On Sat, Dec 5, 2009 at 2:30 PM, Matthieu Herrb <mhe...@gmail.com> wrote:
> On Sat, Dec 5, 2009 at 9:02 PM, rhubbell <rhubb...@ihubbell.com> wrote:
>
>> Yes, I'd like to see some pointers also.  I recall that there was
>> discussion (might've been on linux kernel) a while ago about a
>> partially-open video card.  Why doesn't the community support that?
>
> You mean http://www.opengraphics.org ?
> What makes you say that? How did *you* contribute?
>
>> I recall that price was a factor in lack of uptake.
>> Seems to me that opensource is farsical if it runs on closesource
hardware.
>> So where's the opensource hardware? Seems like the new world order isn't
>> going to allow that. The trend in hardware looks like a race to keep
>> control.  Seems like we are going to be paying for the hardware but not
>> owning; instead leasing.
>>
>> Or am I behind the times and there's salvation from some beneficent
>> hardware maker in Taiwan?
>
> Making hardware is a lot more difficult than writing software. So it takes
> more resources and more skills. This is probably why there aren't so
> many of them.

A graphics card is about both, and I don't think hardware is that
'hard' (if I can do it it must be easy :-).

The real barriers to entry are economic.  A mask set for a current
generation ASIC (say 45nm) is somewhere upwards of one million US $.
(http://www.eetimes.com/showArticle.jhtml?articleID=198900081)  so
unless you have terrific volumes to amortize over, custom chips are
pretty much out of play.  You could see if you can swing a deal with
someone to roll the mask costs into the piece price, but you'll have
to convince them that you'll run enough pieces to make it worth their
while.  In general the specialized design software needed to make a
design that will work costs a similar amount (digital and analog
simulators, logic synthesis, Design rule checkers, place and route,
design for test, etc).

That pretty much leaves you in the FPGA world like the opengraphics
people are.  You can use the FPGA vendor's volumes to get access to
leading edge process technology, but are stuck with their
architecture.  Usually the premium FPGAs start at around $100 ea. and
go up into the $1000 range, so you end up looking at the 'value' parts
(spartan and cyclone for brand X and brand A) to keep costs down which
limits what you can do even more.  The one nice thing about FPGA is
that the vendors usually make the parts for a long time, so
availability and control is good.

Reply via email to