Author: Christophe Theron
Date: 12:27:22 03/28/01
Go up one level in this thread
On March 28, 2001 at 13:58:41, Dann Corbit wrote:
>On March 28, 2001 at 10:20:14, Christophe Theron wrote:
>[snip]
>>My first intention was simply to figure out what was a dual worth really.
>>
>>I did these computations for myself, then decided to publish it because I think
>>it's good informations for the CCC forum.
>>
>>It is true that I believe that adding chess knowledge in a program is more
>>interesting than making a program SMP, and that there is more evolution
>>potential in working on chess knowledge.
>>
>>And it is clear for me that multiprocessor computers will not be the mainstream
>>in the foreseeable future.
>>
>>I know that some will disagree, I can even guess who.
>>
>>These people are just ignoring facts of society like Internet mobile phones,
>>PDAs, tablet PCs and the like.
>
>I think you are wrong about duals. I imagine you may have guessed it was me.
No, I didn't have you in mind actually...
>The reason I think so is that a dual (or quad -- whatever) is the cheapest way
>to get high throughput. As Eugene pointed out, a dual with an effective
>througput of 1200 MHz is better than a single CPU of 1200 MHz as well. We could
>just as well say, "Do not bother to upgrade from 500 MHz to 1GHz, since it will
>only add 50 ELO." This statement is completely true. But having a faster
>machine is very valuable for many other tasks as well.
You are partly right it was not the point of my original post, but we can talk
about it if you want...
> This machine is 960 MHz,
>and with 512 megs of ram and 100 gigs of hard disk. And yet I have projects
>that take over 1/2 hour to compile. At the cost of my time, the machine pays
>for itself over a slower one in a ridiculously short period of time.
The biggest project I have to compile takes one minute on my K6-2 450. And it's
because it is a slow cross compiler at work.
Of course some people can have huge projects (maybe Eugene has even bigger
projects?) but how many people in the world have those needs?
Then OK, for them a dual is maybe a good idea. But that is going to be 1 PC out
of 1000.
It's not an argument in favor of multiple CPUs computer flooding the world (not
even the "free" world as Bob says).
>The future of fast computing will be multiple CPU's.
You are right, but the future is not fast computing.
The future is mobile computing.
That's why I don't expect SMP to rule, that's why I expect SMP to never become
the mainstream.
> It is absurdly easier to
>get a lot of computing power by SMP compared to increasing the frequency. And
>when the CPU's near 2GHz, and the traces .1 micron, how much will it cost to
>increase again by shrinking and speedup?
Yes.
>Calculate the cost of a multiple CPU machine, and then calculate the cost of a
>single CPU machine with exactly the same throughput. You will find that the
>multiple CPU machines are a bargain.
Only for speeds above the speeds of the fastest single processor.
>A workman is as good as his tools. Let's try an exaggeration. There are two
>equally skilled workers. One has a 286 with one meg of ram, the other has a
>dual CPU Athlon running at 900 MHz per CPU. Which one will get quality work
>done first? (Remember, I said that they were equally skilled).
Let me think a while...
My answer is that your skills at asking questions with a forced answer are not
very high. You could do it in a more subtle way... :)
>Now, any tool advantage one worker has over another will translate into more
>productivity. The cost benefit line is something for managers to calculate as
>to how far they are willing to go in beefing up the capital used by the work
>force. But if you are going to outfit people with high quality equipment, it
>makes great sense to use multiple CPU's.
Let me guess... You live in the same country as Bob, isn't it? ;)
>All that having been said, you are clearly right about CPU horsepower. The
>difference added by doubling the speed is only 53 ELO. So, if you take an old
>Pentium Pro 200 and put a chess program on it, and play that same program
>against a dual with 1GHz CPU's, the dual won't win 1000 out of 1000 games.
>
>The ELO difference would be roughly:
>(2000/200)*50=500 ELO and so the Pentium Pro will win 5% of the points.
>
>However, if you have a quad, it will only be .3% of the points (roughly
>speaking).
>
>So, your point is well founded. If you spend a bazillion dollars to get the
>"ultimate" chess machine, it won't dominate like some might imagine. But you
>could spend half a bazillion and get the same horsepower if you bought a 2 CPU
>machine instead.
>;-)
I have given some references for people considering to buy a dual.
I did not tell them "don't buy a dual". I have told them: "here is what you get
with a dual...".
These people will make their choices themselves...
Christophe
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.