Rogelio Serrano wrote:
On 4/20/07, Timothy Normand Miller <[EMAIL PROTECTED]> wrote:
On 4/20/07, Raphaël Jacquot <[EMAIL PROTECTED]> wrote:
> how about using something like this then, wich allows to have a powerpc
> 405 core plus your own stuff next to it ?
>
>
http://www.xilinx.com/products/silicon_solutions/fpgas/virtex/virtex_ii_pro_fpgas/capabilities/index.htm
>
As far as I know, this is within the realm of reasonable. You and the
other CPU enthusiasts need to come to an agreement on some things and
also make an assessment of the relative complexities of things.
If you hit us with a design that's too difficult or complicated, we
won't be able to do it. Consider expense too, of both prototypes and
production.
I suddenly got very excited with this idea after stumbling upon the sgi
o2 uma.
we can put some opengl and video decode in the cpu core and we use
system memory for all video calculations. maybe we dont need memory in
the card anymore. is this possible?
Not possible with AGP, PCI, PCI-X. I don't know if PCIe is fast enough.
They claim 2.5 Gbits per second so 8 lanes would do 2.5 GBytes/second.
Will this scale and is that fast enough to access main memory? I don't
think so.
OTOH, HTX is fast enough since the current release is 2.6 GHz clock and
it is 16 bit DDR and you can use 2 lanes in a slot. Even with the
original (1.0 spec) 800 MHz, one lane would beat 8 lanes of PCIe, and 2
lanes would do 6.4 GBytes/second which would equal 128 bit DDR memory
running at 500 MHz. So, the HTX probably would allow unified memory
architecture. I would still have enough memory on the board to hold
what was on the screen since this allows direct access by the video
controller.
--
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)
