Stockfish up to 8 times faster on Apple M1

[Magnum]

https://github.com/explosion/thinc-apple-ops
https://spacy.io/usage/v3-2

[Ras]

None of your links says anything like that. Just because some applications may experience a speedup, it does not mean that Stockfish would be necessarily among them.

[wickedpotus]

No...On the contrary M1, M1 Pro and M1 Max is actually pretty crappy for everything multithreaded chess related.



But reality has very little to do with Apple fanboys and not so tech-savvy youtubers, bloggers, "tech"-sites who only do flwed benches like "Geekbench" and other uninformed sh*t...

Just go and do some g*d d*mn benchmarking before posting sh*t like this in a forum frequented by mostly intelligent chess friends.

[dangi12012]

No...On the contrary M1, M1 Pro and M1 Max is actually pretty crappy for everything multithreaded chess related.



But reality has very little to do with Apple fanboys and not so tech-savvy youtubers, bloggers, "tech"-sites who only do flwed benches like "Geekbench" and other uninformed sh*t...

Just go and do some g*d d*mn benchmarking before posting sh*t like this in a forum frequented by mostly intelligent chess friends.

Yes everything compiled for M1 natively is not older than 1 year and has a current compiler - then these people compare to old builds with outdated compilers - like geekbench. Its insanity.
At least cinebench did a new version with R23.

[Raphexon]

No...On the contrary M1, M1 Pro and M1 Max is actually pretty crappy for everything multithreaded chess related.



But reality has very little to do with Apple fanboys and not so tech-savvy youtubers, bloggers, "tech"-sites who only do flwed benches like "Geekbench" and other uninformed sh*t...

Just go and do some g*d d*mn benchmarking before posting sh*t like this in a forum frequented by mostly intelligent chess friends.

Yes everything compiled for M1 natively is not older than 1 year and has a current compiler - then these people compare to old builds with outdated compilers - like geekbench. Its insanity.
At least cinebench did a new version with R23.

That doesn't change the fact that NEON is shit compared to AVX2

[wickedpotus]

No...On the contrary M1, M1 Pro and M1 Max is actually pretty crappy for everything multithreaded chess related.



But reality has very little to do with Apple fanboys and not so tech-savvy youtubers, bloggers, "tech"-sites who only do flwed benches like "Geekbench" and other uninformed sh*t...

Just go and do some g*d d*mn benchmarking before posting sh*t like this in a forum frequented by mostly intelligent chess friends.

Yes everything compiled for M1 natively is not older than 1 year and has a current compiler - then these people compare to old builds with outdated compilers - like geekbench. Its insanity.
At least cinebench did a new version with R23.

So why don't you (or some Apple fanboy that wants to "prove" their powerpoint-vapor-benches then use your imaginary compiler/optimizer and show us your stockfish benchmarks, or LC0 ???.. Every CPU is "the fastest" on the sales-slides, until you try to throw real workloads on it...

And besides the benches above are from native compilations on M1 of stockfish using NEON optimizations on the M1 )so these builds are native and built recently).. It still sucks,

[dangi12012]

So why don't you (or some Apple fanboy that wants to "prove" their powerpoint-vapor-benches then use your imaginary compiler/optimizer and show us your stockfish benchmarks, or LC0 ???.. Every CPU is "the fastest" on the sales-slides, until you try to throw real workloads on it...

And besides the benches above are from native compilations on M1 of stockfish using NEON optimizations on the M1 )so these builds are native and built recently).. It still sucks,

You literally didnt read my text.
I am saying apple media does unfair comparisons with brand new builds for apple vs 2 year old software for intel.
To be honest Apple also bought TSMCs 5nm process node and Intel does everything on 10nm. The impressive thing about M1 is rosetta where Microsoft tried 10 years with arm and failed - and apple produced a perfect transcompiler during runtime which works flawlessly.

[Modern Times]

The impressive thing about M1 is rosetta where Microsoft tried 10 years with arm and failed - and apple produced a perfect transcompiler during runtime which works flawlessly.

Yes - that and it is incredibly power efficient for the performance it offers across typical non-chess tasks.

[smatovic]

The impressive thing about M1 is rosetta where Microsoft tried 10 years with arm and failed - and apple produced a perfect transcompiler during runtime which works flawlessly.

Yes - that and it is incredibly power efficient for the performance it offers across typical non-chess tasks.

Non-chess task, I think that is the point, I am curious if Apple will come up with some M2 discrete stuff for the Mac Pro workstation next year, some 128 core M2 or alike, and there migth be a switch from NEON to SVE2 in pipe, would be funny, but then, it would be still pointless in regard of nps/dollar.

--
Srdja

[Werewolf]

One online rumour is that the next Mac Pro will have dual M1 Max CPUs.

I seriously hope they can do better than that.