Page 1 of 14
correspondence chess in the age of NNUE
Posted: Thu Jan 21, 2021 6:26 am
by lkaufman
Now that strong Stockfish NNUE versions have been out for about six months, there should have been lots of correspondence games played by email that have completed during this interval. Based on results we have seen in both normal chess and chess960 when powerful hardware is used with NNUE engines, unless bad openings are mandated, the draw percentage between roughly equal NNUE engines at long time controls is in the high 90s, basically making the game unplayable. So what is the experience of correspondence players using good hardware and NNUE engines over the past six months? I suppose some games are won due to opponents having inferior hardware or not knowing about NNUE or not caring about the result much, but is anyone winning any games from opponents where these factors are not applicable? Is correspondence chess still playable without mandated openings?
Re: correspondence chess in the age of NNUE
Posted: Thu Jan 21, 2021 7:35 am
by Dann Corbit
It takes a very long time for a correspondence tournament to complete.
I guess in about 8 more months we will see outcomes from using NNUE stuff to analyze
Re: correspondence chess in the age of NNUE
Posted: Thu Jan 21, 2021 11:02 am
by Ozymandias
Taking out really weak players (probably not even using a computer) the draw rate for the past three years is already close to 60%. If you take out weak players, like those who only use a computer to check some positions, it raises over 75%. We're still talking over 140,000 games for just three years, and the best ones from this period are likely to finish yet.
To get to 90%, you need to increase the strength of players to a point were only 65,000 games are left. To give you an idea, you still have close to 2,500 players at this level. These are serious players who care about results and have both access and willingness to use HW. This doesn't mean they use the latest, but that's not necessary to face a draw problem, and if you select the cream of the crop (more than 14,000 games and close to a thousand players) you surpass the 95% draw rate. This is why Arno wrote a proposal I submitted to you some months ago, you didn't need the advent of NNUEs to have a real problem in your hands.
Re: correspondence chess in the age of NNUE
Posted: Thu Jan 21, 2021 12:39 pm
by cdani
Any random guy of 2100 using NNUE at ICCF is much more difficult to win than 6 months ago.
Re: correspondence chess in the age of NNUE
Posted: Thu Jan 21, 2021 12:57 pm
by Milos
Ozymandias wrote: ↑Thu Jan 21, 2021 11:02 am
Taking out really weak players (probably not even using a computer) the draw rate for the past three years is already close to 60%. If you take out weak players, like those who only use a computer to check some positions, it raises over 75%. We're still talking over 140,000 games for just three years, and the best ones from this period are likely to finish yet.
To get to 90%, you need to increase the strength of players to a point were only 65,000 games are left. To give you an idea, you still have close to 2,500 players at this level. These are serious players who care about results and have both access and willingness to use HW. This doesn't mean they use the latest, but that's not necessary to face a draw problem, and if you select the cream of the crop (more than 14,000 games and close to a thousand players) you
surpass the 95% draw rate. This is why Arno wrote a proposal I submitted to you some months ago,
you didn't need the advent of NNUEs to have a real problem in your hands.
So essentially what you are saying is that latest SF NNUE just running on 300$ AMD CPU running couple of correspondence games in parallel for maximum time between the moves (with zero human intervention) is better than 99.99% of correspondence players.
Re: correspondence chess in the age of NNUE
Posted: Thu Jan 21, 2021 1:06 pm
by Ozymandias
At no point did I mention how many correspondence players there are. So I never implied any % of players, just the amount I saw at each draw rate.
Re: correspondence chess in the age of NNUE
Posted: Thu Jan 21, 2021 1:36 pm
by jdart
The draw rate in top-level correspondence chess (world championship tournaments for example) is already in the 90% range I think.
Re: correspondence chess in the age of NNUE
Posted: Thu Jan 21, 2021 1:38 pm
by Branko Radovanovic
Ozymandias wrote: ↑Thu Jan 21, 2021 11:02 am
Taking out really weak players (probably not even using a computer) the draw rate for the past three years is already close to 60%. If you take out weak players, like those who only use a computer to check some positions, it raises over 75%. We're still talking over 140,000 games for just three years, and the best ones from this period are likely to finish yet.
To get to 90%, you need to increase the strength of players to a point were only 65,000 games are left. To give you an idea, you still have close to 2,500 players at this level. These are serious players who care about results and have both access and willingness to use HW. This doesn't mean they use the latest, but that's not necessary to face a draw problem, and if you select the cream of the crop (more than 14,000 games and close to a thousand players) you
surpass the 95% draw rate. This is why Arno wrote a proposal I submitted to you some months ago,
you didn't need the advent of NNUEs to have a real problem in your hands.
That's true, even before NNUEs CC draw rate was already significantly above 90% at the highest level of play according to these stats pointed out by L. Ljubicic:
NNUEs will spread this effect into lower Elos and probably kill the current concept of CC within a couple of years.
Re: correspondence chess in the age of NNUE
Posted: Thu Jan 21, 2021 1:57 pm
by Ozymandias
The same way it killed Freestyle years ago.
Re: correspondence chess in the age of NNUE
Posted: Thu Jan 21, 2021 3:04 pm
by whereagles
High draw rate at top level is sort of an heuristic proof chess is a theoretical draw
