Author: Robert Hyatt
Date: 13:48:48 07/19/00
Go up one level in this thread
On July 19, 2000 at 14:47:43, Dann Corbit wrote:
>On July 19, 2000 at 13:53:28, Robert Hyatt wrote:
>[snip]
>>I'm not touching this. I lived about 30 miles north of a small town in
>>Mississippi named "Wiggins". Why do I mention this? Because a very educated
>>physicist lived there, and he made national news more than once with his
>>"perpetual motion machine". Which, as we all know is a machine that at _least_
>>breaks even in the energy used/energy produced. And most are trying to build
>>such a gadget that more than breaks even. _and_ violates the conservation of
>>matter/energy laws of physics.
>>
>>I don't believe it is possible to add eval at no cost. Some things might be
>>addable at minimal cost, of course. But many things have a _huge_ cost if
>>you want to know them.
>
>It's probably impossible to produce perpetual motion, but the patent office
>receives a huge number of such requests anyway.
>
>I think it may be possible that Amir is right. Consider if he already has
>something very expensive. Since he has to wait for that operation, he may be
>able to do all the eval he wants, if he can do it in parallel.
That is possible with more than one cpu, and a non-parallel tree search. But
it has been tried and found worthless for performance. Because there are many
cases where you don't have something (say chunks of eval) to do in parallel.
It works out much better to search different tree branches in parallel.
If we were talking about DB, your answer would be perfect. That is how they
did things. But with today's micros, you are talking about stuffing
instructions into the cpu's maw, and having it execute them. More eval --> more
instructions, and I don't see how to stuff more in unless you resort to tricks
like using mmx hardware, or fp hardware...
>
>Consider a matrix multiplication of a huge matrix. We also have to sort the
>data. Now, the matrix operation is O(N^3) {or even with Strassen's
>multiplication much higher than O(N^2)}. So we can sort with any slow technique
>we want and it will make no practical difference in the algorithm time at all,
>since the slower operation dominates.
Unless you do it on a single CPU. Then the time required is the sum of both,
plus a little for the complexities of trying to do both at the same time (ie
context switches or whatever).
>
>Perhaps that is what he means. He may have some incredibly expensive operation
>that is the real log jam. This slow operation means that "everything else is
>free."
I don't see how you can do this on a current micro. There is but one CPU that
is working on the code... Even if you have multiple cpus, they are doing the
same things to different parts of the tree...
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.