Hacker News Reader

Comment by tiffanyh | Hacker News Reader

How will this compare per watt?

The Apple M-series is 3x more performant per watt than anything else.

Perf Per Watt (bigger is better):

  M1          298
  M2          297
  AMD 5600U    90
  AMD 6800U    78
  Intel 1240P  64

https://www.notebookcheck.net/Apple-M2-SoC-Analysis-Worse-CP...

View on HN

For those just reading the comments, the article mentions the 7000 series chips and the numbers here reflect the prior generation.

What are the numbers for the 7000 chips on this list?

The tdp for the m2 is 20 watts. The article mentions a 15 watt tdp for the AMD chip. We will have to wait to see when the chips are released but it appears performance per watt will be at a minimum competitive with the m2.

I’d maintain a certain amount of scepticism until independent benchmarks are out. Comparison by tdp, particularly between manufacturers, tends to be… less than useful; if nothing else there is no common definition of tdp.

As mentioned elsewhere in the comments, there was an Intel chip that appeared to beat the M1 on performance per watt if you looked at ‘tdp’, the issue being that Intel was using a particularly eccentric definition of tdp.

Many motherboards now report the power (Watts) drawn by the CPU.

If you don't trust that, you can always measure the whole system usage at the plug.

By then you can run your own benchmarks, which was the point of “until independent benchmarks are out”. You need some skepticism when discussing characteristics they give for an upcoming chip.

Many benchmarks today actually do measure these things. It could be improved through.

The point is, fundamentally, that you shouldn't take a manufacturer's word on real-world behaviour, particularly for something that isn't even released yet. Like, you can if you want, but going on history you will nearly always be disappointed.

> Many benchmarks today actually do measure these things

Except, of course, for chips that are not available yet, such as those discussed here.

Sure, and when the chip is available and that test can be done independently, that’ll largely answer the question. For now, it’s pretty much just marketing claims.

I don't have a citation on hand (though I love GamersNexus, their focus on video makes it hard to find citations for such things), iirc TDP is actually not comparable between processor generations, let alone manufacturers. Intel and AMD calculate it differently, and AMD's calculation includes the thermal solution attached to the chip, so hypothetically they can make it almost whatever number they want.

Right, that was the first thing I looked at and I don’t understand the article author’s logic: compare performance to M2, then don’t compare the 25% lower power (admittedly stated) consumption.

Just a super weird writing choice.

Also puts to bed the “x86 is fundamentally unable to be as low power as arm” rationale for the poor intel and amd perf/watt. I think the sole reason for that is that they never _really_ competed on that metric.

M1/M2 is a big.little design isn't it? As in an ultra low power CPU for trivial tasks and it spins up heavyweight cores when the workload demands it?

I mean, 3x the performance per watt might be --lies-- --damn lies-- uh, benchmarks targeting the little CPU?

This is a reasonable guess but I don't think it's correct. Most benchmarks appear to compare P-cores vs. P-cores.

What's the point of an efficiency benchmark that only compares P-cores?

If the P-cores are where the real workload runs, then the efficiency of the P-cores obviously is paramount. P-cores aren't just for running off wall power.

Most people's computers don't have "real workloads" in that sense because a lot of things aren't CPU bound.

Are you assuming that P-cores only get powered on after E-cores are fully saturated?

I only meant "real workload" in the sense that when real people are using their computer, the P-cores are doing real work; that's true even for bursty workloads that don't keep the processor 100% busy, so power is being spent by the P-cores and their efficiency matters.

> Are you assuming that P-cores only get powered on after E-cores are fully saturated?

No, that's not how M1/M2 work.

Apps are generally on the P-cores, but daemons are on E-cores, and the daemons are doing essential work like keeping you connected to wifi. And the apps aren't running all the time if you're just browsing or editing text. It's all part of the end goal of having a lot of battery life.

And then you have Teams and a large Miro board open :-D

SCNR!

> M1/M2 is a big.little design isn’t it?

Isn’t almost everything designed for consumer use, even the Intel i9, these days?

AMD isn't.

M1 value seems wrong, should be 407

Different test:

Cinebench R23 Multi Package Power Efficiency

M1 - 561 pts/watt

M2 - 433

AMD Ryzen 7 6800U - 374

Intel Core i5-1240P - 258

Those are the "multi package" numbers. Though exactly what "single" vs "multi" means in this context is unclear to me.

"Cinebench R23 Multi" and "Cinebench R23 Single" are the software workloads used (multi-thread and single-thread). "Package Power" specifies (approximately) where the power is measured and what components are included—but reviewers like NotebookCheck aren't equipped to actually measure power at the chip or package level, so they're actually relying on software estimates that add another layer of vendor-specific sources of error on top of the inconsistencies about what's in the package (eg. Apple puts DRAM in the package but Intel and AMD don't).

Cinebench uses Intel Embree, which is library that is hand optimized for x86 instructions.

It should not be used to compare ARM vs x86 CPUS. Geekbench is better here.

If you cite notebookcheck, at least cite the article where they compare a processor of the same gen... https://www.notebookcheck.net/AMD-Ryzen-9-7940HS-analysis-Ze...

TLDR: The performance per watt gap is gone, but only when you compare at similar TDPs.

And performance per watt is what (ha!) matters most on a laptop as users demand both - high performance and a long battery life.

Nice try, apple fanboy. Your number are for the prevoius gen.

Reply on news.ycombinator.com