AMD applies make-up to the face of its Linux Control Centre (the Inquirer)

Posted Jul 3, 2007 8:42 UTC (Tue) by Los__D (subscriber, #15263) [Link]

Now, I have no idea what it is that "nVidia" does, you can still run a nVidia card when setting it to "Force AIGLX", this makes the graphics 2-3 times slower (on my hw), but you avoid the black windows. The 2-3 times slower graphics is tolerable without some of the prettier effects like blur and reflection.

Maybe "nVidia" is just another way of driving AIGLX, but it sure is faster on nVidia hardware. I have no idea if it's more "direct", or if it's better, I just know that "Force AIGLX" is very slow on nVidia's hardware.

Posted Jul 3, 2007 9:05 UTC (Tue) by drag (subscriber, #31333) [Link]

And this may not remain true for ever. There is chances that ATI may decide to help open source driver developers (which I doubt)

A much more likely scenerio Intel is going to release new discrete video cards that will compete head to head with Nvidia's high-end offerings.

This is just hearsay and you know how stuff changes as time goes on.. But this is suppose to happen late 2008 or into 2009.

Described as x86-like in terms of programming API they will sport up to 16 GPU cores and have a gig or more of very high speed memory.

One of the more expensive things that happen in CPU making is that when you move to a new manufacturing technology, which is required for each new generation of CPUs being made, is that you have to pretty much abandon your old manufacturing lines. It's cheaper for AMD or Intel to simply build whole new assembly lines then to try to upgrade their existing facilities.

So I figure what Intel will do is when they upgrade to the next generation of state-of-the-art CPUs instead of using the old line to build Celerons or embedded stuff they will use it to build video card cores.

Seems like Intel's previous generation cpu making technology outclasses Nvidia's current generation stuff.

And the sky is the limit. Intel currently a 80 core cpu that they are using for testing and evaluation purposes... http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2925

The future is to have the GPU integrated directly into the CPU. When you have 80 cores I can't imagine that you'd want all of them to be general purpose stuff like we use now. Many of them would be for graphics proccessing and general floating point acceleration. Most of things that use that sort of stuff should be able to scale nicely with cores.

AMD/ATI and Nvidia seem to be heading towards making the GPU designs they have now and making them more CPU-like. Intel seems to be aiming at taking their CPU designs and making them more specialized, like IBM is doing with their Cell proccessor.

This may end up putting Linux in a unique position. It's more able to adapt to changes in hardware platforms compared to Windows.

Posted Jul 10, 2007 9:47 UTC (Tue) by daenzer (subscriber, #7050) [Link]

They have their own GLX implementation, top to bottom. It also supports
accelerated indirect rendering, but the acronym AIGLX usually refers
specifically to that part of the GLX implementation that comes with X.org,
which they don't use at all.

> AIGLX provides indirect 3D hardware acceleration. In order to do this your
> video drivers need to support "texture_from_pixmap" OpenGL extension.

Actually, the traditional AIGLX implementation of GLX_EXT_texture_from_pixmap
is 100% driver independent and involves several data copies, which IME
accounts for a lot of the perceived performance problems. xserver 1.4 will
support zero-copy GLX_EXT_texture_from_pixmap with AIGLX, which performs much
better but requires minimal driver support and only works with EXA.