On 8/2/05, Jack Carroll <[EMAIL PROTECTED]> wrote:
>         The key to making this work would be understanding IBM's needs and
> goals enough to figure out what OGP could do for them.  That would take some
> research and discussion, by someone with the right business talent to do the
> digging and understand the answers.  It doesn't sound to me as though anyone
> here has that talent, with the possible exception of Tim and his partners.

If someone wants to try to talk to IBM, that's fine by me.  Perhaps
there is a demand they have for a graphics card that we don't know
about in some obscure department.  What other sorts of business does
IBM do?  Perhaps, at this stage, we can still insert features that
other GPU designers don't provide.

>         Once that's accomplished, then perhaps it would be possible for
> Traversal to negotiate some kind of deal.  Ya never know.
>         I would venture a guess that if IBM decides to invest something into
> OGP, the least expensive option for them would be to fab the prototype chip
> lot in-house.

Most likely.  But remember, fabs run at full capacity all the time (I
would assume, given the investment to build one).  The likelihood of
them having a gap due to something other than equipment failure or
some other problem is low.  If they were to insert our chip into their
production line, they would cost themselves the profit from something
else they're making.  They have to be able to really solve a big need
with this.

>         Another thing to think about is who are the possible customers for
> OGP chips, besides boards built by Traversal and sold directly to end users.
> Manufacturers of commodity motherboards, maybe?  It ought to reduce their
> costs and give them a stable ABI for default on-board video.  That would
> require a driver for MS-Windows, though.

Well, not Intel, but perhaps AMD.  And of course, PowerPC.  The card
will work in a Sun.  Maybe Sun users would buy it.  I've written more
than my fair share of OpenWindows drivers, and Andy knows Solaris
kernel drivers.

>         Tim, I just had another thought.  Tooling up that first batch of
> chips is the big financial barrier.  The cost of a mask set and a batch of
> chips is heavily influenced by the choice of CMOS process, right?  Could an
> older, less dense process possibly produce a large enough gate count to
> build a useful, less feature-rich OGP chip that would fit a subset of the
> application universe?  (I suspect not, but it would be dumb not to ask the
> question.  There are foundries that specialize in that sort of thing.)

We've decided on standard cell.  There's gate array, standard cell,
and full-custom, in increasing order of NRE expense and decreasing
order of piece cost.  For the volume we want to produce (100k),
standard cell came out to be the best price.  It also performs well,
although not as well as full custom.

A number of different options were considered for this.

_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to