For how long will Stockfish be competitive versus the best NN ?

[Chessqueen]

NN just keep getting better and better, for how long will Stockfish be competitive versus the best NN ?
https://tcec-chess.com/

[Dr.Wael Deeb]

A phenomenal chess engine indeed ....

It's like it emerged from a another dimension ....

Dr.D

[Modern Times]

Is GPU technology improvements moving at a faster pace than CPU ? If so, then the answer is probably not very long. As long as you are willing to spend the money on the GPU and more powerful PSU to run it.

If you have say £600 to spend on a PC, not including the monitor, for that sort of money would LC0 be stronger or Stockfish ?

[Chessqueen]

Is GPU technology improvements moving at a faster pace than CPU ? If so, then the answer is probably not very long. As long as you are willing to spend the money on the GPU and more powerful PSU to run it.

If you have say £600 to spend on a PC, not including the monitor, for that sort of money would LC0 be stronger or Stockfish ?

Certain AB engines like Komodo and Komodo MCTS that used to be one of the top that won several TCEC Championships will be left on the dust by NN Engines unless Larry Kaufman release Komodo NN Soon to become competitive.

[Modern Times]

You need a very expensive GPU to get the best out of NN engines. On the other hand AB engines can perform very well on modest CPUs.

However GPU prices should drop in the sense that you'll get more for your money as time goes by, then things will be different.

[Dann Corbit]

I see a disturbing trend for NN engines.
Look at the Elo curve from learning as a function of time for any of the NN experiments (go, chess, what have you) and you will see it is not a linear looking increase like the graphs made by Pohl of the strength increase of SF.
Look at the graphs here of Elo verses training steps:
https://deepmind.com/blog/article/alpha ... ogi-and-go
And look at the graph here of Elo verses time:
https://www.sp-cc.de/
The three graphs on Pohl's page are continuations from the bottom to the top.
The SF Elo growth is roughly linear.
The NN Elo growth is clearly sub-linear (and not close to linear).
Of course, both of these graphs describe only software progress on the same platform.
I expect continued exponential growth for both GPUs and CPUs for a while, based on parallel progress.
I further expect that the optimum combination will employ both systems, with the alpha-beta version seeding the same hash table as the NN version, and therefore both types of algorithm will be telling the other one where to look. Sort of a next generation lazy SMP.
The AMD unified memory approach should make this trivial in the near future.

Edit: I should mention also that a linear Elo graph means exponential progress.

[Ovyron]

As a correspondence chess player that has dedicated up to 3 weeks for the analysis of a single move (and much longer) I have to say that I haven't been impressed by NNs. There's a good chunk of positions where they have absolutely no idea, and they don't get better at it with more time.

Needing a very fast GPU for them might be true at some level, but at the corr level CPU Leela at 23 nodes per second seems to be enough, this was apparent in the analysis of 1.g4 where after 800M nodes Leela was showing a line that I could find in a very short time with interactive analysis with CPU Leela, and my analysis was still superior deep down the line, because Leela eventually will like the same positions, so with interactive analysis you can get there and surpass it.

Anybody that wants to see the future just needs to give more time to the NNs, the moves they'll play at 40 moves in 40 minutes will not be better than the ones they play a 12 hours/move today.

The people still amazed by the victories that Alpha Zero had over Stockfish 8 just need to match Stockfish Dev against Stockfish 8. SPOILER ALERT you will see positions where Stockfish 8 shows a 0.00, and a 0.80 disadvantage the next move, and a 1.60 disadvantage the next, while Stockfish dev will have aimed for these positions with plans that look much more superior than anything Alpha Zero played. Stockfish dev would have looked like coming from another dimension back then, just because Stockfish 8 was the strongest chess known at the time.

But 1 minute/move is a level not worthy of calling home about, and most of these improvements have been on the software side, not the hardware side. If trends continue the answer to the OP's question will depend on how many ideas the Stockfish developers have left.

[yurikvelo]

If you have say £600 to spend on a PC, not including the monitor, for that sort of money would LC0 be stronger or Stockfish ?

you hardly get any (non-integrated) GPU for this budget.
GeForce RTX 2080 ALONE cost £600 (if you already have a decent PC and want to run Lc0 on a level comparable to Stockfish on your current PC - you have to spend £600+ for upgrade

For PC with slow GPU or without GPU (integrated APU) you might want lightweigt nets:
https://github.com/AlexisOlson/lc0/wiki/Best-Lc0-Nets

[Modern Times]

That was kind of my point - for average sort of money that an average user would spend, Stockfish may well perform better than an NN engine. Getting the best out of NN engines is a very expensive endeavour. The £600 price point was pretty low but for a little more maybe £800 you'd get a machine with something with a GTX 1660 plus perhaps. Anyway, I'm not in the market for a new PC myself, and when I will be it will be a laptop at £600 - £800 current prices. That will just be an integrated GPU.

[Dann Corbit]

You can't just put a top end GPU card into a very inexpensive PC because often the power supply is not up to the task.
However, you can get a much cheaper card that uses less power and still provides very good analysis.
https://www.videocardbenchmark.net/common_gpus.html
Look at the price performance graph and you can see for $200 you can get a pretty good card for LC0 that will also use a lot less power than a 2080Ti.

GPUs can also be used for other interesting things like folding proteins, searching for Mersenne Primes, astronomical or physics grid calculations, etc.

If you want the strongest system in the world, you can spend $100,000 and someone might still outspend you and create a stronger system. But if you just want fabulous chess analysis, you can use time as your ally.

For instance, I bought an AMD 3970x which is 1/3 slower than an AMD 3990x. So if I analyze one and a half times as long, I have the same approximate output. Whatever you spend you can still get great analysis by simply letting the machine think longer. Now, it might be a very long wait if you want the best answer in the world, but you spend what you feel comfortable and live with that.

On an inexpensive smartphone, you can get stockfish and have a GM in your pocket to analyze positions for you at your beck and call. How much more power you need is strictly up to you. If you just want a critique of your game, you don't need a $5,000 system, since your cell phone will do nicely.

It is also true that you get what you pay for. If you have top end equipment and some very strong programs you can solve long, deep, elaborate studies and prove they are correct (or even disprove them). I would put it this way, get what your budget allows and what you feel truly comfortable with.