Timothy Miller wrote:
On 8/14/06, James Richard Tyrer <[EMAIL PROTECTED]> wrote:
I think that we need to give this some thought. Can we build a
board that will be faster? Or, cheaper? Can we have it made in
China? Can we leverage existing technology to do so? Or, is
reinventing the wheel the best way to do it?
Cheaper is hard, because you need more money up front. With $20
million (to pull a random number out of my hat),
Which is why I suggested looking into Venture Capital firms. Yes, there
are a lot of bad things to say about some of them, but they do have money.
we can get lots more chips at a way lower unit cost.
Fixed costs are one of the main factors in the per unit cost. I realize
that the world has changed a lot in the past 10 years. There used to be
foundries that didn't even require that you order a full wafer, and
wafers were smaller.
10 years ago, I would have looked at semi-custom for part of the design
for a video board. Before they were called NorthBridge chips, there
were companies that would make a semi-custom NorthBridge chip. With
semi-custom you save almost all of the actual engineering costs. You
still have the mask costs and masks now cost considerable more than they
did when they were ordinary film. I realize that now such designs are
often done with cores for which IP you have to pay a one time fee (per
unit would be so much better for us).
Still, I would look into a semi-custom NorthBridge as part of the
design. This would give us:
System I/O (AGP, PCI, PCIe, etc.)
VGA core (including video controller)
Memory controller
Digital video output (raw)
Ethernet
Local I/O bus of some sort
Front side bus interface
Bus arbitration
And might also provide:
Analog video output
DVI video output
TV out
Hardware video scaling
These wheels have already been invented and might cost less if we used
off the shelf solutions leaving us to concentrate on our GPU which is new.
Of course, we have to sell them. And yes, we can get cheaper boards
by doing huge orders from Taiwan or China.
Is it necessary to place a huge order in China? Are there suppliers
that would perhaps be willing to do smaller runs than US firms?
That same up-front extra money would buy us more speed as well,
because we'd be going with a full-custom technology.
I fully realize that full custom and a single chip is the way to go for
high volume, but for lower volume, is a single chip the best solution?
IIUC, it is board cost vs. chip cost and the economies of scale kick in
much earlier for boards.
We'd be able to ramp up the clock rate, and with significant extra
engineering work, we could also widen the drawing engine.
I wasn't necessarily thinking about a faster clock. I was thinking
about more GPU hardware.
This is all some time off yet, though. Let's design something that
works, while we raise funds by selling OGD1 boards. We have an idea
of what throughput we'll get from the FPGA. When that's working, we
can shop around with ASIC vendors and make some estimates as to what
clock rate we'll get with each of them.
I have no problem with the first parts of the road map. But, I believe
that it is always good to think ahead and to explore the roads that we
have not chose to travel.
Note that we still expect TRV10 to sell better as a "high-end
embedded" chip than a desktop GPU.
Perhaps that is a large market that I really don't know much about.
However, this is a good fit with using a semi-custom NorthBridge chip
for a video board since this chip would not (always) be used with an
embedded design.
--
JRT
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
