Roy Schestowitz wrote:
> Folding@Home on the PS3: the Cure@PS3 project
>
> ,----[ Quote ]
> | Now in 2006, we are looking forward to another major advance in
> | capabilities. This advance utilizes the new Cell processor in Sony's
> | PLAYSTATION 3 (PS3) to achieve performance previously only possible
> | on supercomputers. With this new technology (as well as new advances
> | with GPUs), we will likely be able to attain performance on the 100
> | gigaflop scale per computer.
> `----
>
> http://folding.stanford.edu/FAQ-PS3.html
>
> The PS3 is running Linux.
I heard a talk recently by one of the guys who designed the IEEE
standard for floating point back in the 1980's. He said the hardware
implementation was not trivial and the designers complained about it,
but it was definitely the best way to ensure accuracy. I'll quote
his abstract, he says it better:
Quote:
-------------
W. Kahan (Mathematics and EECS)
"Your undebuggable numerical program"
Abstract:
Upon producers of computer hardware, compilers and program
development systems, the economic impact of our scientific and
engineering computations has dwindled over the past two decades
to near negligibility compared with games, other entertainment
and communications. Consequently atrophy threatens hard-won
features built into floating-point arithmetic in the 1980s but
almost entirely ignored and therefore inaccessible through the
most popular programming languages like Matlab and Java. (An
exception is fully implemented C99, if you can find it.) Much
of the computing industry is preoccupied with speed to the near
exclusion of everything else. Most scientists and engineers are
unaware of what we are losing, much less of what we can do to
inhibit the loss.
Some of these issues will be illustrated by Matlab examples,
among them examples of programs that are becoming impossible to
debug without help from hardware, compilers and environments
for program development. For more details see items posted on
my web page http://www.cs.berkeley.edu/~wkahan/...,
starting with shorter items like $/Mind1ess.pdf , /SPECbnch.pdf ,
/Grail.pdf and /ARITH\_17.pdf $, and moving to pp. 53-55 of
$/Mindless.pdf$ for specific recommendations of tools that would
ease the debugging of roundoff though nothing will make it easy.}
--------------
His web site is interesting.
|
|
