CPU/Hardware optimized for Erlang

Thomas Lindgren <>
Fri Jul 22 14:25:49 CEST 2005



--- James Hague <> wrote:

> On 7/21/05, Thomas Lindgren
> <> wrote:
> > 
> > I'm normally skeptical to special-purpose hardware
> for
> > specific programming languages -- such chips have
> > always gotten steamrollered by the killer micros.
> 
> In the past I would have agreed, but that was before
> the killer micro
> manufacturers started running into hard limits. 
> Power consumption has
> been increasing disproportionally with performance.

Even so, you will probably have to build an ASIC at
best, while Intel puts 400 people to hand-optimizing
the circuitry on the next x86 ... (they will thus
likely have a big clock speed advantage too).

(And why would the erlang chip have comparable
performance at lower power?)

I think the main drawback of an "erlang" chip is that
the tradeoffs are (not identical but) fairly similar
to a "C" or "Fortran" chip (aka workstation and
supercomputer). It's thus hard to see that it could
compete with a conventional server.

Or why not compare JAM, ECOMP and HIPE? ECOMP was
about 30x faster than JAM, while HIPE at the time was
about 10-20x faster than JAM. Would ECOMP have been
capable of evolving as quickly as JAM/HIPE on modern
hardware? If not, it would go the way of the Lisp
Machine.

That's not to say there is no role for a specialist
chip (perhaps one could spin a successor to APZ? :-)
but I think one needs to do some thinking on its role.
This, in my mind, leads to more concrete
application-oriented features, rather than
language-oriented ones.

Also, who knows, I might be unduly pessimistic. I
really wouldn't mind being proven wrong :-)

Best,
Thomas


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



More information about the erlang-questions mailing list