[Oldies] What's the speed of a recent SF on a 80486-33 MHz ?

[Vinvin]

The question arose in the TCEC chat.

Numbers for 25 or 50 MHz is valid too !

Real numbers would be very nice.
Close approximations would be already informative ...


Second question : what would be the rating of a strong engine (in 2016) on such computer ?

Thanks !

[MikeB]

The question arose in the TCEC chat.

Numbers for 25 or 50 MHz is valid too !

Real numbers would be very nice.
Close approximations would be already informative ...


Second question : what would be the rating of a strong engine (in 2016) on such computer ?

Thanks !

30 MIPS is around a 33 MHz 80486 - was in production for PC's to the mid 90's. Crafty in the mid 90's had about 10K.nps on machine that was about t 4x faster than the 33Mhz 80486 . Crafty today on a decent machine does 6M/nps. I have compiled some of the older versions , using new compilers which distorts the results - but an old version of Crafty will run at least 3 to 4M/nps - some some of the increase is software, some of the increase is compiler improvements and a big chunk is related directly the CPU. So doing the rough math in single core mode, SF would be 2400x slower on a 33 MHz 80486. This is all on the back of a napkin, somebody else may spend more time on this and give you a better answer or an actual # which I would defer to.

So maybe SF would 1000 nodes per second. If you believe doubling speed results in ~50 ELO gain (some do/some don't) that would translate to a rating about ~600 ELO ( 2^11 or so ) less than current SF running in single core model( let's say ~3200 ) so then maybe ~2600 on a 33 Mhz 486. Significantly better than any software available back then.

Chess Genius 1 obtained the highest rating 2140 on a 486/33 machine back in the day:

Code: Select all

 231 Chess Genius 1.0  486/33 MHz            2140   31   -31   483   50%  2143 

source-> SSDF full rating list: http://ssdf.bosjo.net/long.txt

Others may argue - I really don't care - but this may help you in your quest to find the answer.

[bnemias]

I have run sf6 benchmark on a MIPS 4kec cpu, 150 Mhz. This device is very close to a 486/50 for chess, however I used a modern compiler, which I assume you could get working on a 486, but it wouldn't be fun.

Code: Select all

bench 2 1 8

gets 5knps.

So for 25Mhz, you are looking at 2500 nps. I think these numbers are within 20% of reality, but a 1000 nps back of the envelope is pretty close.

I ran a 14 game tournament using the above device (5knps sf6) against a very souped up mchess @ 300knps. sf6 scored 12.5 points. That mchess running on a real 486/50 in the day (about 2knps) was approx 2400 rating. I can't remember if that was USCF or ELO. In any event, I think sf6 on similar hardware is a lot stronger. 2600 to 2800 would be my guess on the same scale.

(Not using that rating list. 2400 comes from memory, when I attended the ACM here in ABQ in the 486 era. mchess scored 2nd, behind deep thought, its only loss)
https://chessprogramming.wikispaces.com/ACM+1991

[bnemias]

I meant not more than 20% off, not "within 20% of reality"

[Steve Maughan]

Back in 1991 I had a 16 Mhz 386sx. I ran M-Chess 1.18 and Rex Chess 2.3. Both did about 1000 nodes per second. So I think your estimate of 1000 for Stockfish is about right.

For those interested, you can read my 1991 review of M-Chess on this hardware here:

http://www.chesscomputeruk.com/SS_35.pdf

That's 25 years ago - Fun times,

Steve

[Werewolf]

Back in 1991 I had a 16 Mhz 386sx. I ran M-Chess 1.18 and Rex Chess 2.3. Both did about 1000 nodes per second. So I think your estimate of 1000 for Stockfish is about right.

For those interested, you can read my 1991 review of M-Chess on this hardware here:

http://www.chesscomputeruk.com/SS_35.pdf

That's 25 years ago - Fun times,

Steve

It's YOU!!!

You're the man who got me into computer chess. I read your review when I was 16 and I thought to myself: MChess is amazing, I must get it.

And I did and I was impressed.

I think the early versions were more positive than the later ones (?), some people call MChess positional. But it was one of my favourite programs.

In an advert (I think in "Chess" magazine) the wording went like this:
There was some hype about Mchess then it read:

"There's only one thing stronger than MChess, and that's the ChessMachine"
Sadly I couldn't afford it.

Wow....happy memories.

[Steve Maughan]

Hi Carl,

It's YOU!!!

Yes indeed it's me. I'm so pleased I got you interested in computer chess!

I've enjoyed looking through the Selective Search back issues. I have many of them. I'm really looking forward to the early editions when they are released as they were lost in a move. You too had quite a few columns over the years!

IMO M-Chess was a historic milestone. I think it was one of the first engines to have a good king-safety evaluation. It was also good at tactics. I emailed Marty Hirsch a couple of years ago and asked if he would write a UCI version. He basically said, "not now but maybe someday".

Steve

[Paloma]

....
For those interested, you can read my 1991 review of M-Chess on this hardware here:

http://www.chesscomputeruk.com/SS_35.pdf

That's 25 years ago - Fun times,

Steve

What i get fron this Link:
The requested URL was rejected. If you think this is an error, please contact the webmaster.

[Steve Maughan]

Hmmm, the link works for me. As an alternative go to this site and take a look at issue 39 which has the M-Chess review

http://www.chesscomputeruk.com/html/sel ... earch.html

Steve

[Vinvin]

Hmmm, the link works for me. As an alternative go to this site and take a look at issue 39 which has the M-Chess review

Steve

It works for me too, on 2 different computers.