Hacker News Reader

Comment by mogoh | Hacker News Reader

> Apple's Arm cores now have significantly higher performance per clock cycle than any traditional x86 PC processor.

I am not an expert but this claim sounds way to high.

At the time of the M1 launch, Apple's chips preformed more instructions per clock cycle than Intel or AMD chips.

> What really defines Apple’s Firestorm CPU core from other designs in the industry is just the sheer width of the microarchitecture. Featuring an 8-wide decode block, Apple’s Firestorm is by far the current widest commercialized design in the industry.

https://www.anandtech.com/show/16226/apple-silicon-m1-a14-de...

They also ran the clocks slower for power efficiency.

ARM is also a bit of cargo cult. Apple CPUs are efficient and performant, other ARM CPUs not so much. If it's really the ISA or other "secret" Apple sauce, which makes all the difference, is not completely clear.

Also some things to consider is that apple dropped the majority of legacy software support and runs native arm64 code for a majority of apps.

also somewhere I saw that amd64 is better performance per watt than m2 or m3

I have a 35W AMD 5750GE (zen 3, 8c16t), along with M1, M1 Max and M2 Pro devices - the AMD processor and M1/M2 chips trade blows depending on workload.

It’s absolutely possible to get x86_64 to the same power/performance envelope as apple’s M series chips :)

> It’s absolutely possible to get x86_64 to the same power/performance envelope as apple’s M series chips :)

And already very close to possibly passing ARM. We will see.

You have to admit, “secret Apple sauce” does sound rather tasty.

It sounds way too high because the claim itself is weird. It doesn't claim that arm cores are faster than x86 cores outright, but rather they're faster if you normalize by clock cycle. Also depending on the benchmark chosen, you get very different results. If you combine the two factors the claim is probably true.

https://www.cpu-monkey.com/en/compare_cpu-apple_m3_max_16_cp...

the claim probably does not include server processors, if that may make a difference.

Server cores don’t tend to have more perf per clock on a single core. They scale by having multiple cores.

So I’m not sure this matters because the statement in question is per core.

Reply on news.ycombinator.com