Engines compiled for Windows 10 | 11 ARM64 Neon

[AlexChess]

Since most rating lists, also having Ryzen 5900 available, are testing engines using only 1 CPU instead of 24, my 4 of 8 fast CPUs and 2 to 6 kN/s are enought for testing...





And I've observed that using different time controls and CPU AVX512 plus plus plus or the monstrous TCEC computing power, the relative rankings are very similar: Stockfish and derivatives on the edge, then all the others . So also reserving only 4 CPUs to my chess passion, Mac Mini M1 is still fast for all the remaining tasks that I need.

[wickedpotus]

But the cpus use hyperthreading = threads and the M1 not, it uses true cores.
Hyperthreading increase the speed by up to 1% but in the same time it shows the double kn/s.
It should be clear that if you have up to 1% speed increase you can’t have doubled kn/s.
= That’s not even a 1 elo increase.
That means 5900HS have only 17,5M using his true cores.
Stockfish on M1 is slow because it’s not designed for the M1. Stockfish developers and M1 professionals confirmed that.
Stockfish is designed for Intel and AMD.
Stockfish could be up to 3 times faster on M1, but every developer can decide what he wants to improve and the developers are working more on the elo strength than making Stockfish 3 times faster on M1. The most have not the M1 MacBook so it is very difficult to check and improve tests and the speed on these devices.
M1X will have 20M.

Also, it seems a large chunk of Apple's user base are sort of religious zealots who do not even try out nor seriously compare other alternatives before proclaiming "truths", and proclaiming anyone bringing up issues or limitations of Apple products to be "haters"...

This is sort of OT.. But I guess it requires an answer

Your statement is somewhat misleading.

The graph above is the max throughput the HW provides with current state of the art compiles (special optimized compiles with NEON support for M1). Period. Your statements about speed up and kn/s are simply incorrect and nodes per seconds are always nodes per seconds. Trying to obscure the performance characteristics by not comparing the max throughput of the hw and only compare under some artificial conditions (like limited-cores, hyperthreading disable etc. are very dishonest, if you want to compare engine speed)

Your statements on stockfish design is moot. its open source. Any Apple pundit are free to compile and enhance it (if it is even possible).. But keep in mind Apple is not the most "open" and "helpful" company in assisting with tools and useful documentation of it's ARM-type CPUs.

I also take your statements that SF could be 3 times as fast on M1 as irrational fanboy speak.. Just as your hyperthreading "theory" this has absolutely no basis in reality.

M1X.. Lol no comment...

[AlexChess]

I don't get it Alex, why are you pushing so hard for devs to support overpriced, vendor-locked-in proprietary hardware? What's in it for you?

If an engine x64 works fine and fast on Windows 11 ARM64, no need for me to try to natively compile it. Almost all popcont /sse3 | sse41 engines work perfectly.
I use Windows 11 ARM64 under Parallels Desktop M1 because not all top engines have been compiled for Mac M1.

Regards, Alex

[AlexChess]

I don't get it Alex, why are you pushing so hard for devs to support overpriced, vendor-locked-in proprietary hardware? What's in it for you?

Why you are pushing so hard against Apple and their hardware? What’s in it for you?

Nothing in it for me. I don't have strong emotions about a "brand"/"logo". I probably coded asm on Apple IIe, Macs, etc. before you even got your first computer

I have nothing against their HW per se, their price/performance though is rather poor, and I don't like their embrace of Security by Obscurity and closed-system approach. I don't like their pricing, don't like their un-openness and locked-in eco-system... Don't like their big-brother mentality and I think their un-openness goes all the way from soldering SSDs to the motherboard to using DRM to mess with 3rd party hw support to of course using DRM on both, HW, OS, and application layers to lock users into their surveillance-heavy, commercial eco-system, to squeeze more money out of em with not so honest methods imop.

Also, it seems a large chunk of Apple's user base are sort of religious zealots who do not even try out nor seriously compare other alternatives before proclaiming "truths", and proclaiming anyone bringing up issues or limitations of Apple products to be "haters"...

Also the current state of "ARM" type value-proposition for chess-lovers is currently not a really good one if we go back to the original premise of the thread. (best use-case for ARM based engines today is probably Android Phones or linux, not Windows, Mac nor iOS...).

Personally I like every operating system: Linux, Mac, Windows and Android on mobile and I use all them because I'm A 360° technology fan, not a specific-brand fanboy

[AlexChess]

ARM64EC TO EASILY RE-COMPILE CHESS ENGINES ON WINDOWS 11 ARM NATIVE:

https://blogs.windows.com/windowsdevelo ... 11-on-arm/

Thank you very much.

Hi!
If you manage to compile them, let me know! I'd like to learn how to compile engines for myself on Windows 11 ARM64, Mac M1 and Linux Ubuntu ARM64 | Raspberry PI 4. We could create some tutorials on Talkchess for the 4 systems (Ubuntu & Raspberry PI 4 share the same engines, the 4th is Android ) with all tools needed to natively compile all open source engines.

Many thanks to you for your positive behavior, from other members here I receive only unexpected critics for not being an orthodox top PC chess programmer or user

Best regards, Alex

[AlexChess]

Combusken 1.99_dev for Windows 11 ARM64 compiled by Marcin on request!!!

Download engine:
https://1drv.ms/u/s!AkW3Hj0Gl_ew0TfO5sM ... G?e=kAQOjE

Combusken 1.99_dev on my Promising 3000+ ELO engines not Stockfish derivatives
forum3/viewtopic.php?p=906289#p906367

Just added to my SuperBlitz as 62th engines. Soon I will publish the new crosstable.

Here the 61 PGN games:
https://onedrive.live.com/?cid=B0F79706 ... 69&o=OneUp

Ciao, Alex