under contract from the Weather Channel (I believe?) are
based on older R1x0/2x0 series cards, and clearly not
engineering-quality.  While some people have gotten the DRI
GLX support to work with some newer RV2x0 series, and even a
few R/RV3x0 series cards, the features are limited because
ATI closed up all their 3D specs as of the R3x0 series.  So
ATI has now joined nVidia and Matrox in creating a
proprietary source-only driver for 3D/GLX support.

nVidia actually opened its "unified driver" 3D specs back in
the XFree 3.3.x days and NV0x (TNT2/early GeForce), but
because of IP ownership by Intel, Microsoft and others, they
found themselves in litigation.  Although some UtahGLX and
DRI work exists on these, the features are old and typically
don't work past NV10/11/15/17 (GeForce2 to GeForce4 MX). 
Sadly enough, the catch-22 is that the best way to stable and
performing OpenGL is to unify the driver code across Windows,
Linux, MacOS X, etc...  At the same time, that's an IP
mindfield, hence why they are proprietary source.  The former
"clean room" endeavors by DRI are noble, but the performance
and feature support (especially for engineering applications)
leave much to be desired, hence why ATI joined nVidia and
Matrox when it started to officially support Linux 3D/GLX.

Now more for engineering, 3DLabs has Linux GLX support as
well with their drivers.  Their newer Wildcat products have
exception performance for engineering (not so good for
general/consumer), although you'll typically pay for it (2-3x
what nVidia GeForce/Quadros will get you).  3DLabs also
adheres to strict OpenGL, and follows the Architectural
Review Board (ARB) process to the letter.  ATI and nVidia
introducing various extensions, and it's up to you to figure
out if they are supported by various applications.  [ NOTE: 
OpenGL/ARB is still not as bad as "DirectX is a wrapper to
vendor-specific APIs," but it's not as "clean" as OpenGL
should be, and I agree with 3DLabs' viewpoint. ]

[ Professional Note:  I know there is the cry for open source
in Linux, but OpenGL is an IP mindfield -- especially on the
"leading edge."  But at least OpenGL over X11 (GLX) _is_ an
"open standard," and had it not been for nVidia back in the
2000-2001 timeframe, many CAM and EDA vendors would have just
ported their codebases to Win32/DirectX, instead of sticking
with POSIX/GLX (doing ports from Irix/Solaris to Linux) thanx
to nVidia at least having a quality Linux 3D/GLX
implementation that was usable for engineering.  When there
is no viable "open source" option, I'll take "open standard,
proprietary source" on Linux versus "proprietary standard,
proprietary source" on Windows that would make it extremely
difficult for the code to _ever_ return to POSIX/GLX (and
making the applications Windows only). ]



-- 
Bryan J. Smith                | Sent from Yahoo Mail
mailto:b.j.smith at ieee.org     |  (please excuse any
http://thebs413.blogspot.com/ |   missing headers)