Clear Linux Beats MacOS in MacBook Pro Benchmark Tests (phoronix.com)

To celebrate its 14th birthday, Phoronix.com used a 15-inch MacBook Pro to run system benchmarking tests on the following operating systems:

- Windows 10 Pro

- The latest macOS 10.13 High Sierra

- Windows 10 Windows Subsystem for Linux (WSL) using Ubuntu 18.04

- Ubuntu 18.04 LTS with the Linux 4.15 kernel, GCC 7.3.0, and an EXT4 file-system.

- Clear Linux 22780 with the Linux 4.16 kernel, GCC 8.1.1, and EXT4.

- Fedora Workstation 28 with updates is the Linux 4.16 kernel, GCC 8.1.1, and EXT4.

- OpenSUSE Tumbleweed with the Linux 4.16 kernel, GCC 7.3.1, and default file-system configuration of Btrfs root file-system with XFS home partition.

The results? When it came to outright wins and losses, Clear Linux 22780 was the front-runner 59% of the time followed by macOS 10.13.4 finishing first 21% of the time and then Fedora Workstation 28 with winning 10% of the time.

For losses, to little surprise considering the I/O overhead, Windows 10 was in last place 38% of the time followed by Ubuntu 18.04 being surprisingly the slowest Linux distribution 30% of the time on this 2016 MacBook Pro.
The article also reminds readers that "For those looking for a Linux laptop, there are plenty of better options..."

Comments

Battery life?

[Anonymous Coward]

Where is the benchmark for battery life? One of the strengths of the MacBook Pro and macOS is the power management and long batter life. I would be surprised if Linux was as good.

Re:Battery life?

[dgatwood]

What the GP meant is that Apple has always been better at Intel power management. Chromebooks and Android are usually not x86, and when they are, they're typically much slower, lower-power x86 hardware than what ships in a Mac laptop. So of course they have long battery life. That goes without saying.

To use a car analogy, comparing a Chromebook to a Mac laptop is like comparing an electric golf cart to a Tesla and saying that the golf cart is as good as the Tesla at battery management because it can run for as many hours on a charge as the Tesla, ignoring that the golf cart went twenty-five miles in those five hours, and the Tesla went 300.

Let's not get silly, shall we?

[Qbertino]

So a super-lightweight quasi real-time IoT Linux OS beats macOS on it's native hardware 60% of the time? Give me an effing break, will ya?
I'm no Apple fanboy and there's plenty of stuff going on with Apple right now to piss on, but performance and integration of their high-end all-out desktop OS into their purpose built hardware is still next to none, by a far margin.

Trying hard to find something that 'beats' them at that game makes you look like an idiot.
So let's not be silly.
Please.

Re:Let's not get silly, shall we?

[dfghjk]

"So a super-lightweight quasi real-time IoT Linux OS beats macOS on it's native hardware 60% of the time? Give me an effing break, will ya?"

As if those features explain the test results. Give me an effing break, will ya?

Re: Let's not get silly, shall we?

[Qbertino]

Not a fanboy at all. No, not at all.

My last Mac is a 13" MBAir from 2010. I'm about to sell it.

My main OS is x86 Linux (Xubuntu on my large ThinkPad and Manjaro i3 on my small one).

And while I really like Apple and appreciate some of the very neat stuff they do and I also like the fact that there is some very neat software available on mac, I am not a fanboy. Right now I find them way too expensive and apparently their new keyboards - allthough I really liked typing on them at the Apple store - need a redo to not fail because of dust. Not getting one of those. To risky at that pricerange.

When their current pricerange drops 400 euros across the board, then I might consider them again on my next hardware buying round. Which, as of now, is at least 4 years out.

Re: Let's not get silly, shall we?

[mccalli]

top hasn't been good way to measure RAM for -years-. Not on Linux either. If your "sitting here on HS with a couple of windows using 13Gb" was correct, then none of the 8Gb machines would even boot. Measuring RAM usage these days, on any OS, has been trickier for quite a while. Buffering, wired etc.. I remember when one of the versions came out, think Lion but might be wrong, and it reported itself as using all the RAM people went mad. Windows shortly changed to do something similar as well, if I recall, and Linux has definitely been doing this for quite some time. Even after all the measuring is done you have to ask another question - what did you buy the RAM for? To sit there being pristine and clean and free? No, to be used. Your oS has detected the resources currently available to it and is making good use of them. If it had less, it would behave differently.

Celebration?

[Rewind]

That does not sound like a very fun birthday party...

The Year of the Linux Laptop!

[Salo2112]

Finally, it's here!

Windows defender hurts

[LostMyBeaver]

Windows Defender is tuned for Windows kernel integration and performs well there. Although to be honest, recent builds of defender have been a hog as I'm assuming it's being a bit more aggressive at sandboxing for CPU prediction bugs.

That said, Windows defender doesn't seem to understand the WSL stuff at all, however it's using the entire system resources to real-time monitor disk reads and writes.

When running without Windows Defender real-time monitoring enabled, it seems to increase performance of the VM to near bare-metal speeds.

Re:Windows defender hurts

[LostMyBeaver]

Well, to be accurate, it's subsystem. Win64, Win32, Win16, OS/2-16, POSIX, and Linux all run as subsystems on top of the Windows Kernel.

I was a bit tired when I wrote that and now that I'm reading it, I'm struggling to understand why I wrote it that way.

WSL is absolutely amazing in the sense that it's basically a clean-room "kinda user space" Linux Kernel. Microsoft, re-implemented the Linux Syscall interface (not that amazing, Sun, FreeBSD, etc... have all done this) and re-implemented components of devfs and procfs as well as lots of other stuff as well. It's a major undertaking since Linux is a bit of an ABI whore.

So, the weird thing is, it is kinda sorta a VM, more of a container with something more VM like than bare metal happening.

It actually is not another kernel as opposed to a subsystem which implements the entire ABI of a kernel. In fact, what's really funny is that it's not technically Linux, rather Ubuntu running on a full Linux kernel replacement.

Now, I'm really really glad you made me read further because the "VM" as I called it and the "not-bare metal" abstraction is the performance issue and it's known.

WSL is running on what's called Pico Processes is which another system for containerization which is yet another layer up. So, the definition of what bare metal is gets really confusing. Also the difference of what a virtual machine is get really blurry.

VMware in the earliest days (and even now for legacy support) runs a just-in-time compiler against guest operating systems to intercept hardware calls that need to be virtualized. Unlike more modern and refined hypervisors, it makes no real attempt at eliminating the need for this silliness. Instead, it's actually scanning loaded machine code for things like calls to INB/OUTB/INW/OUTW... etc... so that it can intercept them, set TRAPs (triggering NMI and/or debugs), inject calls to simulated functions, etc...

Then there's things like para-virtualization where you recompile the operating system kernel to explicitly make calls into the host OS and completely and totally remove any actual kernel from the kernel but instead do something more like user mode Linux. This requires the vendor of the "hypervisor" to change the kernel source of the guest or to supply the APIs needed by the guest. Windows, Linux, FreeBSD, etc... all support these APIs and they're even documented to varying levels. For example, the entire interface as needed by Hyper-V is part of the Linux kernel. Those same APIs are used by Windows and therefore could be used to paravirtualize much of Windows.

Then there's the in-between thing. See, a proper hypervisor as opposed to parahypervisor tends to perform resource allocation and provide a lot of bare metal resources to VMs. So Intel, AMD, NVidia, Cisco, and many others would have you believe that a VM in order to be a VM would require making use of hardware partitioning services.

This is simply not the case. That's legacy virtualization and is no longer needed for pretty much anything other than running non-paravirtualized operating systems.

Real "bare metal hypervisors" are ridiculously inefficient and risk massive security issues because there's no way to do things like have the host hypervisor enforce security. The latest hack on AMD's encrypted memory for VMs made me realize how insanely broken the entire hypervisor model is. In fact, pure hypervisors like VMware should never be allowed in a secure environment. Paravirtualization is the only possible way to start fixing VM security... and that will be a long road.

So... then we have Pico-processes which is a beautiful solution from Microsoft for providing a much much much better solution to the problem... at least when it comes to secure virtualization. Simply don't virtualize. Make a VM which isn't a VM.

Remember, a virtual machine doesn't in fact have to be a perfect replica of a physical machine. That's just VMware voodoo and witchcraft (and I pray they keep making that because I love VMware

Gentoo

[Bert64]

As i understand it, clear linux is a distribution optimized for modern hardware, with all packages compiled with newer compilers and a lot of legacy cruft disabled etc...
So it would be interesting to see how it compares to gentoo, which is also usually configured in that way.

It's also interesting how badly ubuntu fares in many of these benchmarks, despite being only a small step behind clear linux in terms of kernel/gcc versions in use.

Re: Gentoo

[prefec2]

These are all Phoronix measurements. Unfortunately, they do not discuss how they measured it, why they think there are performance differenecs between distributions. Differences could come from any number of sources. Also, I could not find any details regarding test setup, warmup phases, repeats etc.

Things which are greatly suspicious are for instance the Windows WSL performance compared to Windows native. As WSL is more or less an API mapper, it should be slower than native Windows. Except you cache diskspace in memory which would make IO results not comparable.

Re:Nobody cares...

[Ol Olsoc]

So that is why all the posts so far have been from butthurt appleboys... Fail to see your logic there.

Cool story bro! That's odd, because that isn't what I'm seeing.

Using both Linux and MacOS, my reaction is mainly to roll my eyes at the benchmarking fetishists.

p.s. I use W10 as well, but try not to mention that in civilized company

Re:Nobody cares...

[SurenEnfiajyan]

I'm typing this comment on Win7. Actually I've never used any Apple device or OS. But I've tried numerous Linux distros (as a desktop OS) and all of them were unpolished and buggy messes. Most sane people don't care if a Linux distro (especially a stripped down one) shows slightly better performance when it's usability is definitely nowhere near Mac OS's usability. Also I'm neither happy about the recent MSFT's and Apple's moves.

It's a Redhat / CentOS respin

[raymorris]

You're overstating a bit. ClearOS is based on CentOs, and installed on large servers by HP. Not exactly IOT.

Try that with a 2018-2019 Mac (if and when...)

[williamyf]

When a new mac is spawn out of cupertino and unto the world, the linux crowd has not had enough time to adapt drivers and stuff to the system. Therefore the benchmarks always go the way of MacOS. Two years down the road, one needs to use a barebones IoT linux distro to get better performance than MacOS itself, because propper linux desktop distros do not cut mustard.

Do not get me wrong, I like linux a lot, I was a linux evangelist in the early 2000's, my beef is not with linux, is with the moron who wrote the article, and with editordave for summiting it.

Slow news day...

clear linux for the desktop?

[Espectr0]

before this article i have never heard of clear linux before.

the question is, can it be used as a regular distribution? from its webpage i didn't get the impression it could be used as a regular desktop distro

PHP benchmarks

[OrangeTide]

Because the first thing I think when I get a new laptop is how fast can it run PHP.

Big picture includes various factors

[shanen]

Yes, battery drain is a dimension worth considering, but the main reason this story is absurd is because it ignores the overheads such as installing and updating the OSes. Your computer might save a little time on computations, but how much of your MUCH more valuable human time was consumed along the way? How much of your time was saved if Apple tested the OS more carefully? Looming over all of this in the real world is the recovery question. How much time could you lose trying to recover when something goes wrong? And you know that something is always going to go wrong.

I am NOT an Apple fanboi. Most of my machines run Windows, though I have 2-1/3 Linux boxen and one Macbook Pro. I actually regard Apple as a dangerous corporate cancer (but I'll drop that tangent for now). However I have to report my latest experience with Apple was MUCH improved. They fixed the hardware much more quickly than I expected, without charge (and even gave me a bit of grace on the warranty period), and without damaging my software configuration. I spent much more time restoring my Android smartphone the last time one of them had to go to the shop. (Actually ASUS has that phone now, and I have NO intention of paying those bastards for any more repairs. (However it's really my own fault because I had dealt with ASUS once before and this is a case of shame on me.) I hope ASUS enjoys eating the phone.)

The real point of this story is "Penny wise, pound foolish." Didn't find anything along those lines, but at this point I'm not at all surprised to be disappointed with Slashdot. Maybe just a failure of the moderation to make visible some better comments that I couldn't find? Increasingly convinced that the moderation system has become the biggest problem killing Slashdot.

Wouldn't Run Linux on a MacBook

[Philotomy]

Linux is my main OS, but I see no compelling reason to run Linux as the native OS on a MacBook. If you want to run Linux on a laptop, get a laptop with better specs and run Linux on it. If you want a MacBook, just run MacOS.

The primary reason I own a Mac is for Xcode and iOS development. I'm hoping Apple updates the Mac Mini soon. If they do that, I'll get a Mac Mini for iOS development and replace my (aging) MacBook Pro with a Linux laptop. (I run Arch on my desktop/workstation, but I might give Ubuntu a try on a new laptop.)