22 Oct
2006
22 Oct
'06
4:10 p.m.
From: "Chuck Guzis" <cclist at
sydex.com>
On 22 Oct 2006 at 11:21, dwight elvey wrote:
Hi Chuck
Sure a machine the size of a Cray should do faster but the stack
machine I was using was created in an ASIC with less then 4000
gates. I was comparing to traditional micro's of the day.
Take any machine from the 1990's and write a
routine that
can sort 1K 16 bit integers in 19.2ms. That was done on a
stack machine running with a 4 MHz clock!
ANY machine? I daresay the non-stack Cray C-90 could do considerably
better than that!
My point is that in the somewhat rarefied supercomputer world where
computational performance is pretty much the entire game, not ONE
machine deployed as far as I know was a stack machine. It would seem
that if a stack machine had it all over memory-register systems
performance-wise that you'd find at least a single instance of one in
that market.
...and how does one efficiently do large vector
arithmetic on a stack
machine?
Most of the stack machines I've seen were not optimized to deal
with large vectors. This isn't to say that if the design was targetting
a vector machine that it would be fast.
Over the years, several machines have been created that ran quite
fast. A variation of the NC4000 made by Rockwell ( I think called the
RT2000 ) was often used on DSP excelerator cards. This considering
that it wasn't, as you might say, optimized for vectors and had no
MAC. Todays DSP's could beat it but at the time is was one of the
fastest setups.
Most of the reason for lack of success is advertizing and manufacture
labling. The other is that most common high level languages have
grown up around the register based machines. They just work
better on these machines.
Dwight
Cheers,
Chuck
_________________________________________________________________
All-in-one security and maintenance for your PC. Get a free 90-day trial!
http://clk.atdmt.com/MSN/go/msnnkwlo0050000002msn/direct/01/?href=http://ww…