Hacker News Reader

Comment by AnthonyMouse | Hacker News Reader

The per-core efficiency of Apple and AMD CPUs on the same process node is pretty much identical. This has become harder to directly compare because they're now using alternate process nodes from one another, but have a look at this chart for example:

https://www.cpu-monkey.com/en/cpu_benchmark-cpu_performance_...

What do we see at the top of this chart? TSMC 3nm (M3/M4), followed by TSMC 4nm (Ryzen 7000U/8000U), TSMC 5nm (M1/M2), TSMC 5nm/6nm mixed (Ryzen 7000H), and then finally we find something made on an Intel process node instead of TSMC.

The efficiency has more to do with the process node than which architecture it is.

It's too bad they don't have Epyc on that chart. Epyc 9845 is on TSMC N3E and that thing is running cores at a >2GHz base clock at less than 2.5W per core.

View on HN

You're linking to a multi-core benchmark. The story is a lot more in favor of Apple if you look at single-core efficiency, Apple is roughly 2-3x more efficient: https://www.notebookcheck.net/Intel-Lunar-Lake-CPU-analysis-...

And this benchmark doesn't even include M4, which is even more efficient.

> The story is a lot more in favor of Apple if you look at single-core efficiency

Your link is comparing the M3 against AMD chips with higher TDPs. Higher TDPs tank "single-core efficiency" because power consumption is non-linear with clock speed. Give a core near its limit three times the power budget and you're basically dividing the single-core efficiency by three because you burn three times more power and barely improve single-thread performance at all, and then that's exactly what you see there.

To have a useful comparison you have to compare the efficiency of CPUs when they're set to use the same amount of power.

Ofcourse, but that's the whole point right? At similar power levels, Apple M-series is a lot faster. The only way for AMD and Intel to compete on performance is by providing more power, but this obviously has a negative effect on efficiency. So with AMD/Intel you have to choose: efficient and slow or inefficient and fast. With Apple you can actually have it both ways.

> Ofcourse, but that's the whole point right? At similar power levels, Apple M-series is a lot faster. The only way for AMD and Intel to compete on performance is by providing more power

They're saying the exact opposite of that. Their claim is that all the extra power is only juicing performance a little bit, so at similar power levels the performance is not all that far apart.

OK, so I'm still curious about the underlying efficiency. Is there a better test of single core efficiency at the same TDP? Does Apple Silicon provide any noticeable improvement over other x86 or Arm chips in that scenario, or is it really just a straightforward function of process and node size?

Someone needs to run the same single core benchmarks on Intel and AMD while forcing those CPUs to only use 5 watts. In a few minutes of searching I was not able to find anyone doing power-limited single core benchmarks.

Reply on news.ycombinator.com