Apple M2 vs. AMD Rembrandt vs. Intel Alder Lake Linux Benchmarks

[ldesnogu]

The 512GB model is $1,500 and does come with 2 extra GPU cores, but do you know what you could get outside of Apple for that much? A whole lot more laptop.

The comparison is so stupid I'm speechless. Being a hater seems to alter cognitive capacities.

[ldesnogu]

Unless there is a matching alternative available by that time, I'll gladly get a M-based laptop to replace my ThinkPad once Asahi Linux becomes usable for day to day workloads...

I got a Lenovo P1 with Linux as work machine because my company did not have MBP available yet. That thing has to turn on its fans as soon as you plug it in even through USB C. That's very annoying and there's no fix for that. An MBP would have been so much better even if I had to have Linux virtualized. I'll switch in 2 years when the replacement is to be done :-)

[Anux]

People point out that a fanless laptop with much lower power budget has a very competitive performance against machines that require active cooling to provide the same performance.

How do you determine that all other CPUs require active cooling?
Considering the 6850 U runs with 20W there is no reason why it couldn't be cooled passive like the M2. Its just the more costly variant and most users are already happy if the fan is really quiet.

[Raka555]

Not really. These days even simple loads like "office work" (web, office suite, business apps etc.) can easily utilize at least few cores.

Each core do "single threaded performance" in isolation.
That is why a 6 core Ryzen 3600x beats a 8 core Ryzen 1800x in every respects.

SINGLE THREADED PERFORMANCE IS THE MOST IMPORT METRIC whether you have 1 core or a 1000 ... (It is sometimes called IPC - Instructions per clock)

It therefore annoys me that bench-marking websites like Phoronix only glance over single threaded performance.
I would really appreciate it if they would have a section for single threaded performance in each article, so it is easier to compare that.

[MadCatX]

How do you determine that all other CPUs require active cooling?
Considering the 6850 U runs with 20W there is no reason why it couldn't be cooled passive like the M2. Its just the more costly variant and most users are already happy if the fan is really quiet.

I conducted the following test. I have my T14 Gen1 with 4750U placed on a small wooden frame that is just slightly narrower than the laptop itself. The legs are about 10 cm tall so there is plenty of airflow from everywhere. I grabbed the latest Prime95 and ran the small FFTs test in a loop. With the ACPI profile set to "low-power", the chip temperature reached 62 °C, fan was spinning at 4000 RPM, frequency stabilized at 1.3 GHz and RAPL reported about 12 W. I then switched the power profile to "performance". Temperature shot up to 82 °C, power consumption increased to 22 W and the fan ran at full speed of 4400 RPM. Only then the CPU clock boosted to some 2.4 GHz. The room was A/C'd to 21 °C.

I, therefore, conclude that even recent x86 mobile chips absolutely require active cooling to deliver high performance in anything else than very short bursts.

I also discovered that the ACPI power profile would switch itself back to "low-power" after a while of torture testing. I wonder if this is a glitch or if it is the EC's response to the machine getting too hot.

[mdedetrich]

Same people who complain when the passive cooled products loses and asks to be benchmarked against the active cooled product. You didn't want a fan, you get fanless results.

Thats not how benchmarking works, you need to compare the same things (as much as possible) otherwise simply put the benchmark is invalid.

If you wanted to make a valid benchmark then either you should have disabled the fans for the x86/64 laptops (which you can, the CPU will just self throttle to protect itself from overheating) OR benchmark the M1 Pro (which has fans) and if you want to be even more accurate adjust the fan profiles so they are similar and then post the results.

What the people want is completely irrelevant here wrt valid benchmarks, and also to be frank if you care about acoustics and also longevity of laptop going fanless is the best as long as your chips are efficient enough so your performance doesn't tank (which surprise surprise the M1/M2 chips is the only consumer grade sillicon in this category now).

How do you determine that all other CPUs require active cooling?
Considering the 6850 U runs with 20W there is no reason why it couldn't be cooled passive like the M2. Its just the more costly variant and most users are already happy if the fan is really quiet.

Well you can watch the sensors for your fan to see how fast its spinning, or you can just temporarily disable the fans and see how fast your CPU will get throttled which will end up massively tanking your performance

[Anux]

I, therefore, conclude that even recent x86 mobile chips absolutely require active cooling to deliver high performance in anything else than very short bursts.

That's a really bald conclusion. One old x86 chip in one platform with one specific cooling solution is enough for you to judge million other devices?

Here you can see that the M2 with prime 95 starts at 31 W and throttles down to 25 W. No clock speeds are known but you can guess its around 1.5 to 2 GHz, but we know that those chips run at over 90 °C. https://www.notebookcheck.com/Test-A...3.0.html#toc-7 why would a Ryzen 6850 not be able to work under the same conditions?

If you wanted to make a valid benchmark then either you should have disabled the fans for the x86/64 laptops (which you can, the CPU will just self throttle to protect itself from overheating
...
or you can just temporarily disable the fans and see how fast your CPU will get throttled which will end up massively tanking your performance

No shit, if I disable the fans on an active cooling solution it's worse than a passiv cooling solution. Not sure what argument you try to make here, a fair benchmark looks different. Let me quote yourself:

Thats not how benchmarking works, you need to compare the same things (as much as possible) otherwise simply put the benchmark is invalid.

[mdedetrich]

To be honest, efficiency is the last reason why Apple have chosen ARM. Priorities were like:
a) Everything else in Apple ecosystem already was ARM based, so it was a natural and organic decision to unify architecture across the board so optimize the effort supporting two entirely different ISAs; That's an umbrella under which are like 80% of the reasons they went ARM;
b) Practically there were no worthy alternatives to ARM for them. x86 has licensing problem, RISC-V is still in it's infancy, POWER - perhaps, but ecosystem is small, as the market trends, MIPS is dead.
c) Somewhat better ARM efficiency and scaling. I'd say that's about it.

Actually completely wrong, especially if you read the leaked statements directly from Apple. Apple was majorly pissed off at Intel for better past decade because Intel completely botched their power saving and also their thermal efficiency. This is actually what pushed Apple to eventually create their own sillicon for their laptops.

Typical Apple customers don't want their laptops to sound like a jet turbine or require ridiculous power bricks to run and Apple tried to do this with Intel CPU's which completely failed to how inefficient the CPUs are.

[drakonas777]

Each core do "single threaded performance" in isolation.
That is why a 6 core Ryzen 3600x beats a 8 core Ryzen 1800x in every respects.

SINGLE THREADED PERFORMANCE IS THE MOST IMPORT METRIC whether you have 1 core or a 1000 ... (It is sometimes called IPC - Instructions per clock)

It therefore annoys me that bench-marking websites like Phoronix only glance over single threaded performance.
I would really appreciate it if they would have a section for single threaded performance in each article, so it is easier to compare that.

We are talking about different things. My argument is that usually at least several cores are utilized if you do any sort of somewhat serious work with PC. And if you don't, then you do not need PC at all - phone/tablet with ultra low power SoC is enough.

But if you do, then MT workload involves more just cores themselves - active or more stressed other IO/fabric blocks in SoC aswell (and platform in general) and that costs extra non-trivial amount of energy. So yes, of course you can look at single thread performance and make a conclusion "hah, if ST is great, MT will be great", but in reality it's not always the case. Especially talking about Intel CPUs and garbage power presets they have (OC/benchmarking mode is the default now basically ) where 1T will boost to the moon, but MT sustained frequency will be much lower.

It's a delicate dance between different SoC components and how they use the power, the software and uArch itself. My ultimate argument is and always was: ARM ISA is overrated as fuck. The uArch design, SoC design, packaging design,platform design and software support/optimizations - all these as a combination is what matters.

[mdedetrich]

No shit, if I disable the fans on an active cooling solution it's worse than a passiv cooling solution. Not sure what argument you try to make here, a fair benchmark looks different. Let me quote yourself:

What is your point then? The modern Ryzen/Intel CPU's are not as efficient as the M1/M2 and if you want to do a proper comparison with the M1/M2 then pick an Apple laptop that has active cooling such as the M1 Pro/Max.

And to also drive the point home, I have an active cooled M1 pro and the fans have only turned on once and that is when I was stressing CPU for over an hour, and this thing is competing against a desktop 5950x in terms of performance.