CPU loyalty?

Werner Almesberger werner at almesberger.net
Mon Jun 6 10:57:24 EDT 2011


Jane Andreas wrote:
> I guess what I am trying to clear up is the rationale behind the
> current Ben CPU

I think it's mainly the result of opportunity - a CPU that's not too
nasty, for which an inexpensive device exists, and where the makers
of the device agreed to sub-license the design under acceptable
terms, make small changes, and produce the device in affortable
quantities.

For a small company like Sharism, finding anything that has all
these characteristics probably wasn't trivial.

I think we all can find numerous shortcomings in the Ben and its
components. But what matters most is that Wolfgang and his tiny team
managed to have a working product to form the basis for further work
in extremely short time. Better an ugly duckling in the hand than
"perfect" grand castles in the sky [1]. Besides, you never know what
grows out of these ducklings.

[1] http://en.wikipedia.org/wiki/File:Castlefalkenstein.jpg

Regarding CPU loyalty, I think there are several architectures with
readily available chips that are quite comparable and sadly all
imperfect when it comes to freedom. 

And then there are more open architectures, for which no suitable
ASICs exist. So in terms of pragmatism, MIPS doesn't seem a bad
choice to me. ARM would be about the same.

I also wouldn't dismiss an FPGA-centric design as a viable approach
for a general-purpose SoC so quickly. I think the potential of
implementing code in hardware is rather under-explored, in no small
part thanks to the secrecy the FPGA makers surround their synthesis
and routing with.

In a better world, we'd be working on a Ben successor now. Perhaps
with yet another Ingenic CPU (Ingenic have more powerful chips than
the Ben's 4720, and even the capabilities of the 4720 aren't fully
used in the Ben). There are many aspects of Open Hardware besides
the CPU that still need to be addressed. A while ago, I drew this
little chart of what levels of openness the various projects in the
qi-hardware universe have achieved:

http://downloads.qi-hardware.com/people/werner/tmp/achieve.pdf

Sebastien will probably object to the "MilkyMist Two" ;-) That's
just my projection of what shortcomings in terms of openness
existing in M1 would be worth addressing.

- Werner




More information about the discussion mailing list


interactive