The future of computer chess

[Uri Blass]

The point was: is the huge progress in computerchess part of hardware or software development.

With the 16 bit cpus the adress range was much bigger.
Suddenly not only 32 or with bank switching 96 kb was usable. But lot of memory more.

Suddenly we were in megabyte range.
That made hash tables possibe.
The engine could grow and use heavier evaluations and lot of new stuff.

So was this a feature of software or more a feature of hardware change from 8 bit adress room to 16 bit addresses ?

With the new 68000 cpu richard lang was capable to program a very strong engine in assembler.
And with the use of 68020 or 68030 and the increase of the mhz from 12 to 33 or 66 or even 133 mhz
the software was made faster.


The next big step was that the pcs made progress from a few mhz into much more mhz.
We remember 8086, 286, 386, 486-33, 486-66 and more. Then came pentiums much more speed.
And one day we had cpus with 2 cores, then with 4 cores and the race continues,

Today we have very fast cpus with lot memory, high speed and several cores,


So how big was the software progress that was not due to big hardware progress ?

Can somebody create a 6502 engine that beats ed schroeder , dave kittinger ?

People have no motivation to do something that nobody or almost nobody is going to use.

If there is a big prize money for the best 6502 engine in one year from the announcement (let say some rich person donate a prize of a million dollar for it) and nobody get something that is significantly stronger than what we have now then you have some evidence that it is really hard to get a significant improvement with the knowledge of today but today we have no evidence for this claim.

It is possible that only changing weights in the evaluation of what Ed did at that time may give 100 elo improvement and at that time it was hard to test but today I guess that with some work we can emulate Ed's engine in new hardware except being something like 1000 times faster that mean that with new hardware we can test changing weights 1000 times faster relative to what Ed could.

[Viz]

I mean pure search tuning in stockfish - aka just constants changes without any other code changes was 15 elo or so when first implemented somewhat properly.
And this is at times of no less than smth like stockfish 14 IIRC, which was tuned and retuned at smaller sets of data, yet it was so proficient.
Heck, you can even look there - https://github.com/official-stockfish/S ... XInTheDark this is more or less constant gains with nothing but numbers optimization for longer time controls (while regressing at shorter ones).
Sure w/e was present in 1960 had much fewer parameters but let me tell you if they were anywhere near close to optimum it would've been a miracle. Modern optimization techniques can squeeze a lot of elo without even adding any code more or less.

[towforce]

People have no motivation to do something that nobody or almost nobody is going to use.

This.

I am sure that if the resource were invested in it, better engines could be made for 1970s 8-bit CPUs (even if just by tuning) - but nobody is going to make that investment today.

Regarding address space: I don't have expert knowledge on the subject, but here's some simple maths:

2^8 = 256 (so 1 byte can address 256 different memory locations). That's not enough for a good chess program (you'd do well to run ANY chess program in 256 bytes for both program and memory!).

2^16 = 65536 (= "64k")

So 16 bit addressing would have been used. So fetching a memory address from memory (which, in the 1970s and early 1980s, was on a different chip!) would have entailed locking the computer bus for the CPU's usage, then taking multiple clock/bus cycles to fetch the contents of the memory at the 2-byte address!

If your computer had a 16-bit CPU and a 16 bit bus, I imagine that the process of fetching data (or an instruction) from a 16 bit address would take far fewer clock cycles. CPUs improved in many other ways as well.

The 8-bit CPUs were an absolute revolution: they enabled "ordinary people" to have their own computers - which was phenomenal! But these "ordinary people" weren't getting supercomputers.

[mclane]

Thats why the old 8 bit homecomputers have 64 kb memory.
2 to the power of 16 was also in our CD players.

All cds you have use only 65536 levels for the sound.
From the most silent situation in a concert to the loudest in a rock n roll concert. Only 65536 levels of how loud or silent something is.

The human ear is much better then 65536 . Thats why todays music lovers use flac24 files.

The the most silent to the most loudest sound has 16.777.216 levels and comes closer to the human ear.
2 working human ears can find out if music is 65536 or 16.7 million levels .

I can identify the difference because when it is flac24 the hairs in my neck go upwards and the emotion in the music is big, like in a live concert while flac16 sound is technically perfect but almost without emotion.


The 16 bit homecomputers like amiga500 or atari ST had only 10 mhz cpu but megabytes of ram instead of 64 kb.
This made it possible to use different data structures in chess engines.

[jefk]

with sufficiently long time controls, balanced openings will only lead to draws for the strongest
engines. Even unbalanced (but not biased) openings will lead more and more to draws.

That's why in the higher leagues of TCEC some 'advantageous' (biased) lines are chosen;
then ofcourse a specific engine should play such line with both colors.

https://wiki.chessdom.org/Openings_FAQ