You can reuse the FMC standard (VITA 57). There is plenty of space for
LVDS lines and also gigabit lines. It has two flavors - low pin count
and high pin count.
If you go this route, I would suggest to incorporate multiple slots,
even if they won't be fully wired:
1 - towards the outer slot (holding the custom video head combinations)
2 - on opposite direction (internal port - gpu or cpu or something
simular)
3/4 - potentially on the backside of the board (what did not fit into
port 2), but it somewhat violates the PCIe standard.
You can find base boards with FMC but they are rather expensive
developement platforms for desktop use, not suitable for in-PC use.
There already exist a card I described above -
http://www.ohwr.org/projects/spec/wiki You can fit it with a video
output, which you can even already buy ( dvi, hdmi or as displayport ),
but the price is higher (it is for development, not general use).
Of course, you can make some design changes to lower the cost, but the
price of the FMC connectors is "high" ( LPC, around 11.5eur+VAT @ 1 piece )
Daniel
On 12/10/2012 06:16 PM, Jack Carroll wrote:
This is the central issue, I think. A piggyback board is the only way I can
see a retrofittable plug-in GPU being practical. Why do you say signal
integrity would be destroyed? There's been quite a lot of work on fine-pitch
connectors that can pass differential transmission lines from board to board,
and we're only dealing with digital signals, since the DACs and analog cable
drivers would be either on the main board or in a remote adapter module located
at the monitor connector.
A piggyback connector could be standardized within OGP, and decouple the
mechanical interface to the frame buffer board from the interface to the GPU IC
packages. That would allow GPU designs to evolve over time, and plug into
older main boards.
I think the first question to ask is how many lines at what clock rate (or rise
time) would be required to communicate with the GPU assembly. Can I assume
LVDS signals?
Finally, what do you see as the cost impact of dividing the product into two
boards, besides the high-speed connector pair? I wouldn't expect a large
increase in total board area, and area is the main cost measure in multi-layer
boards. The price of the frame buffer board should come down, since it
wouldn't need the area required for the GPU.
Jack Carroll
----- Original Message -----
From: "Timothy Normand Miller" <[email protected]>
To: "Dieter BSD" <[email protected]>
Cc: "ogml" <[email protected]>
Sent: Monday, December 10, 2012 9:09:41 AM
Subject: Re: [Open-graphics] Open Framebuffer board
"Just plug it in"? No. There's a slight chance that we might be able to
fit a different FPGA to the same board, but there's no "just plugging in"
with a ball grid array.
We could make a daughter board, but the signal integrity would be
destroyed, and it would bloat the expense considerably.
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
