M1 Apple Silicon for Chess?

[acepoint_de]

https://acepoint.de/some-chess-stuff-for-apples-m1/

Code: Select all

Ethereal-12.75:       Mach-O 64-bit executable arm64
cfish-17022021:       Mach-O 64-bit executable arm64
crystal_3.1:          Mach-O 64-bit executable arm64
cutechess-cli:        Mach-O 64-bit executable arm64
igel-2.9.0:           Mach-O 64-bit executable arm64
lc0-0.26.3:           Mach-O 64-bit executable arm64
marvin-5.0.0:         Mach-O 64-bit executable arm64
stockfish-12:         Mach-O 64-bit executable arm64
stockfish-17022021:   Mach-O 64-bit executable arm64
sugar-AI-ICCF-140a:   Mach-O 64-bit executable arm64

+ chessx and cutechess, enjoy. More might follow.

Ciao

acepoint

[AlexChess]

Thank you!
I think that M1, M2, M3... will be a real Dream Chess Machines. We must convert all top enginess to Mac M1

Here you find other M1 chess engines:

https://1drv.ms/u/s!AkW3Hj0Gl_ewzxEJ1bA ... K?e=uHWvbt

Black-Diamond-XI
Blue-Honey-FD-XI
Blue-Honey-XI
Bluefish-FD-XI
Bluefish-XI
cfish-9-iMac-mod
Honey-FD-XI
Honey-XI
RubiChess
shashchess_clang
shashchess_macOS_gcc
Stockfish-XI
xiphos-sse

[acepoint_de]

Here you find other M1 chess engines:

The list and the zip file don't contain arm64 engines but (only) x86_64. They will probably run via Rosetta 2 but with less power.

My goal is to build native ones.

Ciao

acepoint

PS. Just an example:

Code: Select all

stockfish-12: Mach-O 64-bit executable arm64
stockfish-12 bench
...
Total time (ms) : 1626
Nodes searched  : 3624569
Nodes/second    : 2229132

stockfish-osx: Mach-O 64-bit executable x86_64
stockfish-osx bench
...
Total time (ms) : 2447
Nodes searched  : 3624569
Nodes/second    : 1481229

[AlexChess]

Yes they are intel macOS, slower than native M1 but they work perfectly with Rosetta 2 and BanksiaGUI (just testing now)

Your idea to compile all them M1 native is absolutely appreciated by me, please let me know when there are new M1 engines.

Engines still missing:
CorChess, Raubfisch
Alpha subzero 130.2 (it's very strong also calculating very slow, anyway 4 times faster than LC0 for CPU, but I think the source isn't available)
Fisherov 0.96, KillFish 1.4 Hybrid, Zeus 11, AI-15, Eman 6.90 (same problem of Alpha subzero)

Ciao, Alex

[acepoint_de]

A few benchmarks: https://acepoint.de/engine-benchmarks-on-apples-m1/

Ciao

acepoint

[Ckappe]

Why not by a computer to run chess instead of a proprietary poser toy?

Perhaps we could have a new benchmark: how many giganodes of analysis can your device do on one battery charge? No power cord allowed.

Sound like an interesting test...(something Mac posers usally are not fond of.. They only want to do 20 seconds apple to oranges benches with geekbench in my experience Post your number and I will do test under same condition on my new AMD laptop, , Shall we use LC0 as a bench? I am extremely confident that my laptop will do more positions than an M1 before power runs out... given my large battery and much faster machine..

But how long are you really away from a powerplug, when doing long-time analysis to be honest...

I choose stockfish as the engine as it is the strongest. (Yeah and of course I have CPU power rather than GPU). Happy to try it out in the next couple of days.

Suggest go infinite from the start position with default hash and however many threads you want.

SF is likely not the strongest on a weak ARM hardware like the M1 compared to laptops with 20xx 30xx graphics.<

So dazzle me how many nodes (from start-pos) can your laptop M1 run before your battery runs out with latest SF and default Network

And btw the strained argument of a "long flight" is also a bit moot as most longhauls these days have 110/220v outlets by the seats...

[George Sobala]

SF is likely not the strongest on a weak ARM hardware like the M1 compared to laptops with 20xx 30xx graphics.<

So dazzle me how many nodes (from start-pos) can your laptop M1 run before your battery runs out with latest SF and default Network

And btw the strained argument of a "long flight" is also a bit moot as most longhauls these days have 110/220v outlets by the seats...

60G to 5% charge left.

[AlexChess]

A few benchmarks: https://acepoint.de/engine-benchmarks-on-apples-m1/

Ciao

acepoint

BANKSIAGUI BENCHMARKS (MY MAC MINI M1 RESULTS SEEM FASTER THAN YOURS ON MACBOOK M1, OR I'M WRONG? (BUT I HAVE 4 CORES ON GUI SETUP)

name nodes/m NPS depth/m time/m moves time #fails
1. Cfish 170221 64 NEON 30915K 8747521 33.4 3.5 67.2 237.5
2. SugaR AI ICCF 1.40a 14264K 3422162 21.1 4.2 47.0 195.9
3. Honey XI 35827K 9416042 31.9 3.8 56.4 214.6
4. Blue Honey XI 36606K 9466255 32.2 3.9 51.0 197.2
5. Bluefish XI FD 38944K 9144635 25.9 4.3 38.2 162.7
6. Bluefish XI 35381K 9063500 30.2 3.9 41.8 163.2
7. Honey XI FD 37644K 9749500 22.7 3.9 54.4 210.0
8. Xiphos 0.6 SSE 38237K 10011729 26.1 3.8 54.2 207.0
9. Igel 2.9.0 CUSTOM 25480K 6303678 22.5 4.0 37.8 152.8
10. Senpai 2.0 1477K 355417 15.8 4.2 41.6
11. Lc0 v0.26.3+git.dirty 11K 3006 6.7 3.7 71.4 265.7
12. Marvin 5.0.0 14540K 3743638 20.8 3.9 63.4 246.2
13. Texel 1.08a13 32876K 8633367 19.6 3.8 51.0 194.2

[acepoint_de]

BANKSIAGUI BENCHMARKS (MY MAC MINI M1 RESULTS SEEM FASTER THAN YOURS ON MACBOOK M1, OR I'M WRONG? (BUT I HAVE 4 CORES ON GUI SETUP)

You are comparing apples and not oranges but steaks. What do you guess will be the result for 1 core vs 4 and arm64 (native) vs x86_64 (Rosetta 2)? And your keyboard seems to be broken .

Ciao

acepoint

[Ckappe]

SF is likely not the strongest on a weak ARM hardware like the M1 compared to laptops with 20xx 30xx graphics.<

So dazzle me how many nodes (from start-pos) can your laptop M1 run before your battery runs out with latest SF and default Network

And btw the strained argument of a "long flight" is also a bit moot as most longhauls these days have 110/220v outlets by the seats...

60G to 5% charge left.

How long did it take? (just starting a test on my zenbook duo laptop to compare as it does 17mnps (on batter powered) I expect 60G will be reached in an hour.. I'll measure how much % of battery that was drained