M1 Apple Silicon for Chess?

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

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
User avatar
Steve Maughan
Posts: 1080
Joined: Wed Mar 08, 2006 7:28 pm
Location: Florida, USA
Contact:

M1 Apple Silicon for Chess?

Post by Steve Maughan » Tue Nov 24, 2020 3:21 pm

Has anyone come any test attempting to measure how good the new Apple M1 Silicon will be for running chess engines?

Steve
http://www.chessprogramming.net - Maverick Chess Engine

Peperoni
Posts: 45
Joined: Sun Nov 01, 2020 4:27 pm
Full name: Richard Porti

Re: M1 Apple Silicon for Chess?

Post by Peperoni » Tue Nov 24, 2020 3:34 pm

From what I understood, the performance are good for single core, for multicore, it looks very average to my eyes.

Jhoravi
Posts: 285
Joined: Wed May 08, 2013 4:49 am

Re: M1 Apple Silicon for Chess?

Post by Jhoravi » Tue Nov 24, 2020 3:44 pm

As far as I have observed chess engine speeds for 2 decades I noticed that it benefits a lot from larger L1 cache size. For the record, our mainstream x86 CPUs have 32kb/32kb L1 cache while this new apple M1 has gigantic 192kb/128kb!

Peperoni
Posts: 45
Joined: Sun Nov 01, 2020 4:27 pm
Full name: Richard Porti

Re: M1 Apple Silicon for Chess?

Post by Peperoni » Tue Nov 24, 2020 3:46 pm

Are some benchmarks available?

Dicaste
Posts: 108
Joined: Mon Apr 16, 2012 5:23 pm
Location: Istanbul, TURKEY

Re: M1 Apple Silicon for Chess?

Post by Dicaste » Tue Nov 24, 2020 3:48 pm

I ordered mac mini but didn't reveived yet. I'll do some testings and give feedback.

Milos
Posts: 4017
Joined: Wed Nov 25, 2009 12:47 am

Re: M1 Apple Silicon for Chess?

Post by Milos » Tue Nov 24, 2020 4:03 pm

Jhoravi wrote:
Tue Nov 24, 2020 3:44 pm
As far as I have observed chess engine speeds for 2 decades I noticed that it benefits a lot from larger L1 cache size. For the record, our mainstream x86 CPUs have 32kb/32kb L1 cache while this new apple M1 has gigantic 192kb/128kb!
Zen3 has for example 64kiB per core, but that's irrelevant since M1 doesn't have L3 cache at all, so essentially what is L3 on x86 is L2 on M1 and what is L2 on x86 is L1 on M1. 5950X for example has 8MB of L2 cache and 64MB of L3 (M1 total 16MB of equivalent L2).

To put things into perspective for Apple sheep, I would be surprised if single core performance on SF-NNUE would be more than 1.2Mnps from a starting position. Overall for 8 cores 8Mnps at best, i.e. far cry from a modern 4 core desktop CPU.

Ras
Posts: 1763
Joined: Tue Aug 30, 2016 6:19 pm
Full name: Rasmus Althoff
Contact:

Re: M1 Apple Silicon for Chess?

Post by Ras » Tue Nov 24, 2020 4:14 pm

Milos wrote:
Tue Nov 24, 2020 4:03 pm
I would be surprised if single core performance on SF-NNUE would be more than 1.2Mnps from a starting position. Overall for 8 cores 8Mnps at best
Only four out of the eight M1 cores are for heavy calculations. The remaining four cores are small ones for saving battery in near-idle. The full performance will be rather around four times the single thread performance, not eight times.

Then again, there's no point in comparing a 10W mobile APU to a 65W desktop CPU, let alone to the even bigger ones. It stacks up pretty well however against AMD's 4000U series.
Rasmus Althoff
https://www.ct800.net

Milos
Posts: 4017
Joined: Wed Nov 25, 2009 12:47 am

Re: M1 Apple Silicon for Chess?

Post by Milos » Tue Nov 24, 2020 4:35 pm

Ras wrote:
Tue Nov 24, 2020 4:14 pm
Milos wrote:
Tue Nov 24, 2020 4:03 pm
I would be surprised if single core performance on SF-NNUE would be more than 1.2Mnps from a starting position. Overall for 8 cores 8Mnps at best
Only four out of the eight M1 cores are for heavy calculations. The remaining four cores are small ones for saving battery in near-idle. The full performance will be rather around four times the single thread performance, not eight times.
I calculated 4x1.2+4x0.8. Ofc it could be that "energy" cores give only 0.5Mnps, but that would still give a total of about 7Mnps.
Then again, there's no point in comparing a 10W mobile APU to a 65W desktop CPU, let alone to the even bigger ones. It stacks up pretty well however against AMD's 4000U series.
That depends on which 4000U series CPU you take into comparison.
4900U will almost certainly blow M1 out of the water.

Ras
Posts: 1763
Joined: Tue Aug 30, 2016 6:19 pm
Full name: Rasmus Althoff
Contact:

Re: M1 Apple Silicon for Chess?

Post by Ras » Tue Nov 24, 2020 5:04 pm

Milos wrote:
Tue Nov 24, 2020 4:35 pm
That depends on which 4000U series CPU you take into comparison.
4900U will almost certainly blow M1 out of the water.
I don't think so, even if the 4900U were not made of pure unobtanium. Geekbench 5 lists the M1 around 7500 in multicore. The fastest 4800U samples are around 7700 (generally rather in the 5000s or 6000s). It depends a lot on what RAM the machine is using (anything from DDR4-2666 to LPDDR4-4266), and the configured cTDP which can be 25W. The 4900U has only 100MHz more boost clock, 4.3GHz instead of 4.2GHz, but the base clock is the same.
Rasmus Althoff
https://www.ct800.net

Raphexon
Posts: 367
Joined: Sun Mar 17, 2019 11:00 am
Full name: Henk Drost

Re: M1 Apple Silicon for Chess?

Post by Raphexon » Tue Nov 24, 2020 5:26 pm

Ras wrote:
Tue Nov 24, 2020 5:04 pm
Milos wrote:
Tue Nov 24, 2020 4:35 pm
That depends on which 4000U series CPU you take into comparison.
4900U will almost certainly blow M1 out of the water.
I don't think so, even if the 4900U were not made of pure unobtanium. Geekbench 5 lists the M1 around 7500 in multicore. The fastest 4800U samples are around 7700 (generally rather in the 5000s or 6000s). It depends a lot on what RAM the machine is using (anything from DDR4-2666 to LPDDR4-4266), and the configured cTDP which can be 25W. The 4900U has only 100MHz more boost clock, 4.3GHz instead of 4.2GHz, but the base clock is the same.
I'll believe it when I see it.

I think ARM chips have always underpeformed their Geekbench score for chess.
And for NNUE engines NEON is a downside compared to AVX2.

Post Reply