"Robert J. Bradbury" wrote:
> If they are taking an example out of the GRAPE book, they are designing the
> instructions for molecular modeling.
I read the IBM announcement on their web site, and a New York Times article that had a little more detail (can't vouch for its accuracy: http://www.nytimes.com/library/tech/99/12/biztech/articles/06blue.html) Neither piece had enough detail to really say what IBM's design plans are, but among the few facts I gleaned from there were:
The "instruction set" -- the total vocabulary of machine-language instructions a computer understands -- will number 57 for Blue Gene, compared with about 200 for most RISC machines. (NYT)
57 instructions is not a lot for a processor - Intel 486 has around 170, the IBM
390 (extreme CISC) has over 400. Of course these are general purpose processors,
and a special-purpose architecture like GRAPE probably has many fewer
instructions; however, I would expect many of those instructions to be relatively
complex to render in the hardware:
(From IBM's announcement):
"We call this new approach to computer architecture SMASH, which stands for Simple, Many and Self-Healing." The SMASH architecture differs from existing approaches in three ways: It dramatically simplifies the number of instructions carried out by each processor, allowing them to work faster and with significantly lower power and chip surface requirements (the traditional approach is to add complex features to gain performance); It will facilitate a massively parallel system capable of more than 8 million simultaneous threads of computation (compared to the maximum of 5000 threads today); It will make the computer self-stabilizing and self-healing -- automatically able to overcome failures of individual processors and computing threads.
To me, this implies that their approach to the hardware design is unit simplicity for raw speed, massive redundancy for fault tolerance and throughput. It seems to me they intend to handle the complexity at a higher level (in the software).
> They may have sat back after
> doing the molecular modeling stuff and said, ok, now how can
> we turn the machine into a good data-mining computer, or a
> good speech recognition processor, or a good image processor
> and added some more instructions to round things out.
Maybe, but I've seen nothing in the published material that says this is anything other than a general-purpose machine. But again, the facts in these pieces are somewhat meager. I'd like to get a peek at the instruction set if they ever deign to publish it.
But even if I'm right, the task of designing software to make full use of the machine's capabilities may be so daunting that no one else will want to take it on, effectively making it a single-purpose machine. And this is likely the only one they will build, like Deep Blue.
P.S. I apologize for my sloppy editing on my original post (which was truly my first post to this list).