General discussion • Re: A Pi Pie Chart

As far as I can tell there's nothing terribly special about Apple M-class processors other than they use the latest (read expensive) TSMC fabrication process. This enables better power efficiency and higher clock speeds, along with lower yields.

They're fast, and they're getting faster with each iteration. Single-core performance on x86 has broadly stagnated over the past few years.

The clever stuff with the M-class chips is actually the hardware assistance for the x86 emulation: having those extra status flags (the BCD thing, and the parity bit of the last calculation, for example), and the strict memory semantics when emulating x86 code. Quite cute.

The Geekbench 6 single-core speeds reported at

https://www.tomsguide.com/features/appl ... benchmarks

as well as

https://browser.geekbench.com/

indicate

Code:

Geekbench 6 Single Core Results            Score  Ratio  ReleasedPi 4 1.8GHz   293   0.13  Jun 2019Pi 5 2.4GHz   777   0.33  Sep 2023Pi 5 2.9GHz   881   0.37  Sep 2023Ryzen 1700   1103   0.47  Mar 2017Ryzen 3700X  1751   0.75  Jul 2019Ryzen 5800   2096   0.89  Jan 2021M1           2334   1.00  Nov 2020M2           2575   1.10  Jun 2022Ryzen 7700   2924   1.25  Jan 2023M3           3093   1.32  Oct 2023

It would appear single-core x86 results actually made more progress recently and nearly caught up with the M3.

Comparing the Geekbench 6 results for the 24-core M2 Ultra Mac Studio with the 24-core Ryzen Threadripper 7965WX yields 21450 versus 23654 respectively based on

https://www.cpu-monkey.com/en/cpu_bench ... multi_core

Weirdly the M1 Ultra is listed at 23970 so it's not clear how reliable those results are. According to the dog developer some carefully computed Pi ratios would be easier to understand. Does anyone reading this have a Mac Studio?

Statistics: Posted by ejolson — Mon Apr 01, 2024 7:08 pm

---