Timothy writes:
> Just my opinion, but we shot too high with OGD1... scale it back until you
> have something simple and cheap.

I wasn't expecting OGD1 to sell well at the time, but you guys were
confident that you could sell at least a couple thousand of them.
I was quite skeptical but figured maybe you knew something I didn't
and kept my mouth shut, not wanting to rain on your parade.

At the time, PCIe was still pretty new, but it was *very* obvious that
PCIe was the new standard and it would quickly take over, especially
for video cards. Dispite this you guys insisted that OGD1 be PCI-X.
How many people have machines with PCI-X? Mostly in servers. Most
servers just need a simple console, text is enough. Yes you can
plug it into a plain 32 bit PCI slot, but you still need the physical
space for the card. Lots of machines don't have that space. Plus
a plain 32 bit PCI slot doesn't have anywhere near enough bandwidth,
even if it didn't need to share that bandwidth with the other PCI slots.

OGD1 was very expensive because of the FPGAs. If you want/need those
FPGAs *and* want video on the same card *and* have a PCI-X slot *and*
can afford the card I'm sure it was very attractive. Lots of hackers
like FPGAs, but can't or won't spend the bucks for it.

OGD1 didn't sell well because it was/is a very specialized, expensive card.

I'm trying for just the opposite. A reasonably priced card that
can serve the needs of a wide range of users. Not the cheapest
card possible that can't get the job done. Not a super high end
does every possible thing but no one can afford it card.

As nice as it would be to create a whole line of cards, we have
limited resources so I'm working on one card. We need at least
one for the GPU chip Timothy and team are designing. I am not going to
ask anyone to design a second card. If someone wants to volunteer to
design additional card(s) so that we can have PCI, AGP, 5 VGA ports,
7 HDMI ports, or whatever alternate features, that would be great.
I have seen comments that it is easy to create additional boards,
but so far I have not seen anyone volunteer to design them.

>> At least 2560x1600      Is 4096*2160 unreasonable?
>
> That would require dual-link DVI and really really good DACs.  Too
> expensive.  Stick to the max for single-link DVI.  This gets 90% of the
> users, even now.

There are still a lot of analog displays in use, so we need at least
1 analog head. What resolution does it need to be? Maybe the analog
doesn't need 2560x1600?

Single-link DVI is not enough. We need 2560x1440 for 27" and 2560x1600
for 30" displays. The 27" displays have come down a lot in price,
we need to support them. I am assuming that once we have 2560x1440,
that 2560x1600 isn't much more expensive, correct?

What I need to know is how much more (money and grief) would it cost
to support the 4K displays? I really want to support 4K if we can,
but I fear it might be too expensive.

>>  Option 1) Socket for FPGA / DSP / other?
>>  use the same pinout for OGP-GPU chip
>
> It can be hard to find a really small and a really large FPGA that share
> the same pinout.

I think we have a disconnect here? I wasn't expecting to offer the
choice of multiple optional FPGAs that could be plugged in.

>  I don't know of any socketed FPGAs.

I remember a FPGA that plugs into an AMD64 CPU socket. You could
take a mainboard with multiple CPU sockets and plug a FPGA into
one. These were discussed on this list once upon a time.

Since we already have the OGD1, picking a DSP chip might be a better
choice? (If there is a suitable DSP, and there might not be.)

> Too many conflicting constraints.

Welcome to engineering. And life in general for that matter.
_______________________________________________
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