Stockfish up to 8 times faster on Apple M1

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

Moderator: Ras

wickedpotus
Posts: 153
Joined: Sun May 16, 2021 5:33 pm
Full name: Aron Rodgriges

Re: Stockfish up to 8 times faster on Apple M1

Post by wickedpotus »

Look in the beginning of this thread for some nps benchmarks.. M1 version of Stockfish has been improved with special ARM compiles lately but u get the idea from those figures.

forum3/viewtopic.php?f=2&t=78673&hilit=Apple#p912017

Pretty much any 8-core AMD mobile CPU beats even the most expensive M1 Max CPU for SF. And M1 is really no contest even for cheap 6-core AMD U versions. Intels new Alderlake CPUs are also killing M1 for chess but those are just making its way out in laptops currently (you see a lot of "special prices" currently for gen.11 laptops that retaliers want to offload :-)
Modern Times
Posts: 3753
Joined: Thu Jun 07, 2012 11:02 pm

Re: Stockfish up to 8 times faster on Apple M1

Post by Modern Times »

wickedpotus wrote: Wed Dec 29, 2021 11:41 pm
Pretty much any 8-core AMD mobile CPU beats even the most expensive M1 Max CPU for SF. And M1 is really no contest even for cheap 6-core AMD U versions. Intels new Alderlake CPUs are also killing M1 for chess but those are just making its way out in laptops currently (you see a lot of "special prices" currently for gen.11 laptops that retaliers want to offload :-)
Yes. And the new Intel mobile CPUs are stunning from the benchmarks I've seen, killing AMD as well.

The one thing Intel and AMD can't match M1 for is power consumption if I'm not mistaken.
ydebilloez
Posts: 186
Joined: Tue Jun 27, 2017 11:01 pm
Location: Lubumbashi
Full name: Yves De Billoëz

Re: Stockfish up to 8 times faster on Apple M1

Post by ydebilloez »

wickedpotus wrote: Wed Dec 29, 2021 11:41 pm Look in the beginning of this thread for some nps benchmarks.. M1 version of Stockfish has been improved with special ARM compiles lately but u get the idea from those figures.
I have seen those figures, but as the discussion went from figures to pro-contra apple fanboys, and we have compared bmi2/avx against non optimized M1, I wonder if we have updated charts taking into account special ARM compiles.
Yves De Billoëz @ macchess belofte chess
Once owner of a Mephisto I, II, challenger, ... chess computer.
ChickenLogic
Posts: 154
Joined: Sun Jan 20, 2019 11:23 am
Full name: kek w

Re: Stockfish up to 8 times faster on Apple M1

Post by ChickenLogic »

Modern Times wrote: Thu Dec 30, 2021 2:57 am
wickedpotus wrote: Wed Dec 29, 2021 11:41 pm
Pretty much any 8-core AMD mobile CPU beats even the most expensive M1 Max CPU for SF. And M1 is really no contest even for cheap 6-core AMD U versions. Intels new Alderlake CPUs are also killing M1 for chess but those are just making its way out in laptops currently (you see a lot of "special prices" currently for gen.11 laptops that retaliers want to offload :-)
Yes. And the new Intel mobile CPUs are stunning from the benchmarks I've seen, killing AMD as well.

The one thing Intel and AMD can't match M1 for is power consumption if I'm not mistaken.
Is that so surprising? Apple specifically wanted to have a very power efficient chip while the focus of top tier AMD and Intel chips is top performance (almost) regardless of power draw. I wouldn't say that Intel is ahead of AMD. Yes, with DDR5 benchmarks show small gains over AMD processors but you pay more than double the prize for DDR5 than for DDR4. On top of that the newer Intel chips are more expensive than their AMD counterpart. Funnily enough, scalpers try to buy a lot of DDR5 ram but if people are smart they just don't buy it. DDR4 will bring about as much performance while one can just wait for 2nd or 3rd gen DDR5. Just like the early versions of DDR4 RAM didn't really gain that much over DDR3.

I must say that Apple's new laptops do provide an impressive battery life - I just don't think the extra money you spend because it is Apple is justified. I'd rather wait until other's realize that ARM is more than enough for 95% of laptop users (wild guess). I agree that ARM or RISC-V are a good future for laptops. I just don't want those in my desktop until they convincingly beat my x86 CPU. I think Stockfish shows that AVX512 and co can be put to good use.

On that note, there will be Zen4c (c standing for "cloud") which will be similar to intels new efficiency cores in Alder Lake. I think this is sort of a compromise between having a lot of cores and throughput while also recognizing that power draw is an increasing issue in server farms.

I think the problems people have with Apple are (no particular order)
a) the proprietary nature of their eco system (both the OS and the IO of the motherboard)
b) they are slow to accept apps into their store and have weird rules in regards to what is acceptable
c) it's not easy to develop a cross platform app with Mac compatibility if you don't have a MAC yourself
d) price
e) inability to fix hardware problems yourself (we'll see how it goes after their recent announcement)
f) pretentious people that need to feel superior by buying apple - but that is a problem with any brand that has a good reputation and uses that to market their product as a lifestyle. Same can be said about some Windows users and even Linux users.
User avatar
Ras
Posts: 2702
Joined: Tue Aug 30, 2016 8:19 pm
Full name: Rasmus Althoff

Re: Stockfish up to 8 times faster on Apple M1

Post by Ras »

ydebilloez wrote: Thu Dec 30, 2021 3:36 pmI wonder if we have updated charts taking into account special ARM compiles.
No compiler can compensate missing CPU instructions, and NNUE chess engines were not on Apple's radar when they designed the M1 because it's just not a frequent use case.
Rasmus Althoff
https://www.ct800.net
syzygy
Posts: 5774
Joined: Tue Feb 28, 2012 11:56 pm

Re: Stockfish up to 8 times faster on Apple M1

Post by syzygy »

Ras wrote: Tue Dec 28, 2021 9:22 pm
Werewolf wrote: Tue Dec 28, 2021 2:21 pmHowever, HT in chess doesn't seem to yield much elo anymore at decent thread counts.
But it does allow running more games in parallel.
Running more games in parallel with hyperthreading increases noise due to speed fluctuations, which means you need more games to get a significant result. Perhaps it is still worth it, but it is not so clear.