CPU/Hardware optimized for Erlang

Thomas Lindgren <>
Tue Jul 26 12:34:12 CEST 2005



--- "Richard A. O'Keefe" <> wrote:

> Will Newton <> wrote:
>     > Also bear in mind that Intel chips are
> constrained
>     > very much by their legacy. 
> 
> But Thomas Lindgren <>
> doesn't think it's a handicap.
> 
> It is quite enough of a handicap for Transmeta to
> carve themselves
> a niche.  The Pentiums do run-time compilation of
> IA32 instructions
> to a sort of RISC with the compilation done in
> hardware a little at
> a time; the Transmeta machines did run-time
> compilation in software
> in great big gulps and were able to achieve
> near-Pentium speed at a
> MUCH lower cost in electrical power.

One should recall that Transmeta first wanted to
compete (in the x86 market!) on performance, then
switched to 'power' when raw performance wasn't
competitive. And ultimately seem to have failed once
Intel/AMD caught up.

Even IBM and PowerPC have abandoned the desktop now.
(Though, again, they seem to be doing interesting
application-oriented SOCs -- Cell, Blue Gene.)

The low-power market for general purpose desktops,
blades, etc. seems to be sewn up by the incumbents
today.

An "Erlang" processor would have to play a different
game compared to Transmeta, since you no longer can
sell to the same market (until MS ports Office). So,
what would induce a telecoms company to switch from
Sun/Solaris to Erlang processor/*? Or Google to switch
from Intel/Linux? Or Cisco to switch from ... Or an
embedded box maker to use Erlang processors instead of
brand X?

For scientific work, a business case is not necessary
of course. But in this case, what unique "Erlang"
features will provide the big edge in, say, power?
Multithreading + slow clock? Something else? And would
it beat an Erlang R10B port to a "similar but
non-Erlang" architecture? How sustainable is this
advantage over time? (Is the advantage inherent in
"Erlang" or something else?)

(I'm being the difficult skeptic here, but these
questions would ultimately have to be answered.)

It seems far more fruitful to consider application
areas instead.

> Complexity certainly has an effect on barriers to
> market entry.

Of course. However, in this case, I'd also say that
keeping backwards compatibility since 1979, with all
attendant complexity, has NOT proved to be a great
disadvantage regarding performance or power. (While
also being a big business advantage.)

> Intel machines are impressive, but basically, they
> are brick outhouses
> that fly only because they have extremely large jet
> engines strapped on.

... flying faster than the competition, using the same
or less fuel. And at cheaper prices, too :-)

>> [APZ]
>
> Reference?

See my other mail, or google "APZ Ericsson Review".

Best,
Thomas

PS. And I haven't even started on compatibility issues
with "software Erlang" yet :-)



		
____________________________________________________
Start your day with Yahoo! - make it your home page 
http://www.yahoo.com/r/hs 
 



More information about the erlang-questions mailing list