X-Message-Number: 12937
From: 
Date: Tue, 14 Dec 1999 11:18:57 +1000
Subject: Blue Gene

Just to clarify the issue of Blue Gene. (John Clark has covered most of this
already but it seems to have fallen on deaf ears) Thomas Donaldson seems to

think that this is no technological breakthrough. Well it is certainly not a new

technology, no chemical/molecular/optical blah blahs. But this is its strength -
it can be done, it is not likely to disappear from the scene never to be heard
of again (with further investigation revealing a insurmountable unforseen
problem encountered).

There are a number of new challenges to producing this beast. The processors
will all be running at 1 GFLOP. Intel Processors do not even come close to this

- don't count MHz it is misleading. The processors will be designed to run a new
form of instruction set. The new instruction set can be thought of as being to
RISC, as RISC is to CISC i.e. a very stripped down and highly optimised set of
instructions and much less of them than in a RISC computer.


Each chip will have 32 processors (on the same chip) each with the capability of
1 GFLOP Each processor will have all its memory requirements on the same chip
running at full clock speed. Basically IBM are going to be into the wafer
business making chips that consist of an entire wafer. The speed of the comms
between the chips internally and externally will be amazing with one article
describing it something along the lines: "it will be able to take the entire
contents of the Internet and spit it out in less than a second".

The failover and redundancy is going to be very important and IBM are planning
to take an existing technology and adapt it and even improve it for this

machine. The floor space of the computer used to beat Kasparov in chess was also
quoted. I cannot remember if this was in feet or meters. Lets assume it was
metres. It took 8000 sqm and the floor space for this new machine will be 2000

sqm. So it will run 500 times as fast and take a quarter of the floor space. The
power requirements and cooling requirements will also be less.

The task of this machine will be to try and work out how a protein with 300
nucleotides folds. The article suggests that typical proteins are 600

nucleotides long but that this should still produce useful results. To calculate
this folding of 300 nucleotides will take this computer a year of constant
calculation. At a Quadrillion calculations per second, that works out at just
over 31.5 Hexillion instructions or in another way:

31 500 000 000 000 000 000 000 instructions.


I am not sure but I suspect that the protein folding problem is exponential so a
600 nucleotide long protein may still be impossible for a very long time. It is
still worth remembering that with the existing technology this problem would
take 500 years to calculate so nothing is definitely impossible.

As for a a cover story, I think this is a bit unlikely. It is going to be tied
up for a year and is going to be open to public scrutiny. If it were to be used

in AI then that would also be something for IBM to show off. Another possibility

would be decryption but this again would be something to show off. Basically IBM
saw the benefit they got in terms of recognition after they beat Kasparov in
chess and they want more. If they can work out how to fold a protein, they will
be in the news and people will talk about it especially if it saves lives.

The U$100 Million is for the research project, not for the computer itself. If

it is built it could cost significantly more but then one million processors for

U$100 Million is a bit unrealistic. A U$1 billion price tag would probably still
justify it being built. Although only one small protein fold can be calculated

in a year, the lessons learned could enable new techniques to be developed to do
it a simpler way. Also a single protein fold might be enough to crack something
like Aids which would easily pay for the computer many times over.

For cryonics, it's benefit is in 2 areas. We will benefit by the possibility of
better health and hopefully thereby increased longevity. This assumes that the

protein folding leads to new medical techniques to combat illness and maybe even

ageing. It is also useful because if we can fully understand protein folding and
large proteins, we can consider how to create a nanomachine by creating a chain
of nucleotides that will fold themselves into a fully working nanomachine. This

would be a long way off if it is even possible but at least IBM is moving in the
right direction.

It is also worth remembering that if IBM is successful in this, then they will
need a new challenge and then another etc. etc. If they keep focusing in this
direction, it certainly will work in our favour. As George Smith effectively
said, it costs us nothing but we will probably benefit. Multiple CPUs with
memory on a single chip may make it to the desktop PC and what is learnt by IBM
may help in this. I feel that IBM may just deliver this beast on time and to
spec so lets not bemoan their efforts.

Take Care

Chris

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=12937