what is the oldest hardware when engines at GM level?

[lkaufman]

1 kn/s at 386-33 Mhz
2 kn/s at 486-33 Mhz
6 kn/s Pentium 75 Mhz
http://talkchess.com/forum3/viewtopic.p ... 51#p821251

For long time control matches:

Stockfish 11 at 386-33 Mhz can fight with 2650- 2700 elo GM's with a strong opening book.
Stockfish 11 at 486-33 Mhz can fight with 2750 -2800 elo GM's with a strong opening book.
Stockfish 11 at Pentium 75 Mhz can beat Magnus Carlsen with a strong opening book.

Below about ten ply Komodo becomes stronger than Stockfish (at same ply or same time). I don't know how fast the time control would have to be for Stockfish 11 on a 386 or 486 to drop below 10 ply, but below whatever that time limit is, Komodo would be the best engine on that old hardware in theory, assuming they both run properly on it.
In blitz Rexchess was already GM level about 1990 on my 25 Mhz 486.

[Chessqueen]

1 kn/s at 386-33 Mhz
2 kn/s at 486-33 Mhz
6 kn/s Pentium 75 Mhz
http://talkchess.com/forum3/viewtopic.p ... 51#p821251

For long time control matches:

Stockfish 11 at 386-33 Mhz can fight with 2650- 2700 elo GM's with a strong opening book.
Stockfish 11 at 486-33 Mhz can fight with 2750 -2800 elo GM's with a strong opening book.
Stockfish 11 at Pentium 75 Mhz can beat Magnus Carlsen with a strong opening book.

Below about ten ply Komodo becomes stronger than Stockfish (at same ply or same time). I don't know how fast the time control would have to be for Stockfish 11 on a 386 or 486 to drop below 10 ply, but below whatever that time limit is, Komodo would be the best engine on that old hardware in theory, assuming they both run properly on it.
In blitz Rexchess was already GM level about 1990 on my 25 Mhz 486.

Now that would really be interesting to see Komodo on a 386 beating a strong GM like Lenderman. I remember back in 1998, 22 years ago when Rebel 8.0 beast Vishy Anand using A KryoTech Cool K6-2 (based on a 450MHz KryoCooled AMD-K6-2 processor),supplied by Kryotech 128 Mb EDO Ram based on the final score I would say that Rebel could prpobably had drawn Anand using a 386 33 Mhz
https://www.rebel.nl/anand.htm

[Ovyron]

How do we know that engines that play stronger against other engines also play stronger against humans?

What if Hakkapeliitta TCEC v2 64-bit is the strongest possible entity against humans achieving 49 wins out of 50 games against strongest human while Stockfish only gets 19 wins out of 20 games or something?

Knight odds will not help answering those questions

[lkaufman]

How do we know that engines that play stronger against other engines also play stronger against humans?

What if Hakkapeliitta TCEC v2 64-bit is the strongest possible entity against humans achieving 49 wins out of 50 games against strongest human while Stockfish only gets 19 wins out of 20 games or something?

Knight odds will not help answering those questions

Unfortunately the opening book used has more to do with which engine avoids the occasional draw with superGM more often (as Black) than the engine itself. The only somewhat meaningful test of which engine is better at beating humans would be chess960, but in that case the winning percentage of the top engines would be so close to 100% that we could never get an answer. We can only hope to get an answer to which engine is better at beating humans when handicapped either by huge time odds on one thread or by material/moves. Of course the answers may be different. I suppose the best way to do time odds is to have the engine do a fixed movetime, since if you use super fast time controls like game in one second + .01" inc. the results depend too much on the GUI, whether it's on the internet or on the actual machine, overhead settings, etc. This is something we could actually test with Komodo vs. GMs. But even here there is a small problem, the engines aren't fully "honest" about movetime.

[Ozymandias]

In blitz Rexchess was already GM level about 1990 on my 25 Mhz 486.

2500 Elo GM? What about standard FIDE time control, when would you say that happened and what engine achieved the feat?

[lkaufman]

In blitz Rexchess was already GM level about 1990 on my 25 Mhz 486.

2500 Elo GM? What about standard FIDE time control, when would you say that happened and what engine achieved the feat?

Yes, we played in a WBCA blitz tournament and made an even or plus score against GMs, all in the 2500s FIDE or so I believe. In Rapid our socrates program was maybe 40% or so vs. 2600 level GMs around 1991-1992, maybe on Pentium by then. I suppose it was mid 1990s when parity was reached at standard time control, I can't be too precise on that point.

[Ozymandias]

In blitz Rexchess was already GM level about 1990 on my 25 Mhz 486.

2500 Elo GM? What about standard FIDE time control, when would you say that happened and what engine achieved the feat?

Yes, we played in a WBCA blitz tournament and made an even or plus score against GMs, all in the 2500s FIDE or so I believe. In Rapid our socrates program was maybe 40% or so vs. 2600 level GMs around 1991-1992, maybe on Pentium by then. I suppose it was mid 1990s when parity was reached at standard time control, I can't be too precise on that point.

Nice, thx

[lkaufman]

How do we know that engines that play stronger against other engines also play stronger against humans?

What if Hakkapeliitta TCEC v2 64-bit is the strongest possible entity against humans achieving 49 wins out of 50 games against strongest human while Stockfish only gets 19 wins out of 20 games or something?

Knight odds will not help answering those questions

I ran some quick tests which indicate that on one thread on a fairly fast machine, the break-even point between latest Komodo and Stockfish 11 is about 15 ms movetime (after making a change to make Komodo observe the time limit more accurately, to be fair). At this time limit Komodo averaged 11.4 ply, Stockfish 11.6 ply. This is around the level that I think would give strong grandmasters (perhaps not the very top ones) a fair battle at 15' + 10" Rapid. So it would actually be interesting to have both Stockfish and Komodo play against strong GMs in Rapid where SF and K each had to limit to 15 ms movetime. There would still be the question of what opening books (if any?) to use for such an event.

[Ovyron]

Unfortunately the opening book used has more to do with which engine avoids the occasional draw with superGM more often (as Black) than the engine itself.

Can someone create a book that beats the strongest human 100 games out of 100?

The only somewhat meaningful test of which engine is better at beating humans would be chess960

Still not chess, not any better than knight odds! It's like we can't answer the question and we create unrelated scenarios that answer other questions. The only way to answer chess questions is by playing chess (in the same conditions that were used to rank the human player.)

I ran some quick tests which indicate that on one thread on a fairly fast machine, the break-even point between latest Komodo and Stockfish 11 is about 15 ms movetime (after making a change to make Komodo observe the time limit more accurately, to be fair). At this time limit Komodo averaged 11.4 ply, Stockfish 11.6 ply. This is around the level that I think would give strong grandmasters (perhaps not the very top ones) a fair battle at 15' + 10" Rapid.

Again, the strength of engines are being used to make guesses about the strength of humans. I don't think any of this has been really tested, like, get a human to play a bunch of 3000 elo engines and a bunch of 3100 engines and see if any difference in strength can be measured. Perhaps the 3000 elo engines are 3000 elo, but the 3400 elo engines are really 3040 against humans (perhaps the 3000 elo engines perform better than the 3100 ones!), and no amount of engine v engine or human with odds/960 stuff would get us closer to the truth.

[Vinvin]

I'd say, the fastest is always the best : computer (even with a 8086 CPU) will always win a game in 1 second.

humans cannot play in 1 second without losing on time(I suspect that same for old computers but humans lose on time faster) but
I talk about time control that humans can play without losing on time.

Ok, so let's say 30 seconds or 1 minute.