CPU/Hardware optimized for Erlang

[Thomas Lindgren]

--- Will Newton <will@REDACTED> wrote:

> I don't have any hard numbers of course (this is the
> Internet after all), but 
> I would imagine that the number of transistors
> required to implement the wide 
> array of instructions in such a chip limits the
> possibilities of putting more 
> cores on a die because eventually your power
> dissipation will become a 
> problem.

Note that after decoding, the CISC instructions have
been translated to uops, traces of uops, or what have
you. I believe the difference with a RISC is small
after that point.

I also think other architectural features are more
costly at this time, e.g., big branch predictors,
large multiported register files, large instruction
windows, fast clock ...

> http://www.metagence.com/Products/index.asp

I'll have a peek :-)

> Are you aware of any overviews of this processor
> architecture?

The APZ family run the Ericsson AXE switches, and are
programmed in PLEX. Lots of features will be
recognized by Erlang programmers.

While they are getting a bit long in the tooth (even
pre-2000, I heard of multiple efforts to write
emulators and JITs to get rid of them) they have had
at least a 20-year run. When I left Ericsson, pre-2K,
there were tens of thousands of people working with
AXE and PLEX.

Here's a sample article:

http://www.ericsson.com/about/publications/review/1999_03/85.shtml

I think ericsson.com has a bit more info if you
rummage around. (Presumably, there are also *plenty*
of readers on this list who do :-)

Best,
Thomas


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com