Author: Vincent Diepeveen
Date: 07:25:32 10/28/02
Go up one level in this thread
On October 27, 2002 at 19:24:00, Bob Durrett wrote:
>On October 27, 2002 at 16:03:02, Vincent Diepeveen wrote:
>
>>On October 27, 2002 at 10:42:05, Bob Durrett wrote:
>>
>>are you looking for $0.01 processors which do near nothing, or
>>are you looking for good processors that can compete at that time
>>with processors of that time?
>>
>>>How Small Can a Microprocessor Get?
>>>
>>>Maybe the question should be: "How many microprocessors per square inch on a
>>>wafer will be doable in the future?
>>>
>>>Part of the my problem is that I do not know how many electronic components
>>>[resistive, transistor, etc.] are required to make a microprocessor. How many?
>>>
>>>
>>>Bob D.
>
>Vincent:
>
>My interest was for chess engine applications.
>
>Intuitively, is seems that one million "$0.01 processors which do near nothing"
>might actually do more, collectively, than one modern microprocessor.
I will not directly disagree, but there are some issues when you
press for example 100 processors from 100Mhz instead of a 1.2Ghz McKinley
(which uses pretty much transistors, even 3MB L3 cache on chip).
In theory the 100 x100Mhz is faster, and when pressed for the same
price i can imagine you prefer those 100 chips. However then you need
a couple of millions of dollars to build an actual system from it to
quickly connect the cpu's.
the next example shows the actual problem:
1 processor K7 system - $1,000
2 processor K7 system - $2,000
8 processor Xeon - $100,000
16 processor Alpha - $10,000,000
512 processor SGI origin 3800 - ==> a lot more than $10MLN
So you go to a system like this, but then with say 100 cheap $0.01
processors and then to pay a billion to make a machine with 10000
processors is pretty stupid IMHO.
>Since I see advancement in two fronts, complexity and speed, I would expect to
>see blindingly fast "$0.01 processors which do near nothing" when compared to
>today's computers [depending on how far out into the future].
>
>Similarly, I would expect to see many "tiny" microprocessors which do as much as
>a current-day Intel Pentium.
In theory it is possible, but i must note that i have put years of work
in DIEP in order to run parallel and another few months now to get it to
work on this machine very well:
http://www.sara.nl/Customer/systems/sgi3800/overview.html
Of course a supercomputer offers so much more than probably your idea
is now. But in order to let a chessprogram work well you need at least
2 things which are expensive to make:
a) each processor needs its own RAM. Even if it is 128 MB, it is
still not easy to do this in combination with the next thing
b) you need an OS that can handle so many processors, don't say linux
can do it. That's not the 100% truth.
c) each processor needs to communicate to other processors.
this is the expensive part of course of the machine.
>I don't know where the tradeoff will be for future chess-playing machines. But
>perhaps millions of relatively simple processors would be the choice of chess
>engine programmers of the future?
you want a machine with millions of processors? How much is that going
to cost? Trillions?
A crucial thing for chess is a fast communication speed between nodes. As
you can see the Teras machine delivers up to a TERA byte a second.
That is not so easy to deliver. That costs $$$$$$$$$.
Yes with that many digits... :)
>With your "future hat" on, and your magic crystal ball, what do you see?
>Bob D.
I am not hardware expert but i see very clearly that software runs behind
more than the hardware does. If you see how well those SGI system are,
then i am really amazed.
However it requires of course special programming to get something
parallel to work in such a way that you do not poll continuesly and
preferably do not write in too many cache lines which are hosted/allocated
at other nodes.
What i see too is that the power that most cpu's require now, that it
might give a big problem in the future. I do not know how i can cool
my dual k7 without it producing a lot of sound.
I was looking for watercooling now, but the big fans that are used there
are spinning at 90KM/hour, or in short, producing a kind of 'car' sound :)
Best regards,
Vincent
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.