On Aug 23, 2007, at 7:09 AM, Dieter wrote:
Yes _good_ FPGAs are thousands of dollars ;-)
"Good" meaning large and fast, or something else?
Yes, large and to more of an extent fast. For example, some of the
Lattice Semiconductor high end FPGAs go up to 700 Mhz, but they
are > $2000 each.
There is a large (mobile phone/PDA/etc) market for lower end FPGAs
that
provide a processor bus for ARM and StrongARM CPUs, so they have gone
really far down in price.
I was under the impression that FPGAs were used for low volume and
development stuff. Things with the volume of mobile phone/PDA/etc
would support making a custom chip.
It really depends on the product. With the lower cost of FPGAs these
days, it
is sometimes a better deal to integrate one instead of an ASIC. Of
course, if
you are making millions and millions of them then ASICs provide a
cheaper
price in the long run, but being able to reprogram without getting the
silicon
refabbed is a big advantage.
MB bandwidth? MB = macroblocks? Megabytes? Mercedes Benz? :-)
Macroblocks, sorry.
ATSC uses 1080i rather than 1080p because they couldn't quite squeeze
1080p into the 19.something Mbps available. I wonder if 1080p/24 would
have fit, that would have been a better match for most 1080 material.
Hmm, interesting. Anybody know how much bandwidth we can reasonably
expect from 100 Mbit ethernet?
It is not obvious to me why OGD needs a large FPGA but we can get by
with a small one. Other than having two heads, what is OGD doing that
we don't need to?
They are implementing a full blown graphics card with OpenGL. This takes
up a lot of space. The more I read, the more it seems that H.264 is
really not
that computationally expensive (compared, at least, to OpenGL), it is
just that
the industry is used to the low compression, easy-to-deal-with MPEG
standards
that came earlier.
Just look at some testing I did this spring: I could run, using just
color transform
in hardware, the Elephant's Dream movie on my CPU no problem. There is
no
way that I could use Mesa to get anything near the performance of a
9200 gen
video card.
Just look at any OpenGL 1.5 block diagram, and compare it to the the
H.264
equivalent. Also, remember that H.264 is up to about 250000
macroblocks/sec;
the generation that OGD is trying to replicate was filling about a
billion triangles
a second. So even if it took 4000 times as long to process a
macroblock, an FPGA
the size of OGD's would be able to handle 1080p; in reality, it seems
as if it takes
about 1000 times as long, so it requires an FPGA that is 1/4 the size.
Have you read the "1st draft of Ethernet Video box requirements"
I posted in May? Should be one of the first articles in the archives.
Yeah, I temporarily forgot them. Looking at it again, it appears that
everything
except some proprietary stuff (RealPlayer) we can implement. We can
probably
just either reconfigure the FPGA (not good) or run a different DSP
program (good)
whenever we encounter a new format, which means that the FPGA just has
a bunch
of memory mapped functional blocks (as discussed) that are then pieced
together
and improved upon by the DSP.
Back to being allowed to implement things, we should probably find a
legal
way to implement an H.264 decoder without running into patent trouble
-- there
is a section of the wikipedia article about this that I suggest
everyone read. Though
nobody has decided to legally track down x264 or similar projects,
wikipedia made
it sound like they could at the drop of a pin.
The 6446 is designed for transcoding, so it has a resizer, etc, but I
think
you are right -- the 6443 should be fine.
The 6443 doesn't have a scaler? Ouch.
We should be able to implement one in the FPGA without too much
trouble. If it
can fit, shaving $10 off the BOM is worth it.
We will not have the Apple brand name, or a big advertising budget.
We do have the Internet. If we can build a good box for a low price,
word will spread and it should sell.
And we are connected to the OGP/OHF, so that should increase awareness.
If either TRV or this project really takes off, we can boost each other.
The volume question is mostly due to chips having significant volume
discounts.
Yes, of course. I think that for now at least, we can work with the
prices we have (x1),
and assume the worst case -- there is no demand for this and all we get
out of it is a
lot of wasted time and a couple dozen sales to other list members ;-)
_______________________________________________
Open-hardware-ethervideo mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-hardware-ethervideo
