On 5/16/06, Ray Heasman <[EMAIL PROTECTED]> wrote:
On Tue, 2006-05-16 at 12:28 +1000, Hugh Fisher wrote:
> Dieter wrote:
>
> > Would it be possible to split things up to allow an upgradable board?
> > A basic 2D board with a socket. Want fancy 3D stuff? Plug in the
> > extra chip. 2-3 years later plug in a new improved chip.
>
> A lot of people such as Ray Heasman seem to have this idea
> that "3D" = full OpenGL with hardware transform, lighting,
> programmable shaders ...
I was under the impression that the recent discussions about a shader
were done with the intent of including one on the chip, at which point I
considered how many gates we were talking about and my head exploded. I
tried to check on the wiki, but it was down.
I thought it was pretty clear that the programmable shader discussion
was forward-looking, hypothetical, and something that's to be
experimented with on a small scale with the FPGA. I did mention that
I'd be willing to entertain the idea of skipping a generation if it
were necessary, but that's only if it's necessary.
Also, more than once in this discussion, I referred to OGA as a
"fixed-function fragment shader". That means it's not programmable.
That means it's quite minimal.
That was part of the reason I was upset about 3D support. If we're
trying to get things into a limited number of gates, even texture
mapping is a waste, when just being able to scale a bitmap down would
probably be sufficient.
But you also said you wanted rotation. If you have scaling and
rotation, then you have all the hardware you need to do arbitrary
distortion.
The only texture feature we have that is "advanced" is MIP mapping,
but I intend to implement that with a relatively simple state machine
that does sequential fetches, thereby drastically limiting the
hardware required.
On the whole, I still see the idea of a mixed signal ASIC in a recent
process as being a pipe dream without some significant funding that
won't be coming from any of us.
We've had ASIC vendors tell us we could do this. I don't know HOW
they mean to do it, but if they can, we'll use it. Remember that many
ASICs are pre-fabbed. All the silicon is done and the first few
layers of metal. When ours is produced we only need masks for the
last few layers of metal. If they've got blanks with DACs on them,
GREAT.
If not, we'll have to do something different.
I'm also really disappointed that there won't be a cheaper dev card to
play with. The current OGD design might be nice from a developer
convenience perspective, but it sucks from the "affordable to an FPGA
hobbiest that wants to dabble" perspective.
You've got to be kidding. Aside from one or two products that have
ancient and very small FPGAs on them, OGD1 is dirt cheap. If you
compare it to something even remotely comparable in terms of features
and logic area, it's a steal.
Things cost what they cost. Do you want us to sell it at a loss?
Given how much time I would
have to spend on playing, I can't justify buying an OGD. I'd happily
settle for a cut down card that can crash my PCI bus if I screw up.
Second hand PCs are free for most geeks. I'd program it over a USB port,
from another machine, and be happy.
There may be some room for non populating some parts, but it probably
wouldn't bring the price down much.
I have realised that my impression of what OGP is, is wrong. It's not an
open source effort to make something that lots of people can develop on
and play with. It's a company that has an open source policy and also
happens to have a mailing list.
This is exactly opposite of the truth. OGP started out as a project
at Tech Source. They decided to drop the project, so I was left with
two colleagues and the OGP list. Going it alone, I needed a name and
a way to conduct business on my own. Traversal is just a front for
the OGP. Companies don't do business with the Linux Kernel Mailing
List, but they DO do business with OSDL. Traversal is a way to
centralize business. Of course, being a business, it has to make a
profit, and we may find it necessary to work on projects not directly
related to open source stuff (but not to the detriment of open
source!). But the primary intent is for Traversal to build what the
community wants. The drawback is that if the community gives us bad
information, and we don't figure that out, Traversal will go down.
Traversal doesn't need everyone to have
one of their dev cards, even if the card is cut down, because it's not
the goal. Making something that will encourage lots of hobbiest
developers to do weird things is not the goal.
Actually, it IS a goal. We have graphics as a long-term goal to serve
as a focusing principle, but my longer-term goal is an array of
FOSS-friendly products, including (or particularly) those that help
hobbyists design their own hardware. OGD may change form, but it'll
never go away as long as we can afford to exist.
The basic point behind Traversal is to:
- Enable the community to spec hardware for us to design or design it
themselves.
- Build that hardware in a way that is affordable to a wide audience of people.
So, OGD1 isn't "just" a dev platform. It's a major focus. Turning
complete designs into ASICs is what will allow others to afford to
benefit from the results of the community's efforts.
I also want an array of "high end" product that are natively
FPGA-based so that they are always reprogrammable. It's just graphics
that probably requires an ASIC for it to be affordable.
Don't get me wrong; I'd love to see open hardware out there. I hope
Traversal succeeds. I'll buy their stuff out of loyalty if it isn't too
expensive. I just think don't people like me constitute a market that
can be counted on, so there will have to be a reason for the average Joe
to buy it.
This is true in any case.
Judging from the questions that Timothy hasn't answered, I suspect OGP
will get their mixed signal design done for them in some kind of
backroom arrangement using someone else's seat licenses, and then all he
has to do is find money to tape out. I really wish him luck, but it's
not a project that I can contribute meaningfully to.
Why?
> This was discussed extensively at the start of the design
> process. The OGD/OGC are *not* intended to fully implement
> a complete 3D rendering pipeline.
It's hard to tell what is intended. Nothing is ever the last word, and
people are forever chiming in with new stuff. If you've missed one
email, you could have missed a change in direction.
There hasn't been a primary change in direction since early 2005. If
you want to know what OGA is, all you have to do is read the source
code. Oh, and there are PDFs that describe in English what it does
and the math behind it and everything. If you had done that, you'd
have realized long ago that it's not a programmable shader design.
> Note that the names given to these new X11 implementations
> include the letters GL, meaning OpenGL. The X Windows authors
> have already decided what hardware requirements they want
> from the next generation of video cards. They want OpenGL!
I think their choice is a pragmatic one, rather than one driven by any
real set of ideal requirements. If they could get the chip of their
dreams made for every X session out there, they would opt for a subset
and superset of OpenGL. ie. They would opt for something else. Current
generation 3D cards are both overkill and a very inefficient way to get
to the desired result, but they will do a fine job once the piper is
paid.
OGA is both a subset and superset of OpenGL. Really, it's just a
bunch of generic graphics math. In one sense, you could say that I
put together a design that was very general for both 2D and 3D and
then rearranged it a bit to conform to OpenGL. That doesn't mean I
locked it down into something too restrictive or something.
Do some general research on 2D and 3D graphics and how the math works.
Then read the OpenGL spec and see what it specifies. Then read about
OGA and see how it conforms to those two sets of ideas.
It would be senseless to intentionally incompatible with OpenGL if
it's trivial to make it conform without causing any harm.
For I while I was hoping OGP could break that, but I was unrealistic in
my expectations.
Exactly how much intertia do you want us to push against?
Plus, you've given us a chicken-and-egg problem. Have you talked to
the X.org people about this? Have you gotten any feedback on what
features they want? I'm not going to arbitrarily break things just to
be different. I have some information on what the needs are and
assembled a design based on that. If you can get better information,
I will alter the design (if necessary).
> X Windows is already being rewritten for 3D graphics cards,
> so designing and providing a 2D only interface for the OGD/
> OGC won't make adoption any faster and might even slow the
> process down.
Yeah, but it would get a low gate-count all-digital non-custom ASIC into
the hands of embedded developers, which would require little NRE and at
least bring in some money.
Gate count is a question we'll be able to answer in the not too distant future.
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
