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

[mdedetrich]

What I am arguing is it is a mistake to attribute most of the M1/2 properties/user experience to ARM ISA. That's all. And unfortunately, that's a trend in the tech community. They see how M1/2 performs and they jerk off agains ARM ISA like it's the only or even the main reason for that. It's not.

Sure there are also other reasons as well, i.e. SOC (system of chip) design, using LPDDR5 memory with lots of channels however to argue that ARM isn't a contributing/main factor is deceptive and misleading. ARM is the reason why Apple's silicon no longer has to account for instructions that is literally many decades old in design and the ARM ISA is also deliberately designed for maximum power efficiency, that was like ARM's forté ever since it existed. The only historical issue with ARM is that its performance was many factors worse than x86/64 and there was no real market for ARM desktop because of chicken-egg problem so it took a company like Apple to change that, i.e. deliberately design ARM silicon that has extremely good performance while still retaining its historical power/thermal efficiency.

[mdedetrich]

Which is a point that I think often gets overlooked. Not even one Macbook in a thousand will ever be used for "real" work that needs the peak performance of the M1, but getting an extra 6 hours of Word etc matters, and getting that in something with half the weight of a "real" laptop and none of the fan noise is hugely desirable.

Is compiling programs from source which fully saturates the CPU considered "real work" for you? Because thats what I use my M1 pro for and its competitive against top end desktop CPU's in that regard.

I had a T14s Gen 1 before the M1 pro and compiling programs is 3-10x faster depending on what is being compiled.

[drakonas777]

I am not saying that ARM ISA is not a contributing factor for M1/2 properties. I'm saying it's not the "practically single", most important contributing factor.

Listen guys, instead arguing I suggest to do an experiment in the future. Hear me out. Lets wait for the Intel N4 / TSMC N5/N4 x86 SoC to emerge, so that we would have comparable node. When, let's wait for the last CPU SKUs within these nodes, to have the latest and most advanced x86 core in it. I guess it will be ZEN4+/ZEN5 for AMD and Arrow/Lunar lake for Intel. OK, after this happens let's take one of those Intel/AMD APUs, which has die space/transistor count/etc (matter of discussion) used as close as possible to match with say, M2 right. After that, let's install as close Linux distros on them, as possible and let's disable all the accelerators if any. Ant then lets do some testing on pure general purpose cores, multi threaded workload at the same power limit. Ant then we will see how M2 ARM is destroying those x86, OK? LOL

[mdedetrich]

I am not saying that ARM ISA is not a contributing factor for M1/2 properties. I'm saying it's not the "practically single", most important contributing factor.

Right, but that is an assumption and from what all of the evidence we have if we account for all factors it definitely is an important factor. Its definitely not node processing, since we have AMD running on the same node that Apple does. Really the ISA is the main differentiating factor right now.

Listen guys, instead arguing I suggest to do an experiment in the future. Hear me out. Lets wait for the Intel N4 / TSMC N5/N4 x86 SoC to emerge, so that we would have comparable node.

How is a node shrink going to change anything? If a node shrink happens it will apply to all ISA's, not just x86

When, let's wait for the last CPU SKUs within these nodes, to have the latest and most advanced x86 core in it.

This already exists with TSMC 5nm. We have M1 which uses TSMC 5nnm and AMD'z Zen3+ which is TSMC 5nm which laptops are starting to use. Why do we have to wait for 5nm (i.e. the renamed N5), are you arguing there is going to be some magic with N5 that will benefit x86/64 more compared to ARM in some future theoretical product?

I guess it will be ZEN4+/ZEN5 for AMD and Arrow/Lunar lake for Intel. OK, after this happens let's take one of those Intel/AMD APUs, which has die space/transistor count/etc (matter of discussion) used as close as possible to match with say, M2 right.

But why, this isn't going to change anything? I mean unless you are expecting Intel's node to leapfrog TSMC's (which is not likely for the near future) its going to be the same story. And if this does magically happen, its because of Intel's better node technology and not the ISA.

In fact you can argue that historically the reason why x86 was "better" is not due to the ISA but the fact that Intel just bruteforced performance with far superior node manufacturing compared to competitors (this is also evidence by the fact that AMD for most of its history had terrible performance with x86 due to its deficiencies with node technology). This worked up until a decade ago when TSMC overtook them.

After that, let's install as close Linux distros on them, as possible and let's disable all the accelerators if any. Ant then lets do some testing on pure general purpose cores, multi threaded workload at the same power limit. Ant then we will see how ARM is destroying x86, OK? LOL

You can do that right now with Asahi Linux. You can install it on an existing mac and at least if you don't care about GPU performance the CPU side of things is what you would expect from the M1 (in fact in some cases programs running on Asahi Linux M1 are faster than MacOS on the M1). So if anything this is widening the gap, not making it smaller as you are hoping.

[ddriver]

Exactly. Air M1 was too amazing (Apple wanted to ensure people will like the switch to ARM64), so Apple had to cripple Air M2 to make consistent steps in the next models line. M2 consumes a bit more power (like newer Intel/AMD/NVidia chips), but Apple also decided for weaker/cheaper passive cooling (and there's no active cooling naturally in the Air models).

The storage problem is well known and affects only the cheapest model. Such capacity is not usable for people, who need to do a real work, so they take the 512GB+ models anyway.

LOL, fanboi in denial much?

There was nothing amazing really, just a lot of low hanging fruit. It was the first 5nm cpu, with close to double the transistor count of products it was compared against.

This test only goes to further add substance to my claim that it is absolutely not a "wow cpu" more of a decent soc that leverages the vertical integration and full control over software and hardware.

Further M versions will be far more incremental, M2 is the same process, it is just various optimizations, and in some case even regressions, there is nothing to get that extra perf out of.

Quite amazing that amd manages to beat that on what's essentially 7 nm process. A solid general purpose performer. None of that hit or miss hybrid soc approach that apple is taking, accelerating some stuff while general purpose code struggles.

[MadCatX]

I am not saying that ARM ISA is not a contributing factor for M1/2 properties. I'm saying it's not the "practically single", most important contributing factor.

When it comes to absolute performance and power consumption under load, the M1/2 chip design is the major factor.

Listen guys, instead arguing I suggest to do an experiment in the future. Hear me out. Lets wait for the Intel N4 / TSMC N5/N4 x86 SoC to emerge, so that we would have comparable node. When, let's wait for the last CPU SKUs within these nodes, to have the latest and most advanced x86 core in it. I guess it will be ZEN4+/ZEN5 for AMD and Arrow/Lunar lake for Intel. OK, after this happens let's take one of those Intel/AMD APUs, which has die space/transistor count/etc (matter of discussion) used as close as possible to match with say, M2 right. After that, let's install as close Linux distros on them, as possible and let's disable all the accelerators if any. Ant then lets do some testing on pure general purpose cores, multi threaded workload at the same power limit. Ant then we will see how ARM is destroying x86, OK? LOL

[/QUOTE]
By that time, Apple will be out with 3nm M3, if not even further since Arrow Lake is scheduled for 2024.

[Ladis]

LOL, fanboi in denial much?

There was nothing amazing really, just a lot of low hanging fruit. It was the first 5nm cpu, with close to double the transistor count of products it was compared against.

This test only goes to further add substance to my claim that it is absolutely not a "wow cpu" more of a decent soc that leverages the vertical integration and full control over software and hardware.

Further M versions will be far more incremental, M2 is the same process, it is just various optimizations, and in some case even regressions, there is nothing to get that extra perf out of.

Quite amazing that amd manages to beat that on what's essentially 7 nm process. A solid general purpose performer. None of that hit or miss hybrid soc approach that apple is taking, accelerating some stuff while general purpose code struggles.

I don't care if the vendor gives me double amount of tranzistors for the same money. It's their problem, whether they're profitable. Apple is known to use this path - more tranzistor running at a lower frequency. Gives the same performance, but at a lower power consuption. Since Apple doesn't sell CPUs to others, he doesn't need to care about profitability of the CPU itself - they need to care only about profitability of the final product (device).

[drakonas777]

I will do it at the time when node, core sizes/transistor count used and SoC designs will be closer than they are now

Also, which M1 is on TSMC N7? They are all on N5 as I see it. If that's some "leak/rumor" - don't even bother do argument on it. I reject it. AMD does not use N5 yet. N6 yes, but it's a version of N7 basically. Some misunderstandings here I guess...

Anyway, my offer still stands: Apple M2 vs ZEN4+/5 or Arrow/Lunar Lake at 10-20W fixed. Software (en)decoding/3d render/compilation/HPC average. Let's see how it won't matter, let's see ARM's supremacy LOL Or even better - upcoming comparable Nuvia/Qualcomm/NVIDIA consumer ARMs LOL

[Anux]

What is your point then?

That it's perfectly fine to compare passive vs active chips. Just keep in mind that one is silent and the other not.

What about comparing cars, is it ok to compare a Porsche 993 vs McLaren F1?

The modern Ryzen/Intel CPU's are not as efficient as the M1/M2

And where is your proof for that? Have you read this article and the ones preceding it? Because the 6850U is faster than the M2 at roughly the same power consumption.

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.

Again why is a x86 CPU that consumes 17 W at average and 28 W peak (CPU only) (https://www.phoronix.com/review/ryzen7-6850u-acpi) not passiv coolable but the M2 with 25 W and 50 W peak (at wall) is?
And just because one is cooled with another cooling solution you're not allowed to compare them? Doesn't make sense to me.

Is it ok to compare 2 active Laptops where one has a stronger fan? And if so why?

I'd argue that a T14 Gen1 is a pretty good representative of what a decent x86 laptop can do temperature wise.

I'd argue otherwise.

The improvement between Zen2 and Zen3 is not that large.

Yes it is, look at the benchmarks here on phoronix.

There certainly aren't millions of devices, more like hundreds, and probably just a handful of them would have a considerably better cooling design than a TP.

Yeah millions was an exaggeration, but TP and best cooling design? Press X to Doubt xxxxxxxxxxxx

the M-based Airs do not have particularly sophisticated heatsinks. Some kind of over-engineered cooling solution might be able cope the latest x86 mobile chips but the point is that the M Airs don't seem to need one.

Why would an x86 chip with the same power consumption need a better heatsink? The Heatplate in the M2 is actually pretty simple https://www.youtube.com/watch?v=BSkcMwzZEFo but it covers 1/3 of the Notebook and you can't place heat sensitive parts near it, so you need to design the motherboard and battery accordingly.

The test you link seems a bit inconsistent in the comparisons it makes

Ignore them, I just needed the power figures under prime95 for that argument, all other data is in the tests here at phoronix.

[Weasel]

Look how many "CPU experts" dreams were shattered with actual facts. Imagine getting this destroyed still against products on inferior process nodes.

x86 >>>>>> ARM.