Re: Re: HARDWARE: Open-Source-Friendly Graphics Cards -- Viable?]

Re: Re: HARDWARE: Open-Source-Friendly Graphics Cards -- Viable?]

[Alexander Stohr]

Hi Jon, Hi audience,

(I just got aware of that discussion because i got that mail CCed 
trough a resend on a general discussion list about software patents.)

Even if ATI and nVidia, plus maybe even IBM, would sign a well written
and working agreement between them all, it would not stop anybody else
out there in the world that is holding patents from inspecting the unveiled
data and then looking for specific things that might work for pressing out
some Billion dollars from those companys. Patents do work, but they do
work mostly for the lawers income, and for companys that have the only
purpose for getting revenues from a "bought up" patent portfolio. Those
can be really nastys, even if you are Microsoft you dont like to pull out
some tons of code from your web browser just because someone else
is (really!) holding a patent for plugin technology like used for ActiveX.

BTW, did you know that the main study on SCO source in the Linux core
is from 1999, according to the SCO lawyers. Lets say it took two engineers
some two months for fiddeling out all those details in the 500.000 lines 
where there were similarities to SCO code - then that was an effort of 
some 30.000 USD - and according to SCO there was no bigger study
before that and no bigger study after that research. And now SCO is
reporting a 30.000.000 USD (thirty million) effort shedule for upcoming
lawyers work in the SCO vs. IBM case.

You dont want to go to court unless someone really forces you to do so.
You dont even want to do that despite you are as big as ATI or nVidia.
For such amounts you better want to hire some 300 high rank developers
rather than some 6 high paid lawyers with their full office staff - just to
stop others from charging you repeatedly with a per-chip-tax that sums
up to some 10 Million a year _per patent_ with options for up to back
propagation of charges for the previous 20 years. Much worser, as soon 
as you have redesigned your chip desing to something else less performant
there is absolutely no guarantee that tomorrow there will be no other
person starting charging you once again and again for another topic.

You might now understand, keeping the IP of a company a company secret 
despite having several patents is a vital measure for keeping the business
running well for the very own benefit and not for the pocket of some other
people, including the big big pockets of his lawyers. Technology was meant
for getting a nice refund from other people for providing them wanted products.
That is the true base of the "demand and delivery" market concept. Getting 
distracted from that economy core by courtroum events is contra productive.
Therefore its a well funded strategy of any company to avoid getting to court
so that they can in turn preserve their productivity and end up in a nice benefit.

-Alex.

> -------- Forwarded Message --------
> > From: Jon Smirl <jonsmirl@gmail.com>
> > Reply-To: Jon Smirl <jonsmirl@gmail.com>
> > To: Dave Airlie <airlied@gmail.com>
> > Cc: Kendall Bennett <kendallb@scitechsoft.com>, Linux Kernel Mailing
> > List <linux-kernel@vger.kernel.org>
> > Subject: Re: HARDWARE: Open-Source-Friendly Graphics Cards -- Viable?
> > Date: Mon, 25 Oct 2004 23:55:58 -0400

> > I wish they could just get together and agree not to sue each other
> > over stupid things like register designs and programming models. The
> > designs are horrible on both cards due to accumulation of historical
> > cruft. Save the lawsuits for the core of the engines if you really
> > have to sue each other.
> > 
> > -- 
> > Jon Smirl

Re: Re: HARDWARE: Open-Source-Friendly Graphics Cards -- Viable?]

[Jon Smirl]

On Fri, 29 Oct 2004 19:12:15 +0200, Alexander Stohr
<alexander.stohr@gmx.de> wrote:
> Hi Jon, Hi audience,
> 
> (I just got aware of that discussion because i got that mail CCed
> trough a resend on a general discussion list about software patents.)
> 
> Even if ATI and nVidia, plus maybe even IBM, would sign a well written
> and working agreement between them all, it would not stop anybody else
> out there in the world that is holding patents from inspecting the unveiled
> data and then looking for specific things that might work for pressing out
> some Billion dollars from those companys. Patents do work, but they do
> work mostly for the lawers income, and for companys that have the only
> purpose for getting revenues from a "bought up" patent portfolio. Those
> can be really nastys, even if you are Microsoft you dont like to pull out

The best way to make this work would be to get ATI, nVidia, IBM and
Intel into a room and do a cross licensing deal on the interface
portion of the designs. If the four companies also enter into a mutual
defense pact that will stop any third parties from causing trouble.

On the other hand, this strategy doesn't work if you are currently,
knowingly violating an existing valid patent.

Keeping things secret does nothing to protect you from a patent
infringement suit. All it does is make it a little harder to initially
detect that there are grounds for one. Once someone files suit they
will use the legal process to get all your secret designs anyway.

I also question if keeping interfaces secret is gaining anyone any
advantages over the competitors. Everyone involved possess an
excellent engineering staff capable of easily figuring out what the
other groups are doing. GPUs compete on functional units, chip
processes, parallelism, marketing and manufacturing cost, not on the
device driver programming model.

My belief is that everyone involved would gain from contributing to a
common pool of code for Linux. I don't believe that doing this is
going to alter anyone's market share; but it will make the users a lot
happier and breed goodwill for all involved.

-- 
Jon Smirl
jonsmirl@gmail.com

Re: Re: HARDWARE: Open-Source-Friendly Graphics Cards -- Viable?]

[Alexander Stohr]

Comments inline...

From: "Jon Smirl" <jonsmirl@gmail.com>
> The best way to make this work would be to get ATI, nVidia, IBM and
> Intel into a room and do a cross licensing deal on the interface
> portion of the designs.

Unless there is some benefit for them i dont expect them to spent any efforts 
on a think like that. They are stock based companies and act this way.

Some of them do dominate a market. If news are correct then nVidia has 
lost a not that small market share in the last 12 month so they are less likely
to even turn their closed linux driver business (that is closed for years) into
open source. Publishing old specs is not very likely, publishing new specs
is much less likely in such a situation. They are reall not a wellfare organisation.

On the other hand, ATI realy does not want to give nVidia or Intel
or any other sort of startup any sort of clues where there is space
for chip design improvement or where there are trapdoors when
implementing a specific grafics standard. They dont want to hint
others in designing better chips for sole comercial reasons.
Unlike main processors, grafic chips just need a driver - but they
dont need large scale information for user programmability.

> If the four companies also enter into a mutual
> defense pact that will stop any third parties from causing trouble.

That wont stop anything out there from sueing one after another, 
winning a few battles and then and receiving a multi million dollar 
payment from those big ones any now and then.

> On the other hand, this strategy doesn't work if you are currently,
> knowingly violating an existing valid patent.

It is much more the case that tons of patents are valid but you have
spent tons of research on the chip development as well, so the problem
of finding a case of infrignment before starting up with production is
a "N x N" problem which results in the need for checking some
million a-to-b-match cases with an intelligent human. You never 
ever can make sure, so you better not publish at all.

> Keeping things secret does nothing to protect you from a patent
> infringement suit. All it does is make it a little harder to initially
> detect that there are grounds for one. Once someone files suit they
> will use the legal process to get all your secret designs anyway.

If you publish anything then its just a nice reader and all is open.
If nothing is published then anthing needs much effort to get unveiled,
so the average time between two infrignment discoverys is significantly
different in its magnitude. Sorry, we are in a closed age of grafics
technology, its long ago when IBM sent out exhautive chip and board 
docu with its IBM-PC deliverys.

> I also question if keeping interfaces secret is gaining anyone any
> advantages over the competitors. Everyone involved possess an
> excellent engineering staff capable of easily figuring out what the
> other groups are doing. GPUs compete on functional units, chip
> processes, parallelism, marketing and manufacturing cost, not on the
> device driver programming model.

If you are a big-fish company then it is easy, but consider the startups.
As grafics business turned into a head to head show, there is good
reasons for the two heads to not raise the danger of changing anything
with that. Both heads are located in North America, but who would
ever stop for e.g. china from deploying its very own grafics company?

Compare to the situation in about 1995 where there was really much
of duplicate development, call it overhead, which was just a nasty
but it never really impacted market. Other things indeed did.
Hardware 3D was one major aspect that changed the situation
so that it was highly important to protect the small advantages
and this situation still has not changed.

If you really want an open source grafics standard, then you have
to launch your very own chip project. This will get you into the
state of beeing a hardware vendor and then you can check if it
is that easy. But be warned, others with much more comercial
backgrounder have failed already, like bitboys and their kyro.

> My belief is that everyone involved would gain from contributing to a
> common pool of code for Linux. I don't believe that doing this is
> going to alter anyone's market share; but it will make the users a lot
> happier and breed goodwill for all involved.

Drivers never were common code. Driver code is the mediator
between proprietary hardware and a general software interface.
It is okay to find generally needed resources and interfaces,
such as the AGP-gart does represent it, but it all fails where
some chip vendor is bound to its existing designs which tend to
be unique compared to any other vendors design. At least those
hardware does want that much different code that it rarely makes
sense to merge in any other dirver code. Look at the DRI codebase
and you will see a bunch of kernel modules, each for another user
space driver. If there were so much in common then there would
have been never a code split or already a merge between those code.

I do understand that you do want to have open source drivers for
all grafics hardware, maybe even in a unified form, but that does
conflict with the other vital interests of market oriented companies.

May i mention the for characters "S3TC" here?
It was long time clear how to implement and that it was know to be
implemented in an academic fashion for personal interest and research.
And then the debate ran for a longer time on how that could get 
implemented into DRI without getting the project and the users
into the legal danger of getting sued for patents infrignment.
On one hand it was so simple to resolve that technically, on the other
hand it was nearly impossible to integrate the solution into a release.

But the big fish companys of graphics business do have a license
and therefore only they were in state to release respective drivers.
OpenSource had no license and therefore felt impossible to merge
that. So even if some, many or all of the secrets of todays graphics 
chips would get published, there is still the question if it will ever be
open source and free software than. The industry is multi connected
trough its patents cross licensing, the OS/FS is not connected to 
that network and it does not look like it would ever be.

Open source must find its very own way - hoping that suddenly
something rather boosting will happen is not really a likely thing.
Not that you should stop asking for getting specs and documentation
but if it were all that easy and with no risk and no impact to other
sensitive aspects of the companys business - i am sure it would
have happened then. As it dont happens, assume there are such 
cases of possible bad impact, and none must of them must be
related to intentional patents infrigment, as long even the unintentional
can have the very same rather deep impact. Further the concurrency
battle is nothing that has suddenly ended in that market jsut because
the amount of market participants has massivly diminished and 
even the danger of a new company entering the market still exists.

You have to understand how stock based companies do operate
in order to see why they do behave as they do behave towards
you and towards other open source programmers, but even to
the rest of the world. Their business is not done wellfare reasons.

-Alex.

Re: Re: HARDWARE: Open-Source-Friendly Graphics Cards -- Viable?]

[Jon Smirl]

There are two ways to protect hardware innovations, trade secrets and
patents. Patents are fully published and trade secrets are not.  Trade
secrets are not a very good way to protect things since once they leak
they are gone. So if you have any good ideas get a patent on them, it
is a much stronger protection and it grants you a legal monopoly.

But patents are all published. So it makes no sense to hide things
that are patented, you can always just read the patents and find out
all of the details.

I don't see any other reason for keeping the programming model secret
other than fear of infringement suits. Many pieces of hardware have
their specs published and they aren't being sued. Why would ATI fare
any differently? I have the R200 specs, there is nothing in there that
hasn't already been done on dozens of other cards.

Why don't you publish the R200 specs on your website, it is older and
interest in it is rapidly falling. I'll bet nothing earth shattering
happens from publishing the spec except that a bunch of open source
developers stop pestering your development support group. You would
also get a lot of goodwill from the press announcement.

I also don't see how you conclude publishing programming specs makes
you a welfare organization. I still have to buy a card to use it. Open
specs will most likely increase your sales not lower them.

I'll keep working on building a base for X on GL. Right now I'm
working on integrating fbdev/DRM into something more coherent. The
basic idea is to bring up a standalone OpenGL with a few added things
like mode setting and cursor support. X will then run on top of that
using only the OpenGL API plus a few extensions for modes and cursors.
Hopefully you'll use my code to build proprietary drivers that support
the newer ATI cards in this model.

-- 
Jon Smirl
jonsmirl@gmail.com