what is the oldest hardware when engines at GM level?

[Uri Blass]

I wonder what is the oldest hardware that there are chess engines that can use and are still at GM level at different time controls.

The intel 80386 is from 1985

Are there chess engines that can use this hardware and get GM strength level?
Can people modify stockfish to use this hardware(assuming it does not run today on 386) and if yes how many nodes per second it is going to search?

I also wonder what is the best time control against humans with slow hardware.
I suspect that longer time control for slow hardware help the engines today more than humans.

I know it was the opposite in the past and more time gave the advantage for humans but this was with bad search algorithms and bad evaluation and today engines use better search algorithms and better evaluation.

[MikeB]

I wonder what is the oldest hardware that there are chess engines that can use and are still at GM level at different time controls.

The intel 80386 is from 1985

Are there chess engines that can use this hardware and get GM strength level?
Can people modify stockfish to use this hardware(assuming it does not run today on 386) and if yes how many nodes per second it is going to search?

I also wonder what is the best time control against humans with slow hardware.
I suspect that longer time control for slow hardware help the engines today more than humans.

I know it was the opposite in the past and more time gave the advantage for humans but this was with bad search algorithms and bad evaluation and today engines use better search algorithms and better evaluation.

Too bad you didn't say say 'slowest' instead of 'oldest' - it would be easier to answer. You mentioned stockfish, based on all the testing I have done with stockfish as fewer nps, my estimate is that SF needs somewhere between 80K/nps to 150K/nps to be legitimate GM strength and that is probably only attainable with using at least 5 man EGTB as well. It could be at low as 40 to 50K nps, but it is for sure at the 150K/nps. The issue is how do you figure this out since so few GM's play any computer regulary? The best you can do is expolation from CCRL and SSDF testing - but that process rifed with assumptions that may not be true. So perhaps the proper is answer is "we don't know, but certainly anywhere from 45knps to 150Knps is as good as a guess as any"
Now the second part of your answer is what processors run Stockfish from 40 to 150K ? The raspberry 1.4GHz 64-bit quad-core processor that can run at level 60k/nps in single thread mode and over 200k/nps using all 4 cores. All of the intel processors that would be comparable to the Pi ( late 90's early 2000's), in most cases would not run a Windows OS that would be compatible with current Stockfish, you would probably need to put an older version of linux on one these machines and hope for the best. But yes, I believe a computer with a CPU from the late 90's or early 2000's , running Stockfish , would probably be grandmaster strength. (And ,obviously, I believe the 80386 from 1985 would not be GM strength, even with Stockfish).

At the end of the day, one also has to consider the MIPS of the CPU and the CPUs of 1985 are way down the list for the average consumer.

https://en.wikipedia.org/wiki/Instructions_per_second

[Uri Blass]

I wonder what is the oldest hardware that there are chess engines that can use and are still at GM level at different time controls.

The intel 80386 is from 1985

Are there chess engines that can use this hardware and get GM strength level?
Can people modify stockfish to use this hardware(assuming it does not run today on 386) and if yes how many nodes per second it is going to search?

I also wonder what is the best time control against humans with slow hardware.
I suspect that longer time control for slow hardware help the engines today more than humans.

I know it was the opposite in the past and more time gave the advantage for humans but this was with bad search algorithms and bad evaluation and today engines use better search algorithms and better evaluation.

Too bad you didn't say say 'slowest' instead of 'oldest' - it would be easier to answer. You mentioned stockfish, based on all the testing I have done with stockfish as fewer nps, my estimate is that SF needs somewhere between 80K/nps to 150K/nps to be legitimate GM strength and that is probably only attainable with using at least 5 man EGTB as well. It could be at low as 40 to 50K nps, but it is for sure at the 150K/nps. The issue is how do you figure this out since so few GM's play any computer regulary? The best you can do is expolation from CCRL and SSDF testing - but that process rifed with assumptions that may not be true. So perhaps the proper is answer is "we don't know, but certainly anywhere from 45knps to 150Knps is as good as a guess as any"
Now the second part of your answer is what processors run Stockfish from 40 to 150K ? The raspberry 1.4GHz 64-bit quad-core processor that can run at level 60k/nps in single thread mode and over 200k/nps using all 4 cores. All of the intel processors that would be comparable to the Pi ( late 90's early 2000's), in most cases would not run a Windows OS that would be compatible with current Stockfish, you would probably need to put an older version of linux on one these machines and hope for the best. But yes, I believe a computer with a CPU from the late 90's or early 2000's , running Stockfish , would probably be grandmaster strength. (And ,obviously, I believe the 80386 from 1985 would not be GM strength, even with Stockfish).

At the end of the day, one also has to consider the MIPS of the CPU and the CPUs of 1985 are way down the list for the average consumer.

https://en.wikipedia.org/wiki/Instructions_per_second

I believe that stockfish clearly needs less nodes based on the data that I read for GM strength at long time control.

http://www.talkchess.com/forum3/viewtop ... 4&start=20

Stockfish 2020-04-29 (1 kn/s) vs Fruit 2.1 (2800+)

stockfish won 70.5-29.5


Note that stockfish did not use exactly 1 kn/s but it was ponder off games and average of 187 kn/move
and stockfish clearly can use 180 kn/move at long time control when it use 1 kn/s.

[MikeB]

I wonder what is the oldest hardware that there are chess engines that can use and are still at GM level at different time controls.

The intel 80386 is from 1985

Are there chess engines that can use this hardware and get GM strength level?
Can people modify stockfish to use this hardware(assuming it does not run today on 386) and if yes how many nodes per second it is going to search?

I also wonder what is the best time control against humans with slow hardware.
I suspect that longer time control for slow hardware help the engines today more than humans.

I know it was the opposite in the past and more time gave the advantage for humans but this was with bad search algorithms and bad evaluation and today engines use better search algorithms and better evaluation.

Too bad you didn't say say 'slowest' instead of 'oldest' - it would be easier to answer. You mentioned stockfish, based on all the testing I have done with stockfish as fewer nps, my estimate is that SF needs somewhere between 80K/nps to 150K/nps to be legitimate GM strength and that is probably only attainable with using at least 5 man EGTB as well. It could be at low as 40 to 50K nps, but it is for sure at the 150K/nps. The issue is how do you figure this out since so few GM's play any computer regulary? The best you can do is expolation from CCRL and SSDF testing - but that process rifed with assumptions that may not be true. So perhaps the proper is answer is "we don't know, but certainly anywhere from 45knps to 150Knps is as good as a guess as any"
Now the second part of your answer is what processors run Stockfish from 40 to 150K ? The raspberry 1.4GHz 64-bit quad-core processor that can run at level 60k/nps in single thread mode and over 200k/nps using all 4 cores. All of the intel processors that would be comparable to the Pi ( late 90's early 2000's), in most cases would not run a Windows OS that would be compatible with current Stockfish, you would probably need to put an older version of linux on one these machines and hope for the best. But yes, I believe a computer with a CPU from the late 90's or early 2000's , running Stockfish , would probably be grandmaster strength. (And ,obviously, I believe the 80386 from 1985 would not be GM strength, even with Stockfish).

At the end of the day, one also has to consider the MIPS of the CPU and the CPUs of 1985 are way down the list for the average consumer.

https://en.wikipedia.org/wiki/Instructions_per_second

I believe that stockfish clearly needs less nodes based on the data that I read for GM strength at long time control.

http://www.talkchess.com/forum3/viewtop ... 4&start=20

Stockfish 2020-04-29 (1 kn/s) vs Fruit 2.1 (2800+)

stockfish won 70.5-29.5


Note that stockfish did not use exactly 1 kn/s but it was ponder off games and average of 187 kn/move
and stockfish clearly can use 180 kn/move at long time control when it use 1 kn/s.

Interesting , it is true that Stockfish's Elo really ramps up as more nodes are played . I never tested at LTC such as you referenced. All of my tests were at much faster times controls. I do know that SF using one thread can beat many engines using 64 threads and it's not even close. Engines rated 3000 or less with one thread .

[mehmet123]

I believe that stockfish clearly needs less nodes based on the data that I read for GM strength at long time control.

According to my tests:

The human elo of Stockfish with 1 kn/s is 2612.

The human elo of Stockfish with 10 kn/s is 2950.

The human elo of Stockfish with 20 kn/s is 3035.

http://talkchess.com/forum3/viewtopic.php?f=6&t=72485


Stockfish benchmark of Intel Pentium I (150MHz) is 5626 nodes/s

https://sites.google.com/site/computers ... benchmarks

Double speed gains ~100 elo for Stockfish at low nodes.

The human elo of Stockfish with 5 kn/s is ~2850 (Intel Pentium I 150MHz)

The speed of Stockfish at Intel Pentium I 150MHz would be much higher if Stockfish codes was adapted to older softwares.

[Colin-G]

Apart from the nps, what about the hash value? How does this affect the elo?
Very old computers did not have much ram and could only have a small hash value.

[mehmet123]

Apart from the nps, what about the hash value? How does this affect the elo?
Very old computers did not have much ram and could only have a small hash value.

At old computers the hash values of chess engines is low but the speed of chess engines at old cpu's very low too.

Less hash than required may cause negative impact but more hash than required may cause more negative impact too.

https://sites.google.com/site/computers ... en-vs-5men

[Vinvin]

I wonder what is the oldest hardware that there are chess engines that can use and are still at GM level at different time controls.

The intel 80386 is from 1985

Are there chess engines that can use this hardware and get GM strength level?
Can people modify stockfish to use this hardware(assuming it does not run today on 386) and if yes how many nodes per second it is going to search?

Some memory about Stockfish on old CPUs here : http://talkchess.com/forum3/viewtopic.p ... 51#p821251

I also wonder what is the best time control against humans with slow hardware.
I suspect that longer time control for slow hardware help the engines today more than humans.

I know it was the opposite in the past and more time gave the advantage for humans but this was with bad search algorithms and bad evaluation and today engines use better search algorithms and better evaluation.

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

[mehmet123]

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.

[Uri Blass]

I wonder what is the oldest hardware that there are chess engines that can use and are still at GM level at different time controls.

The intel 80386 is from 1985

Are there chess engines that can use this hardware and get GM strength level?
Can people modify stockfish to use this hardware(assuming it does not run today on 386) and if yes how many nodes per second it is going to search?

Some memory about Stockfish on old CPUs here : http://talkchess.com/forum3/viewtopic.p ... 51#p821251

I also wonder what is the best time control against humans with slow hardware.
I suspect that longer time control for slow hardware help the engines today more than humans.

I know it was the opposite in the past and more time gave the advantage for humans but this was with bad search algorithms and bad evaluation and today engines use better search algorithms and better evaluation.

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.