Mailinglist Archive: opensuse (5130 mails)

< Previous Next >
Re: [OT] Meta-discussion on the IP issues of GPUs and 3D -- WAS: Xgl on legacy nvidia cards
  • From: Alvaro Kuolas <kuolas@xxxxxxxxx>
  • Date: Wed, 24 May 2006 13:13:28 -0300
  • Message-id: <44748628.9050800@xxxxxxxxx>
Bryan J. Smith wrote:
> On Tue, 2006-05-23 at 10:08 -0300, Alvaro Kuolas wrote:
>> What's wrong with you people!
>> Were is your GNU GPL spirit?
> It's called intellectual property (IP). IP is _always_ going to be an
> issue with GPUs (Graphical Processing Unit), _period_. Today's GPUs are
> _more_powerful_ than CPUs. It's like having a CPU with specialized
> vector units. And must is driven in specialized software. A lot of
> that IP is either Intel's or, increasingly via cross-licensing
> agreements, Microsoft.
> An additional problem at the _hardware_ level is the Intel platform
> design itself. The reason why ATI and nVidia need kernel drivers is
> largely Intel's fault. Intel continues to have _no_ "system"
> interconnect and connects the GPU via an "peripheral" interconnect. So
> it has to use hacks in software for coherency between the GPU and CPUs.
> If we all used AMD HyperTransport eXtension (HTX) for GPUs like we do
> Infiniband and select other expansions, then we wouldn't have a problem.
> But as of right now, except for a few non-commodity, multi-GPU,
> specialized systems, HTX isn't available for GPUs. And that's not
> likely to change with Intel's control over the GPU industry --
> especially when it comes to IP.
> But at least it's an _open_ API in OpenGL, including extensions via an
> Architectural Review Board (ARB) to keep vendor extensions from
> dominating out of next generations. That's completely _different_ than
> how DirectX is.
> So even if we don't have "open source," we _do_ have "open standards."
> It's only half of the equation, but given all the IP ownership issues,
> it's actually advantageous (from a legal perspective) for ATI, Matrox,
> nVidia and others to keep the source code closed. Especially with some
> of the key IP at the kernel level being Intel's (which they don't share
> in their own drivers).
>> For now I just buy computer parts that have data sheets freely
>> available or enough information for a good driver.
> Writing a GLX and corresponding kernel-level memory/interconnect
> interface driver for a GPU is about 1,000x time more difficult than a
> network driver for a MAC (Media Access Control) IC.
> It's like comparing a single stateful packet filter (SPF) firewall for
> layer 2-4 to a complete security appliance that also proxy and filters
> layers 5-7 as well.
> This is reality. It's not going to change. Especially since GPUs
> double in performance _twice_as_fast_ as CPUs.
>> That's a good way to support the really GNU/Linux Friendly enterprises.
>> ...and support Linux in general.
>> But, on the Video Card segment... is though!
>> Just nVidia and ATI... both on war. Not a friendly situation.
> Not true! Not true at all!
> ATI and nVidia, as well as Creative, Matrox and others, _do_ work
> together in the OpenGL ARB! They keep the _open_ standard going! So
> anything written today in OpenGL will _still_ work on OpenGL in the
> future!
> GPUs are _not_ MACs. They are 1,000x more complex -- especially in
> their software support. But at least we _do_ have an open standard.
What will happen to OpenGL?

SGI filled for Chapter 11 bankruptcy protection and 3Dlabs going off the
Professional Workstation market.
Worst yet, since 2003 Microsoft is the "number one" enemy of OpenGL.

GPUs are important today, they will be more in the future. It's
irrational how the market is turning, we need the same control over the
GPU as we have over the CPU.

< Previous Next >