Please improve KataGOs Code to run much faster on Apple devices when using the new Mac Studio

[lex312]

It’s very powerful and it should be able to use this power: https://www.apple.com/mac-studio/specs/

[OmnipotentEntity]

It would be very helpful if you gave us more information. What performance level are you seeing? Why do you think it should be higher? Have you compared this level of performance with other users?

Ultimately, I doubt any contributor has Mac Studio, and I don't really anyone wants to purchase a Mac Studio to try and see on the off chance that they can reproduce whatever performance issue you believe you're having.

Also:

https://github.com/dje-dev/Ceres/issues/73
https://github.com/LeelaChessZero/lc0/issues/1711
https://github.com/official-stockfish/Stockfish/issues/3959

Please don't spam low quality bug reports to multiple projects.

[merlinpan]

Someone have benchmarked M1 on tensorflow:

https://github.com/tlkh/tf-metal-experiments/blob/main/README.md#experiments-and-benchmarks

So on the M1Max (macbook pro), they reached ~40W at ~1/8 the performance of nvidia 3090 (at 1/8 the power draw approximately as well).

I don't know whether the "tensorflow support" discussed there is based on apple's neural core or just based on the GPU (they just say "metal").

The M1Ultra of the apple studio is more or less 2x M1Max connected together, with a more aggressive power cap. So we can expect it to reach ~1/4 to 1/3 of the performance of 3090 without additional optimizations. (~3060 level I would say).

Nothing to write home about, but considering there is no cuda on mac, might be worth supporting if a contributor can be found.