Mailinglist Archive: opensuse (5130 mails)

< Previous Next >
Re: [SLE] Re: [OT] Meta-discussion on the IP issues of GPUs and 3D -- WAS: Xgl on legacy nvidia cards
  • From: "Bryan J. Smith" <b.j.smith@xxxxxxxx>
  • Date: Wed, 24 May 2006 22:50:55 -0400
  • Message-id: <1148525455.4713.14.camel@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>
On Wed, 2006-05-24 at 21:59 -0300, Alvaro Kuolas wrote:
> Yes it's true, like a new CPU architecture on your AGP.

Ahh, no.

> But they don't change so fast.

Ahh, yes they do!

> nvidia has the TNT architecture from, well, the TNT trough the GeForce2.
> The GeForce3 was new, because of the vertex programs and pixel shaders.
> GeForce4 was a "power up" GeForce3.
> The new arch. came with the GeForce FX, it's said that was a VLIW
> architecture... a short lived one.
> Now it's the NV4x cards and G70 IS a NV4x.

Over-simplified. It's a heck of a lot more complicated than that.

Intel's base architecture remains _unchanged_ from 1994 -- 2 ALU, 2 FPU
(1 complex or 2 ADD), 3 control. They slapped on 1 SSE with P3, another
with P4. AMD is _unchanged_ from 1999, 3 ALU, 3 FPU (2 complex + 1
ADD/MULT), 3 control (4 in Opteron) -- the FPU is used for SSE.

Intel and AMD don't add more ALU, FPU and other units on-a-dime. They
go unused. Not all libraries can and will be re-written to be more
optimized and scheduled for the new units. Even AMD knows how hard it
is to do out-of-order execution and register renaming so it can leverage
it's additional ALUs and FPUs versus Intel -- and it only wanted to do
that _once_!

But GPUs do change them on-a-dime in _every_ generation it comes to
pixel, texel, vertex, etc...! The software libraries are then optimized
for that -- to standard APIs like OpenGL, their ARB extensions and the
occasional vendor extension (prior to assimilation into the next set of
ARB extensions).

> Now here we need to define the act of buying a hardware.
> What is buying hardware?
> For me the idea is: I bought a Hardware, not a service.

No, you bought a _high_complex_ set of _vector_processing_ that is
_extensively_driven_ by *SOFTWARE*. Again, CPUs are _not_ like GPUs!

> More if it's programmable, I want to use it at it's full extent. If not so,
> it's a service, I buy it to do it's job in a lock down way.
> In reality computers are not like cars.
> Too many patents really hurts economy. And why, if they only want 12
> months of head start, a patent grants to a Hardware/Software based model
> 20 years of ownership?

Then blame Intel.
Then blame Microsoft.

And don't forget IBM -- the biggest IP monster of them all (just ask the
Linux kernel developers) -- and the 1000lbs. gorilla that knocks
companies out of the way, while donating $1B to maturing its product
line (and _not_ actual GPL donations -- which makes me sick).

Blame a lot of companies.

ATI and nVidia didn't start it. In fact, nVidia _fought_it_, in the
name of Linux, during XFree 3.3.x by _releasing_ the source code!

That's because nVidia was already 80+% Linux internally by 2000, like
most other semiconductor firms running EDA applicatons. While Intel got
lambasted by SVLUG for their lack of supporting Linux, nVidia was moving
forward. I know because I was working in the semiconductor industry at
the time, and _knew_ the people involved with many firms.

Furthermore, without nVidia's _quality_ GLX support in 2000+, we would
have lost a _lot_ of POSIX/GLX CAM and EDA applications to Win32/DX
ports. Had those CAM and EDA applications gone Win32/DX, they would
have definitely been Windows-only forward! But because nVidia offered a
_quality_ GLX option on Linux, those apps stayed POSIX/GLX, and were
ported from AIX, HP/UX, Solaris and other UNIX flavors to Linux --
instead of Win32/DX.

CAM and EDA Engineering has been the _leading_ adopter of Linux on the
desktop/workstation. Thanx to them, Linux is a _very_mature_ GLX
platform. And thanx to nVidia, who recognized the need for a _quality_
GLX implementation on Linux over 6 years ago, we have it.

No, I don't like the IP issue. nVidia doesn't either. But because
nVidia (as well as ATI) have signed agreements with Intel, Microsoft,
SGI and others, we have them. Even if the community comes up with its
own 3D API, it will infringe on that same IP.

So get used to it for the next 5+ years, because that's just how it is.

Bryan J. Smith Professional, technical annoyance
Americans don't get upset because citizens in some foreign
nations can burn the American flag -- Americans get upset
because citizens in those same nations can't burn their own

< Previous Next >
Follow Ups