Apple M2

[wickedpotus]

Classical SF is very fast on the M1 architecture.

This is mostly anecdotal right? I have never seen any benches where M1/M2s have been able to post SF-classic benchmarks close to current gens AMDs or Intel CPUs. But prove me wrong if you have any "real" benches to support that statement. Can you find ANY strong chess engine that runs better/faster on an "M1/max/pro/M2" than an alderlake i7 or i9 laptop?

[George Sobala]

Classical SF is very fast on the M1 architecture.

This is mostly anecdotal right? I have never seen any benches where M1/M2s have been able to post SF-classic benchmarks close to current gens AMDs or Intel CPUs. But prove me wrong if you have any "real" benches to support that statement. Can you find ANY strong chess engine that runs better/faster on an "M1/max/pro/M2" than an alderlake i7 or i9 laptop?

My MBA M1 running classical SF dev benchmark on its 4 performance cores gets about 12.1Mnps. This is faster than my desktop Intel 12700F on 4 cores classical - 10.8Mnps. Sure the Intel has many more performance cores, and can also hyperthread, but on a per-core basis the M1 is respectable. (And of course this completely ignores performance per watt).

Now on pure NNUE on 4 cores the M1 only manages 6.6Mnps vs the 12700F's 7.9Mnps. The situation is reversed.

[wickedpotus]

Classical SF is very fast on the M1 architecture.

This is mostly anecdotal right? I have never seen any benches where M1/M2s have been able to post SF-classic benchmarks close to current gens AMDs or Intel CPUs. But prove me wrong if you have any "real" benches to support that statement. Can you find ANY strong chess engine that runs better/faster on an "M1/max/pro/M2" than an alderlake i7 or i9 laptop?

My MBA M1 running classical SF dev benchmark on its 4 performance cores gets about 12.1Mnps. This is faster than my desktop Intel 12700F on 4 cores classical - 10.8Mnps. Sure the Intel has many more performance cores, and can also hyperthread, but on a per-core basis the M1 is respectable. (And of course this completely ignores performance per watt).

Now on pure NNUE on 4 cores the M1 only manages 6.6Mnps vs the 12700F's 7.9Mnps. The situation is reversed.

Thx George,

Your numbers sort of prove my point and indicate that the M1 does not do much better at all without NNUE either. It's simply outpaced by all the current competition.

Old SF11 (non nnue) gets more than 30Mnps on modern Intel and AMD gaming laptops. My old 900usd TUF 4800H based laptops got about 25Mnps sustained with energy mode, on 8-cores (16threads SF) which is like 2 to 3 times the performance of your M1. So the perfromance-ratio disadvantage using Apple Silicon seems pretty constant with SF11 compared to SF15 (given full use of hw).


Compared with desktops is sort of a moot point, especially if you only use 4-threads on an MT CPU capable to run 20 threads on 12 cores, You are comparing it wrong IMHO. I assume you realize the 12700F is capable of running 20 threads for SF. So your comparison would be sort of like comparing the performance of a V12 Sports car with a severely overpriced V4 mini would be sort of a option "better", becuase it uses less fuel to take you to the race-track.

Besides the perf/watt is just plain silly if you are not running a data center. It's like a gamer with a 3080 RTX card would switch down to a GTX 1650 card.. with the "argument" that yeeha the framerate in my game per watt is so much greater.. so I am happy with a lower overasll sucky framerate.

[Sopel]

https://forums.macrumors.com/threads/st ... 0.2326552/ has some results and they are directly comparable with https://ipmanchess.yolasite.com/amd--in ... ckfish.php. No proper benchmarks because apple people for some reason can't really benchmark. TLDR; M1 Ultra could maybe match AMD Ryzen 9 3950X

[wickedpotus]

https://forums.macrumors.com/threads/st ... 0.2326552/ has some results and they are directly comparable with https://ipmanchess.yolasite.com/amd--in ... ckfish.php. No proper benchmarks because apple people for some reason can't really benchmark. TLDR; M1 Ultra could maybe match AMD Ryzen 9 3950X

I am pretty sure at least some Apple users "can" benchmark.. but many Apple fanboys seem to have a selection bias when it comes to benching... And for some reason only post benches or test-cases when they think Apple doesn't look too bad

When asked I have gotten answers like "I don't really care for chess so benchmarking is not important to me" .. But the same fanboy obviously "cared" a great deal about Geekbench as a "usecase" ....

[George Sobala]

I indeed have a use case.

My MBA is light, cool, completely quiet and runs forever. It delivers far more chess than I know what to do with for casual analysis. Should I desire even more chess power I just run remote engines on my properly cooled desktop, which apparently can run as many as 20 threads (who knew?!).

I have a Mac, a Windows system, and indeed several Linux boxes too. I am aware of the pros and cons of different technological solutions. And I would hate to have anything else on my lap than my M1 MBA.

I am genuinely interested in what other people's use cases are that require cutting edge laptop hardware. I can think of (a) playing against other computers on Playchess etc (vanity project and you don't stand a chance with a laptop), (b) simply for a sense of superiority (vanity), (c) Super GM prepping openings for a tournament (wrong hardware), (d) Serious Correspondence Chess (wrong hardware). For normal mortals, the analysis provided by SF15-nnue on an MBA is way beyond their chess comprehension; I am uncertain what extra nps offer. Can you enlighten me?

[Modern Times]

For normal mortals, the analysis provided by SF15-nnue on an MBA is way beyond their chess comprehension; I am uncertain what extra nps offer. Can you enlighten me?

You're not missing anything. it is the brainwashed Apple users who proclaim loudly that SF15 on their M1/M2 is way faster than anything AMD or Intel can offer that are the problem. It is a problem with Apple users generally, who seem to be brainwashed that anything and everything Apple produce is automatically the best there is. Sometimes that is true but often not.

[smatovic]

I indeed have a use case.

My MBA is light, cool, completely quiet and runs forever. It delivers far more chess than I know what to do with for casual analysis. Should I desire even more chess power I just run remote engines on my properly cooled desktop, which apparently can run as many as 20 threads (who knew?!).

I have a Mac, a Windows system, and indeed several Linux boxes too. I am aware of the pros and cons of different technological solutions. And I would hate to have anything else on my lap than my M1 MBA.

I am genuinely interested in what other people's use cases are that require cutting edge laptop hardware. I can think of (a) playing against other computers on Playchess etc (vanity project and you don't stand a chance with a laptop), (b) simply for a sense of superiority (vanity), (c) Super GM prepping openings for a tournament (wrong hardware), (d) Serious Correspondence Chess (wrong hardware). For normal mortals, the analysis provided by SF15-nnue on an MBA is way beyond their chess comprehension; I am uncertain what extra nps offer. Can you enlighten me?

Following this line of arguments, a 300 bucks budget laptop or a 50 bucks Rapsberry Pi would do the job?

--
Srdja

[Werewolf]

I am genuinely interested in what other people's use cases are that require cutting edge laptop hardware. I can think of (a) playing against other computers on Playchess etc (vanity project and you don't stand a chance with a laptop), (b) simply for a sense of superiority (vanity), (c) Super GM prepping openings for a tournament (wrong hardware), (d) Serious Correspondence Chess (wrong hardware). For normal mortals, the analysis provided by SF15-nnue on an MBA is way beyond their chess comprehension; I am uncertain what extra nps offer. Can you enlighten me?

It’s a good point.
I’m an Aquarium user who likes to study openings. Over the last few years I’ve accumulated around 150 cores on eBay machines, and networked together, they analyse in IDeA.

Each core analyses a single position for around 15 minutes. And the opening systems have many positions. It takes it about 3 months to build up a meaningful tree of positions.

Right now I’m looking into the King’s Indian and am somewhat shocked that my latest book on this opening (Grandmaster Repertoire 2022 no less) contains losing variations from the very first chapter.

To attempt this on a laptop would be unthinkable, especially a Mac.

[wickedpotus]

Following this line of arguments, a 300 bucks budget laptop or a 50 bucks Rapsberry Pi would do the job?

--
Srdja

This!!

Why pay upwards of 3-4000 USD for an "M1 Max" laptop if your use case is something you can do just as well on a cheap tablet, phone, or netbook for less money and with even greater battery stamina?

I like to have a laptop powerful enough to game, analyze, develop, build and test on (and maybe even do so NN training), that is open and flexible (for example to plug in an external GPU, add some extra RAM or bigger SSD for example). I don't see the attractive proposition of paying a premium for a slower less capable computer. This is not "Apple hate" this is just common sense in my view.