NPU core use by lichess web app - any plans?

This is obviously a developer question, so it would be understandable if nobody knowledgeable is around :

I am aware that standard APIs like WebNN already exist, but I doubt that they are available / usable in any stable browser releases yet.

However, online chess seems to be an almost perfect use case for on chip NPUs to speed up the in-browser analysis - or so I would think.

Does anybody have any insights to share how much consideration is given to this as it applies to lichess?

This is obviously a developer question, so it would be understandable if nobody knowledgeable is around : I am aware that standard APIs like WebNN already exist, but I doubt that they are available / usable in any stable browser releases yet. However, online chess seems to be an almost perfect use case for on chip NPUs to speed up the in-browser analysis - or so I would think. Does anybody have any insights to share how much consideration is given to this as it applies to lichess?

Rabergsel

For those not knowing what an NPU is: it is basically a special type of processor core accelerating neural network computing.

And yes, Stockfish transitioned to NNUE technology a few years ago, but the thing is that this kind of Neural Networks is optimized for normal CPU cores. Maybe there would be a bit of acceleration as it is still the same calculations but still....

Also, I don't think it is worth the effort to implement this, as NPUs are (as of today) only part of modern, higher-end CPUs of Intel. The effect for the average player on lichess wouldn't even be usable.

Plus the Stockfish engine, which would use the NPU, is not part of the Lichess Project, and is its own project: https://stockfishchess.org/

I hope I got everything right.

For those not knowing what an NPU is: it is basically a special type of processor core accelerating neural network computing. And yes, Stockfish transitioned to NNUE technology a few years ago, but the thing is that this kind of Neural Networks is optimized for normal CPU cores. Maybe there would be a bit of acceleration as it is still the same calculations but still.... Also, I don't think it is worth the effort to implement this, as NPUs are (as of today) only part of modern, higher-end CPUs of Intel. The effect for the average player on lichess wouldn't even be usable. Plus the Stockfish engine, which would use the NPU, is not part of the Lichess Project, and is its own project: https://stockfishchess.org/ I hope I got everything right.

coderunner86

I agree with everything you wrote and I am not holding my breath to see actual, useful implementations of anything browser-related for a long while.

But since the API exists it is conceivable that several years from now we would have NPUs that better reflect actual neural network topology and may be orders of magnitude more efficient. I vaguely understand that most NPUs these days are roughly very fast matrix multipliers, but that's where it ends for me.

Again - just curious; not expecting anything. And thanks for responding.

I agree with everything you wrote and I am not holding my breath to see actual, useful implementations of anything browser-related for a long while. But since the API exists it is conceivable that several years from now we would have NPUs that better reflect actual neural network topology and may be orders of magnitude more efficient. I vaguely understand that most NPUs these days are roughly very fast matrix multipliers, but that's where it ends for me. Again - just curious; not expecting anything. And thanks for responding.

Solal35 edited

Does anybody have any insights to share how much consideration is given to this as it applies to lichess?

Hello, for the moment, as far as I know, none. We try to modify as little stockfish as possible, and generally our position is that even in its current form it is strong enough for all analysis usage.

Generally players want to upgrade to the latest stockfish for psychological reasons (having access the best version), rather than practical ones.

> Does anybody have any insights to share how much consideration is given to this as it applies to lichess? Hello, for the moment, as far as I know, none. We try to modify as little stockfish as possible, and generally our position is that even in its current form it is strong enough for all analysis usage. Generally players want to upgrade to the latest stockfish for psychological reasons (having access the best version), rather than practical ones.

coderunner86

Ok, my curiosity has been satisfied. I may ask again in a couple of years :)

This topic has been archived and can no longer be replied to.