Apple MacBook ULTRA 20-inch M5 ULTRA is coming :-)

[Hai]

[Werewolf]

I bet there won't be an M4 Ultra.

Just because Apple do internal experiments doesn't mean the CPU will see the light of day. M5 Ultra is possible though.

[Werewolf]

I've done my own testing now and I've found the following:

Test Machine: M2 Ultra Mac Studio 24 cores (16+8)
OS: Latest MacOS as of July 2025

Installed Parallels. Installed Windows 11 Pro afterwards.

Stockfish 17.1 AVX2 - failed to run
Stockfish 17.1 BMI2 - failed to run
Stockfish 17.1 popcnt - fastest running available

On 1 thread SF 17.1 got about 550 KN/s, that is about half as fast as the best from AMD / Intel.

[royb]

You may want to try not running Parallels and Windows under that just to run Stockfish. If you download the MacOS binary from the Stockfish website, that should run faster than the method/configuration you chose?

I say that because on my M2 Macbook Air, I get 1.0 Million nodes per second from the starting position for a game with Stockfish on a single core. Not sure of the clock speed of the M2 Ultra cores...

[Werewolf]

You may want to try not running Parallels and Windows under that just to run Stockfish. If you download the MacOS binary from the Stockfish website, that should run faster than the method/configuration you chose?

I say that because on my M2 Macbook Air, I get 1.0 Million nodes per second from the starting position for a game with Stockfish on a single core. Not sure of the clock speed of the M2 Ultra cores...

Yes of course SF is faster in MacOS but the issue is you can't run all the Windows chess programs (Chessbase / Fritz / Aquarium etc etc).

An earlier comment by Hai - sorry it may have been in a different thread - claimed AVX was supported in Parallels and it isn't.

[royb]

You may want to try not running Parallels and Windows under that just to run Stockfish. If you download the MacOS binary from the Stockfish website, that should run faster than the method/configuration you chose?

I say that because on my M2 Macbook Air, I get 1.0 Million nodes per second from the starting position for a game with Stockfish on a single core. Not sure of the clock speed of the M2 Ultra cores...

Yes of course SF is faster in MacOS but the issue is you can't run all the Windows chess programs (Chessbase / Fritz / Aquarium etc etc).

An earlier comment by Hai - sorry it may have been in a different thread - claimed AVX was supported in Parallels and it isn't.

Oh, I see your point. I run Hiarcs Chess Explorer Pro on MacOS and thus I can run native M-series binaries there. Thanks for clarifying your post!

[Hai]

Yes of course SF is faster in MacOS but the issue is you can't run all the Windows chess programs (Chessbase / Fritz / Aquarium etc etc).

An earlier comment by Hai - sorry it may have been in a different thread - claimed AVX was supported in Parallels and it isn't.

I've done my own testing now and I've found the following:

Test Machine: M2 Ultra Mac Studio 24 cores (16+8)
OS: Latest MacOS as of July 2025

Installed Parallels. Installed Windows 11 Pro afterwards.

Stockfish 17.1 AVX2 - failed to run
Stockfish 17.1 BMI2 - failed to run
Stockfish 17.1 popcnt - fastest running available

On 1 thread SF 17.1 got about 550 KN/s, that is about half as fast as the best from AMD / Intel.

You can't run all windows chess programs on every windows device...
You can use some chess software from ChessBase which they have build for macOS devices.
They even have Apple App Store apps.
People can write an email to the ChessBase support so they create ChessBase 19 for macOS and Fritz 21 for macOS.

--------------------------------------------------------------------------------------

https://www.cpu-monkey.com/en/compare_c ... max_38_gpu
Apple M1 Max (32-GPU) Characteristic Apple M2 Max (38-GPU)
Rosetta 2 x86-Emulation ISA extensions Rosetta 2 x86-Emulation

https://www.cpu-monkey.com/en/compare_c ... cpu_40_gpu
Apple M3 Max (16-CPU 40-GPU) Characteristic Apple M4 Max (16-CPU 40-GPU)
Rosetta 2 x86-Emulation ISA extensions Rosetta 2 x86-Emulation, AVX2, SME

https://www.cpu-monkey.com/en/compare_c ... cpu_40_gpu
Apple M2 Ultra (76-GPU) Characteristic Apple M4 Max (16-CPU 40-GPU)
Rosetta 2 x86-Emulation ISA extensions Rosetta 2 x86-Emulation, AVX2, SME

--------------------------------------------------------------------------------------

I get in the starting position on my M1 Max with only 8 of 10 cores 8400. After 60 seconds I have 7400, due to TBs hits.
A lot of people still don't know how to compile Stockfish.
A lot of people still don't know how to compile a much faster Stockfish.
A lot of people still don't know how to install newest clang and newest gcc and newest... on macOS.

--------------------------------------------------------------------------------------

Stockfish developers mentioned that they won't spend time to change the Stockfish code so you can compile Stockfish AVX2 on macOS, because it will be in theory always slightly slower than the native compiled version of Stockfish.
Of course BMI2 also failed to run. That was obvious.

--------------------------------------------------------------------------------------

popcnt - fastest running available
Really...
Neon
Dotprod
SME???

The Stockfish developers mentioned also that they could increase the speed of Stockfish on Apple M chip devices a lot.
I hope they will do it soon and also in the future.

--------------------------------------------------------------------------------------

Most people don't know how each thing works in detail.
Most people also don't know how all these things work together.
There is something top secret and it's called google.

---------------------------------------------------------------------------------------

Scalable Matrix Extension (SME) - Can Stockfish use this?
I haven't heard anything about this from the Stockfish developers.
But it will make NNUE a lot faster.

[Werewolf]

I get in the starting position on my M1 Max with only 8 of 10 cores 8400. After 60 seconds I have 7400, due to TBs hits.
A lot of people still don't know how to compile Stockfish.
A lot of people still don't know how to compile a much faster Stockfish.
A lot of people still don't know how to install newest clang and newest gcc and newest... on macOS.


popcnt - fastest running available
Really...
Neon
Dotprod
SME???

Can't you see my post wasn't talking about running Stockfish natively on MacOS, but running Windows on Paralells which runs on MacOS? That's totally different and your Neon comments etc are therefore not relevant.