MacBook Pro 16-inch M1 Max Stockfish Benchmarks

[JimmyjamesEU]

Alderlake is newer than Apple Silicon.. I think you are confusing CPU architecture with OP-codes.. You could apply same reasoning to M1, that it is the same as an 1987 Acorn and can't be optimized for further

Alderlake isn’t an entirely new architecture. Optimisation for it is well understood. I appreciate your honesty and openness about your confusion though.

 

[Leifi]

Alderlake isn’t an entirely new architecture. Optimisation for it is well understood. I appreciate your honesty and openness about your confusion though.

I am pretty sure you should educate yourself more about CPU architectures... before discussing furtther,

Intel Unleashes Alder Lake Architecture: 50% Increase In MT Performance For Just $589 USD MSRP

Intel has finally unleashed a sub-10nm (officially the Intel 7 process, which is comparable to TSMC 7nm) desktop CPU with its Alder Lake processors and the company’s multi-year slippage at 14nm is officially over. The Alder Lake architecture also features a plethora of innovative changes...

wccftech.com

 

[leman]

Yeah.. I just ran a test on a laptop here (5900HX) 8-core..

using 14.1 st avx2 i get:

AMD Ryzen 9 5900HX with Radeon Graphics, 3301 Mhz, 8 Core(s), 16 Logical Processor(s)
stockfish_14.1_avx.exe bench 1024 16 24
===========================
Total time (ms) : 66452
Nodes searched : 2126455531
Nodes/second : 31999873

compares pretty favorably to Krevnicks bench

M1 MAX -
/stockfish bench 1024 10 26 default depth nnue
===========================
Total time (ms) : 76914
Nodes searched : 1192240595
Nodes/second : 15500956

You are comparing two different networks. Can you post Ryzen results using nnue?

 

[Leifi]

You are comparing two different networks. Can you post Ryzen results using nnue?

These are default bench (default on bench option ,nnue).. and the network is built into the exe.

 

[leman]

These are default bench (default on bench option ,nnue).. and the network is built into the exe.

I don't think the default bench is with nnue. I get very different results when running bench with and without nnue on my i9-9880H

./stockfish bench 1024 16 26 default depth nnue
===========================
Total time (ms) : 287152
Nodes searched  : 2757046678
Nodes/second    : 9601349

./stockfish bench 1024 16 26 default depth
===========================
Total time (ms) : 263606
Nodes searched  : 3037480753
Nodes/second    : 11522805

Still I have to say I do not understand your results. Ryzen 9 5900HX is on average around 20-30% faster than my Coffee Lake i9-9880H (and up to 50% faster in extreme cases such as Cinebench multi). How are you getting 3x the score??

Edit: I just noticed that you are using different benchmark settings than everyone else. Can you please run your benchmarks with the same settings as the usual bench platforms (e.g. http://ipmanchess.yolasite.com/amd--intel-chess-bench-stockfish.php)? For your CPU (16 threads) it will be

bench 1024 16 26 default depth nnue

 

[Leifi]

Yea. I went back and tested - checked.. seems you are indeed right about nnue not beiong used in the default bench,.. running with 14.1 default depth nnue it is lower. (but still much superior to M1 Max)

So a fair comparison should be

5900HX
stockfish_14.1_avx.exe bench 1024 16 26 default depth nnue
===========================
Total time (ms) : 133268
Nodes searched : 3108347209
Nodes/second : 23324015

M1 MAX -
/stockfish bench 1024 10 26 default depth nnue
===========================
Total time (ms) : 76914
Nodes searched : 1192240595
Nodes/second : 15500956

A healthy 50% increase over M1 Max (and as mentioned earlier more than 2x the performance od M1)


and without NNUE:

5900HX
stockfish_14.1_avx.exe bench 1024 16 24
===========================
Total time (ms) : 66452
Nodes searched : 2126455531
Nodes/second : 31999873

M1 Max
./stockfish bench 1024 10 26 default depth (P+E Cores w/o NNUE)
===========================
Total time (ms) : 86774
Nodes searched : 1737632105
Nodes/second : 20024801

The diff is about 60% there.

 

[Leifi]

Yea I checked.. seems you are indeed right, about not using nnue per defualr.. running with 14.1 default depth nnue it is lower...

So a fair comparison should be

5900HX
stockfish_14.1_avx.exe bench 1024 16 26 default depth nnue
===========================
Total time (ms) : 133268
Nodes searched : 3108347209
Nodes/second : 23324015
M1 MAX -
/stockfish bench 1024 10 26 default depth nnue
===========================
Total time (ms) : 76914
Nodes searched : 1192240595
Nodes/second : 15500956


and without NNUE:

5900HX
stockfish_14.1_avx.exe bench 1024 16 24
===========================
Total time (ms) : 66452
Nodes searched : 2126455531
Nodes/second : 31999873

So i guess the conclusion is that it is 2-3 times faster than M1, and about 50% faster than M1 Max, after M1 optimizations... right?

 

[leman]

Yea I checked.. seems you are indeed right, about not using nnue per defualr.. running with 14.1 default depth nnue it is lower...

So a fair comparison should be

5900HX
stockfish_14.1_avx.exe bench 1024 16 26 default depth nnue
===========================
Total time (ms) : 133268
Nodes searched : 3108347209
Nodes/second : 23324015
M1 MAX -
/stockfish bench 1024 10 26 default depth nnue
===========================
Total time (ms) : 76914
Nodes searched : 1192240595
Nodes/second : 15500956


and without NNUE:

5900HX
stockfish_14.1_avx.exe bench 1024 16 24
===========================
Total time (ms) : 66452
Nodes searched : 2126455531
Nodes/second : 31999873

So i guess the conclusion is that it is 2-3 times faster than M1, and about 50% faster than M1 Max, after M1 optimizations... right?

I have to say, I still don't understand why your scores are so high. They are almost 2x higher than the scores of i7-11800H on http://ipmanchess.yolasite.com/amd--intel-chess-bench-stockfish.php, a CPU that otherwise performs within 5% of 5900HX! What kind of laptop is that? Is your CPU overlocked or otherwise non-stock? Either Zen is just ridiculously good for Stockfish (which does not seem to be the case looking at results from openbenchmarking.org) or your 5900X is running with desktop-like configuration.

 

[Leifi]

I have to say, I still don't understand why your scores are so high. They are almost 2x higher than the scores of i7-11800H on http://ipmanchess.yolasite.com/amd--intel-chess-bench-stockfish.php, a CPU that otherwise performs within 5% of 5900HX! What kind of laptop is that? Is your CPU overlocked or otherwise non-stock? Either Zen is just ridiculously good for Stockfish (which does not seem to be the case looking at results from openbenchmarking.org) or your 5900X is running with desktop-like configuration.

It's an Rog Duo 15-inch dual OLED laptop with RTX 3080. Running arch linux & windows.

 

[leman]

It's an Rog Duo 15-inch dual OLED laptop with RTX 3080. Running arch linux & windows.

What do you get in Geekbench and R23?

 

[Leifi]

What do you get in Geekbench and R23?

Geekbench are on my black-list but cinebench r23 just gave about 13997 mt-score on that laptop.

 

[leman]

Geekbench are on my black-list but cinebench r23 just gave about 13997 mt-score on that laptop.

Thanks for the update. It is a bit odd that your Ryzen seems to punch way about its weight wehen I look at performance of similar CPUs on Stockbench I could find online, but then again, the laptop reviews suggest that this particular Asus can run the CPU on turbo all day long. That is probably what we are seeing here. It performs on the level of an 8-core desktop Zen3 CPU because, well, it runs at the same clock and power consumption as a desktop Zen3 CPU. Still, very impressive for a laptop of this size.

 

[Krevnik]

I really appreciate your time spent.. I have been bit*ing alot in various thread that no one was willing to spend 5 minutes to bench

I did it because the developer who provided the patch was wondering if someone with an M1 Pro/Max would run it as they don't have any M1-based hardware. I'm more inclined to help someone asking for feedback/data on something constructive (in this case, new NEON optimizations that help ARM across the board) than to wade into an internet argument just to prove someone wrong/right.

If you download an exe it have been compiled for som generic versions like AVX2 or BMI. If you compile yourself you will get the best compile for your machines architecture and can use better or worse compilers and options.

AVX2 isn't "generic" though. That's one of the points folks have been harping on in the first place.

 

[Appletoni]

DennisdeWit, I am very interested as well. Just start from the original position, select in the menu Engine/Start Infinite Analysis and let it run for 5 minutes. Then take a screenshot. That would be very much appreciated.

note: it would help if you do not run anything else at the same as to not split the power of the cpu.

My M1 is not great for Stockfish, very curious to know the difference with the M1X.

We have not a single good Stockfish benchmark result.
We have got one patch with up to 5% speed improvement on nearly every os also on M1X chip.
Some Apple guys said, when looking at the Stockfish code, there is a lot what can be improved or changed.
But nothing concrete was mentioned.
Looks like only the one patch has passed and proved to be a very little speed up.
That‘s all.
It doesn‘t look good for Apple M1 MAX.
If you look at the cpu results online, it is even slow at integer math xD.

It would be great if you could ask some Stockfish developers on github, discord, fishcooking or other chess forums to improve Stockfish on MacBook Pro 16-inch M1 MAX chips.

I hope that we will get a MacBook Pro 16 or 18-inch M3 MAX chips soon, because 20 cpu cores (16+4) would be the minimum?
But at that time AMD will sell probably 32 cpu cores which you can use inside a notebook.
MacBook Pro 16 or 18-inch M2 MAX will be to weak because it only gets ARMv9 but still only 10 cores?
But we can hope for a MacBook Pro M2 MAX PLUS and M3 MAX PLUS which should be stronger and also probably be available in 18-inch.

 

[leman]

We have not a single good Stockfish benchmark result.

False

We have got one patch with up to 5% speed improvement on nearly every os also on M1X chip.

False

Some Apple guys said, when looking at the Stockfish code, there is a lot what can be improved or changed.
But nothing concrete was mentioned.
Looks like only the one patch has passed and proved to be a very little speed up.

All of this is false

If you look at the cpu results online, it is even slow at integer math xD.

False

But at that time AMD will sell probably 32 cpu cores which you can use inside a notebook.

Baseless assumption