Apple M4 Mac Mini With macOS vs. Intel / AMD With Ubuntu Linux Performance

[db23]

According to Apple, the M2 Ultra-powered Mac Studio can use up to 295 W.

https://support.apple.com/en-us/102027

The only things in the case are the SoC, SSD, Ethernet phy, and a fan. So, that's pretty much all SoC. Even if we take it to be their "at the wall" figure and assume a power supply efficiency of 87%, that means the SoC is still using up to ~250 W.

That is not right, just because the PSU can draw that much doesn't mean it is all going to the SOC. I'd be shocked if it uses over 90W considering the M3 version uses less than 80W. Keep in mind all the other ports including the 6 USB4/Thunderbolt4 ports which spec allows for 100W output to devices(not that it will deliver that much), this is what that excess power is for.

[cutterjohn]

There is Joules reported as footnote on the power graphs...

He just wants to be pedantic... I see that SAME post in EVERY review... (and many times, your same response, or by someone else...)

[botipua22]

Hot damn does Apple make some excellent chips and hardware. They do have monopoly over TSMC's best nodes, but beyond that they still seem to make faster and more efficient chips than their ARM and x86 competitors. Now only if their stuff could run proper Linux instead of macOS and iOS!

[xenospace]

well, this ruined my day...

[sophisticles]

- Proper package management.
- Access to Mesa drivers and Proton compatibility for gaming
- An operating system that can be trusted

There is no such thing as a Linux package manager, Linux by default does not have a package manager, and neither does any other OS for that matter.

No OS needs to have a programmed "installed" in order to use it.

What happened decades ago when GUIs started to be used, a method was devised so a downloaded program could be accessed by clicking on an icon or an entry in a menu. The method basically amounts to the executable being copied to a specific folder and another file placed in a different folder.

All package managers work the same way, doesn't matter if it's any of the Linux ones, such as dnf, apt, portage, any of the "stores" like Gnome's software center, doesn't really.

There is nothing unique about the way Linux based OSes do things, just the name changes.

In fact, you can "install" any program you want, to any folder you want and have it listed in the menu of any Linux GUI you want to use. All you have to do is create a text file named with whatever the program is called dot desktop and put that text file in /usr/share/applications, in that file you will have these two lines:

Exec=/opt/programfolder/program # path to the program
Icon=/opt/programfolder/icon # path to the custom icon

That's it, there's nothing "proper" about it at all, if anything it's very insecure because you can easily screw up a system by just deleting all the files in /usr/share/applications if your package manager screws up.

Mesa drivers are a joke, it's been shown time and again that proprietary drivers offer superior performance and why on Earth would anyone want to run any game using Proton instead of natively?

As for Linux being trustworthy, maybe if you have the IQ of a Kumquat, then yes, it is trustworthy.

But if you really understand the OS you claim to love?

Not a chance.

[coder]

This might be another reason for the outstanding single core performance. A single P-core has access to all the L2 for the whole cluster.

Well, if we'd have more single-threaded benchmarks, then maybe we could test that hypothesis. The other way would be to look at the performance counters when running the FLAC benchmark to see just how often L2 misses occur on one of the other CPUs.

On the former point, I can point to the Geekbench 6 single-threaded numbers. Although we can't rule out that it's not also unduly affected by L2, it's meant to be a composite score and thus shouldn't be too weighted towards cache size, latency, bandwidth, etc.

https://www.notebookcheck.net/Apple-M4-SoC-analysis-AMD-Intel-and-Qualcomm-currently-don-t-stand-a-chance.839332.0.html

Here, you can find the M3 owning Cinebench 2024 single-threaded performance, by a substantial margin. Presumably, the M4 will inherit this crown:

https://www.notebookcheck.net/Intel-Lunar-Lake-CPU-analysis-The-Core-Ultra-7-258V-s-multi-core-performance-is-disappointing-but-its-everyday-efficiency-is-good.893405.0.html

I'd suggest Michael consider creating a single-threaded suite in PTS and then presenting the composite score from it, in CPU reviews.

Likewise, there should be a multi-threaded suite, in which all of the tests are capable of exhibiting very good scaling on at least one CPU. Or, take the same same approach as in SPEC2017 and create a Rate-N composite that's measured by running N instances of each test in the single-threaded benchmark suite.

Michael , perhaps what's limiting the notoriety and growth of PTS is the lack of well-defined composite scores, like SPECint, SPECfp and the Rate-1 vs Rate-N variants of each. One beef I have with SPEC2017 is that Rate-N is shared-nothing. While that's useful, it would've been nice if they also had a suite of highly-scalable benchmarks that would each be capable of spinning up N threads.

P.S. where I think GeekBench6 went wrong is they made their Multi-Threaded score consist partly of apps that specifically do not scale well, which I find a bit ridiculous.

[Espionage724]

- Proper package management.
- Access to Mesa drivers and Proton compatibility for gaming
- An operating system that can be trusted

I'm my own package manager: Did it since Win95 no problem. Windows gives me the freedom to vet my own stuff; Linux is get it from official repos or you're on your own. macOS I had no issue with App Store and some dmgs/pkgs for other stuff. Linux is more walled-garden in this perspective than Windows and macOS.

Mesa: Can't say I need that over a proper graphics stack; Windows also presents that fine, and so did macOS for the stuff I played

Eh, I trust macOS and Windows. I also trust mainstream Linux distros. MS and Apple have duties to provide a good OS under payment; Linux is community-driven, which apparently has Wayland being some star-performer for years which was an outright lie 2016-2024. No Xorg vs Wayland silliness on Windows and macOS: They ship something that just-works well.

[Espionage724]

BTW, why not CachyOS?

Because Intel Clear Linux makes more sense on an Intel CPU, and realistically nobody is using some non-mainstream OS on specialized hardware

I'm not for the appeal of CachyOS when people can just make the base OS Arch do whatever its doing, and I'm sure Gentoo offers the same possibilities; CachyOS just sounds like a fork easy-distro for gaming performance, and if it isn't, is anyone daring to be running that on a Server?

[coder]

I'm my own package manager: Did it since Win95 no problem. Windows gives me the freedom to vet my own stuff; Linux is get it from official repos or you're on your own.

There's a fundamental difference you're glossing over. In the modern Windows ecosystem, most programs ship with all of their dependencies bundled, with a few exceptions that they get out of system32 or whatever it's called nowadays.

In Linux, the default is that everything is shared between all of the packages, so force-upgrading one thing can break dependency chains for other things. That's why packaging is so crucial for Linux.

One benefit of the Linux approach is that when a security vulnerability is discovered in one library, you just update the shared instance on your system and it fixes every package that uses it. In Windows, you have to hope & pray that each user of a vulnerable library releases an update, and it's mostly on you to seek out and install those updates.

Linux puts a foot back into this world, with things like Snap.

[dcdev]

Because Intel Clear Linux makes more sense on an Intel CPU, and realistically nobody is using some non-mainstream OS on specialized hardware

I'm not for the appeal of CachyOS when people can just make the base OS Arch do whatever its doing, and I'm sure Gentoo offers the same possibilities; CachyOS just sounds like a fork easy-distro for gaming performance, and if it isn't, is anyone daring to be running that on a Server?

You didn't understand me.
In this test you can see Ubuntu. Not CachyOS or Clear Linux.
Ubuntu is not the fastest distro.
And there are AMD CPUs, too. I'm not sure if Clear Linux works on AMD.