Apple M1 Ultra - is this enough for chess?

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

Magnum
Posts: 195
Joined: Thu Feb 04, 2021 10:24 pm
Full name: Arnold Magnum

Apple M1 Ultra - is this enough for chess?

Post by Magnum »

https://www.apple.com/mac-studio/
It should have more than enough power.
Werewolf
Posts: 2058
Joined: Thu Sep 18, 2008 10:24 pm

Re: Apple M1 Ultra - is this enough for chess?

Post by Werewolf »

I knew it wouldn't take long to see something on the M1 ultra.

The issue is for media-related work its performance may be high (Geekbench suggests close to Threadripper 3990, but I think it overrates ARM related CPUs) but for chess it'll be lower for sure.

Secondly, you'd have to compile SF Dev yourself.

Thirdly, the choice of UIs is much lower on MACOS.

But I suppose if there's interest people could start developing engines / UIs for MACOS now.
Luecx
Posts: 138
Joined: Thu Jun 18, 2020 9:20 pm
Full name: Finn Eggers

Re: Apple M1 Ultra - is this enough for chess?

Post by Luecx »

Koivisto has support for ARM (M1) CPUs. I tested the performance compared to my 3950x and the speed is very similar.
The ability to speak does not make you intelligent. https://github.com/Luecx/Koivisto

Image
Werewolf
Posts: 2058
Joined: Thu Sep 18, 2008 10:24 pm

Re: Apple M1 Ultra - is this enough for chess?

Post by Werewolf »

Luecx wrote: Wed Mar 09, 2022 12:45 pm Koivisto has support for ARM (M1) CPUs. I tested the performance compared to my 3950x and the speed is very similar.
Which M1 did you test?

M1, M1 Max etc?
Sopel
Posts: 391
Joined: Tue Oct 08, 2019 11:39 pm
Full name: Tomasz Sobczyk

Re: Apple M1 Ultra - is this enough for chess?

Post by Sopel »

For stockfish 14.1 pure nnue we should expect about 30MNps, extrapolating from https://forums.macrumors.com/threads/st ... 0.2326552/ and https://browser.geekbench.com/v5/cpu/co ... e=13330272. Comparing to http://ipmanchess.yolasite.com/amd--int ... ckfish.php it should place around AMD Ryzen 9 3900X/ Intel Core i5 12600K. Respectable, but nothing special.
dangi12012 wrote:No one wants to touch anything you have posted. That proves you now have negative reputations since everyone knows already you are a forum troll.

Maybe you copied your stockfish commits from someone else too?
I will look into that.
Magnum
Posts: 195
Joined: Thu Feb 04, 2021 10:24 pm
Full name: Arnold Magnum

Re: Apple M1 Ultra - is this enough for chess?

Post by Magnum »

Sopel wrote: Wed Mar 09, 2022 4:08 pm For stockfish 14.1 pure nnue we should expect about 30MNps, extrapolating from https://forums.macrumors.com/threads/st ... 0.2326552/ and https://browser.geekbench.com/v5/cpu/co ... e=13330272. Comparing to http://ipmanchess.yolasite.com/amd--int ... ckfish.php it should place around AMD Ryzen 9 3900X/ Intel Core i5 12600K. Respectable, but nothing special.
Write code to make Stockfish use more of the power/potential from the M1 Ultra.
dangi12012
Posts: 1062
Joined: Tue Apr 28, 2020 10:03 pm
Full name: Daniel Infuehr

Re: Apple M1 Ultra - is this enough for chess?

Post by dangi12012 »

Here is one gold nugget on the M1 ARM:

You get the innocent looking instruction __rbitll
Its the bitreverse (not byte) instruction! :)

That makes it possible to have a uniform hyperbola quiescence algorithm like implemented in this thread:
http://www.talkchess.com/forum3/viewtop ... 8&start=20

Shortcut:
https://github.com/Gigantua/Chess_Moveg ... rotation.h


What will also be interesting is the gpu performance - since Vulcan CRUSHES any CPU.
Worlds-fastest-Bitboard-Chess-Movegenerator
Daniel Inführ - Software Developer
Sopel
Posts: 391
Joined: Tue Oct 08, 2019 11:39 pm
Full name: Tomasz Sobczyk

Re: Apple M1 Ultra - is this enough for chess?

Post by Sopel »

Magnum wrote: Wed Mar 09, 2022 10:53 pm
Sopel wrote: Wed Mar 09, 2022 4:08 pm For stockfish 14.1 pure nnue we should expect about 30MNps, extrapolating from https://forums.macrumors.com/threads/st ... 0.2326552/ and https://browser.geekbench.com/v5/cpu/co ... e=13330272. Comparing to http://ipmanchess.yolasite.com/amd--int ... ckfish.php it should place around AMD Ryzen 9 3900X/ Intel Core i5 12600K. Respectable, but nothing special.
Write code to make Stockfish use more of the power/potential from the M1 Ultra.
I already did. Performance on M1 is close to the limit.
You get the innocent looking instruction __rbitll
Its the bitreverse (not byte) instruction! :)
You can get the same and more from AVX512 galois field instructions
dangi12012 wrote:No one wants to touch anything you have posted. That proves you now have negative reputations since everyone knows already you are a forum troll.

Maybe you copied your stockfish commits from someone else too?
I will look into that.
Milos
Posts: 4190
Joined: Wed Nov 25, 2009 1:47 am

Re: Apple M1 Ultra - is this enough for chess?

Post by Milos »

Sopel wrote: Wed Mar 09, 2022 4:08 pm For stockfish 14.1 pure nnue we should expect about 30MNps, extrapolating from https://forums.macrumors.com/threads/st ... 0.2326552/ and https://browser.geekbench.com/v5/cpu/co ... e=13330272. Comparing to http://ipmanchess.yolasite.com/amd--int ... ckfish.php it should place around AMD Ryzen 9 3900X/ Intel Core i5 12600K. Respectable, but nothing special.
And 12700F that costs 300$ will already by slightly better. And one can build a system with 12700F or even better 12700K with DDR4 compatible board and 32GB of memory for under 800$. Ofc without GPU.
And for running NNUE based engine it beats 4000+$ M1 Ultra. Isn't that sweet :lol:.
dangi12012
Posts: 1062
Joined: Tue Apr 28, 2020 10:03 pm
Full name: Daniel Infuehr

Re: Apple M1 Ultra - is this enough for chess?

Post by dangi12012 »

Sopel wrote: Wed Mar 09, 2022 11:53 pm You can get the same and more from AVX512 galois field instructions
Pls dont ever quote me ever again. You half knowledge is just too annoying.
Have you looked at the throughput and latency? Not to mention that AVX512 is DOA and Intel is suspected to drop it soon... Enabling AVX512 often makes every other process on the system run slower.

If you are using this in production code - measure what you did to everyone around you.
If anyone reads this: Sopel is a confirmed forum troll who is just very annoying to every thread he writes to.


Back to topic:
M1 looks interesting - will wait for INDEPENDENT reviews.
Worlds-fastest-Bitboard-Chess-Movegenerator
Daniel Inführ - Software Developer