The first ATI r5xx/6xx X11 driver release

I'm confused.  How can you have DVI output working and yet not have HDMI support finished?
They're the same signal...  Are they talking about the HDMI audio device that some ATI cards
have?

I think LWN missummarized the announcement... it's the first release of this driver; it's of
course NOT the first driver for this hardware; one can argue it's the third (after the normal
ati and then the avivo drivers).

Dave Airlie and co have the same features... and more in the "other", full driver already
since some time, including various pieces of acceleration:

http://airlied.livejournal.com/53129.html

To compete even with Intel driver, in terms of overall usability.
And a little more work to get in line witn NVidia closed-source offer.
Too early to announce about its existence, I think :>

I would love a 'cranky editor reviews video cards' article :) I'd like to upgrade the video in
my current desktop box (very, very old card) but have had a hard time figuring out what (if
anything) will get me decent dual-head performance with a Free driver. (I don't care about 3D,
but I imagine others care about that too.)

Hrm. 

So much hate towards the Intel drivers. In my experiance they've been adiquately fast, run
compiz acceptably well unless your running 3D apps on top of it, and have been stable. Xrandr
now allows hotplug, and yes there are a couple different ways to get DVI out on a Intel
onboard if you realy want it. (ie. by a motherboard with DVI out or purchase a ADD2 card and
add it to a existing0 motherboard)

Obviously if you need to a lot of 3D work and want to do gaming then Intel shouldn't be your
first choice. ATI's new proprietary drivers seem fine and Nvidia provides good performance.


If you realy want to know. Probably the biggest problem with Open Source drivers is the dual
driver nature of it. 

Currently, in Linux, with open source drivers, you have a minimum of 3 drivers running to
operate your video card at any one time.  

* You have the vga console or framebuffer stuff for the virtual console. This is from the
Linux kernel. 
* Then from X.org you have the xfree intel driver that provides acceleration for either XAA or
EXA framework. 
* Then on top of that you have the Mesa/DRI driver for opengl acceleration. 
(then the in kernel DRM driver to allow hardware access)

So essentially you have 3 different drivers from 3 different projects running one video card.
This is not a ideal situation.. in fact the whole architecture can be best described as just
very very bad. It's not so much designed, as just happenned over time.  

And not only that there has been a significant shift in how video cards are designed. There is
realy no such thing as a '2D' portion of the video card anymore. And not only that it's now
possible to use the 'GPU' as a more general purpose proccessor for running stuff other then
just 3D graphics. 

Now if you want to have your computer to have a entire proccessor that requires a significant
amount of proprietary middleware to run... then this is were things are headed. The GPU is
very powerfull stuff. For certain operations a modern video card can easily outpace a cluster
of Pentium machines running Linux. The way Nvidia and ATI have things setup right now is that
in order to utilize this power you have to program your software using their proprietary
tools. They have middleware that provides a abstraction for your application that in addition
to hiding the video card interfaces from you it provides a generic ISA that will work across
many of their video cards. 

This is the same thing that AMD and Intel do with their cpus, but it's done in hardware to
make the modern RISC-style proccessors they use appear as x86. 

So the way video cards and GPUs are going is that they are creating a new computer archecture,
a extension to x86, were they are refusing to disclose how to actually program for it. 

So this is why it's important to pay attention to projects like Gallium3D
http://www.tungstengraphics.com/wiki/index.php/Gallium3D

The way DRI is designed leads to very complex drivers. When combined with having to work side
by side with another driver or two that is made by other parties then it's obvious that it's
not going to lead to stable drivers. Now if all we used for acceleration was DRI.. have
everything purely OpenGL with no 'just 2d' exa or xaa then it has a much better chance of
working. 

But the goal of Gallium3D is a new driver achecture for Linux that is not only will provide
the features of Mesa/DRI it will make programming for modern hardware much more simplier. 

It is also suppose to not only provide acceleration for OpenGL it can provide acceleration for
all sorts of different APIs.. Glucose, OpenVR, DirectX, and support it for different window'ng
systems.. no-X framebuffer-style for embedded systems, Window, X windows/Linux/*BSD/etc

So you can get rid of the 2-3 driver nature of Linux graphics and go back to a single driver.
A single driver that is easier to deal with and supports multiple APIs, rather then _just_ EXA
or _just_ OpenGL.

That's the idea at least.

Just remember.. Those video cards are not just for graphics anymore. 

Looking at the comments, it seems the r200 is still the best overall floss driver supported
chip.

I used it in the past but went for an r300 a year or so ago, and I can confirm it's alot
buggier in 3d and slowish in some 2D operations (mozilla, evince/epdf (popler) adobe flash,
gnash). On the positive side, it's a little faster in 3D and some applications, like
openarena, quake3 are rock solid.