[lpi-discuss] What's New in the SUSE Linux Enterprise Desktop

Bryan J. Smith b.j.smith at ieee.org
Sun Apr 23 16:42:46 EDT 2006


On Sun, 2006-04-23 at 10:47 +0200, Anselm Lingnau wrote:
> Yes, if you're willing to run closed, proprietary,

First off, please _differentiate_ the term "closed/proprietary" whether
it means source, standard or both.

E.g., ATI and nVidia's GLX (OpenGL on X11) drivers are
closed/proprietary source, but _open_ standard.

> crash-prone graphics drivers

Huh?

The "main issue" with both the stability and source code availability of
the _kernel_ driver that is the memory interface is largely an _Intel_
created problem.  It's not only Intel's IP that ATI and nVidia license,
but it's largely due to the _design_ of the Intel platform.  If we used
AMD platforms with its _native_ HTX (peripheral slot on system bus), we
could avoid the need for this driver altogether.  Intel itself does
_not_ release that IP in its own i8xx/9xx GLX drivers for Linux.

As far as the ATI and nVidia X11/LibGL driver, they are quite stable
IMHO.  nVidia and now ATI have a tri-platform driver -- they wrap object
code for their unified GDI-Aqua-X11 (Windows-MacOSX-UNIX) around a
Linux-specific driver.  That is _legal_ from the standpoint of the
XFree86-4/Xorg MIT license.

In the XFree86 3.3.x days, before the new driver model, nVidia released
an open source driver for NV0x (TNT2/GF[1]) cards -- and they got
"cease-and-desist" letters from Intel, Microsoft, SGI and others.  Now
driver did help the UtahGLX project release drivers that support through
some NV1x cards (GF2), but still didn't solve the kernel-memory and
other performance issues.

It's really the kernel-memory interface driver where the stability
issues are it.  But in nVidia's defense, their AGPgart and other
interfaces are _more_stable_ than some of the GPL ones.  Can't speak for
ATI's.

>  from nVidia or ATI. Great »open source« desktop, indeed.

OpenGL is an IP mindfield, but as least it _is_ and _open_standard_.

Intel only supports a subset (and the performance versus Windows shows),
and they won't share their IP on _critical_ things, including the
kernel-memory interfaces that ATI-nVidia use.

Microsoft controls quite a bit of OpenGL, which they licensed in the
late '90s through the last few years.  Even if we have open source
drivers, it could be an IP issue.  And even if we come up with our own
replacement API, it still infringest on some of the 3D patents they've
secured that nVidia-SGI's donations don't cover.

> If you believe that all that has so far kept people from adopting Linux

Hold on!  Understand nVidia _saved_ a _lot_ of POSIX-OpenGL engineering
applications from moving to Win32-DirectX in 1998-2000, and moved them
to Linux instead.  It wasn't surprising though, as nVidia has been over
80% Linux since before Y2000, and it dominates their operations now.

Back then Wintel became much cheaper than RISC/UNIX, and many CAM and
EDA vendors were starting to port their POSIX-OpenGL to Win32-DirectX.
Thanx to nVidia, with a _complete_ GLX implementation for Linux, vendors
kept the apps POSIX-OpenGL, and Linux got a _massive_boost_ in user
share on the engineering desktop.

Today (and pretty much for the last 4-5 years), Linux basically
_dominates_ the mid-to-high-end EDA market, and has a majority share
(depending on if you include low-end CAM solutions, typically based on
commodity CAD applications) of the CAM market.

> on the desktop is the absence of utterly extraneous eye-candy à la
> Compiz/XGL,

While I agree with you that it's not causing people not to adopt Linux
on the desktop, your _technical_assumption_ is _dead_wrong_.

It is _not_ about "eye candy."  It is about _efficient_ and
_non-redundant_ use of windowing buffer.  The current mechanisms of
overlapping 2D in main memory, instead of leveraging the 2D (and 3D)
framebuffer of the GPU is the inefficiency.

Apple addressed this with QuartzExtreme, which basically uses OpenGL to
render -- even for 2D.  Red Hat and nVidia are using GLX composite
rendering atop of an existing X11 window manager, whereas SuSE is using
a complete GLX window manager.

> then I suggest you think again. I could list a dozen things that would be more 
> worth improving on the desktop, without even having to *start* thinking hard.

I disagree.  Microsoft's Avalon has _stalled_ for NT 6.0 "Windows Vista"
because they _failed_ to complete WGF 2.0 (based on DirectX 10).  They
are going with WGF 1.1 (based on existing DirectX 9) at release.  Even
with a high-end video card, it's _not_ doing GPU framebuffer rendering.

Both the composite and Xgl solutions on Linux _best_ it.

I personally _prefer_ MacOS X and it's QuartzExtreme _because_ it
renders "regular," *NON* "eye candy" 2D so much better than Linux or
Windows -- by a _wide_ margin.  It's far more smooth and easy to use.

> (Then again I wouldn't be surprised if you turn out to be right -- but that 
> doesn't say much for our users, if they're willing to put up with all sorts 
> of other atrocities just because they get to rotate their virtual desktops on 
> the sides of a cube in real-time while watching a DVD video in a window that 
> oozes across the edge, Dali-style. Personally I need these »features« about 
> as much as I need a hole in my head. Give me a better e-mail client any day, 
> thank you very much.)

Sun's LookingGlass (which is GPL) is actually the greatest advancement
in user interface in decades.  People dismiss it as "eye candy" and that
has _nothing_ to do with it.  It's a _true_ 3D API, not just rendering.

It's _never_ been about "eye candy."  That's just what the IT media
touts it as.  It's about the _proper_ way to render the desktop.
Today's approaches are _hacks_ and _slow_.


-- 
Bryan J. Smith             Professional, technical annoyance
mailto:b.j.smith at ieee.org       http://thebs413.blogspot.com
------------------------------------------------------------
****** Speed doesn't kill.  Difference in speed does! ******





More information about the lpi-discuss mailing list