macOS 10.13 vs. Windows 10 vs. Clear/Fedora/openSUSE/Ubuntu Linux Benchmarks

[Guest]

Is it just me or has Mac been doing a lot better in benchmarks? I thought from previous benchmarks Mac lost in almost every test, but maybe I'm remembering wrong.

No, it's just lucky set of benchmarks.

[Guest]

Shhhh. Don't ever say that out loud that anymore. Last time I did, I got a shitload of people all over me giving me lots of reasons why Ubuntu has an optimal configuration already and why it shouldn't change a thing. So be careful what you say...

It's easy to prove them wrong. Once I pointed out what's sub-optimal in Ubuntu's kernel config. Now I'm using a custom one. Ubuntu kernel has some debugging enabled, uses 250 Hz timer and afaik it's not using CFQ disk scheduler.

[F.Ultra]

Good question, slightly slower on BLAKE, much slower on Go Garbage/JSON, poor results on GraphicsMagick, 7-Zip, Stockfish, crazy slow on C-Ray etc, it doesn't make sense, how much have Canonical botched this release ?

Many of these should also be completely CPU bound (e.g 7-Zip in particular) so there OS shouldn't play any major part, which makes these results even more strange.

[Guest]

Many of these should also be completely CPU bound (e.g 7-Zip in particular) so there OS shouldn't play any major part, which makes these results even more strange.

Different governors, perhaps?

[Charlie68]

What's been nice to see lately is how well Tumbleweed keeps up in these showdowns despite Michael installing it with Btrfs and XFS instead of the faster EXT4 file system that the others are using.

However, they are subtle differences to the user, so I consider the benchmarks just fun. For this reason I think the great convenience of using Btrfs with snapshots etc. is more important than what millisecond in speed.

[cynical]

Why is Ubuntu so slow? Userland? Kernel config? Or what?

It's using an older kernel, for one. So slow is a bit of an exaggeration though, there wasn't much difference overall between any of these.

[Royi]

This test doesn't test the OS.
It tests set of compilers and their configuration.

If one want to test the OS one should use the same compiler and compiling configuration on all systems.

It can be done with LLVM or Intel Compiler (GCC doesn't have full AVX support in Windows).
Otherwise those tests have no meaning.

[Grinch]

This test doesn't test the OS.
It tests set of compilers and their configuration.

Yes, in a very large part you are exactly right, with the exception of the Go tests since it's the same compiler toolchain across all systems, with the same optimization.

As you stated, in order to actually test the OS impact, you should use the same compiler and optimization settings across all OS'es, else the results are useless as a representation of actual OS performance.

[Royi]

Grinch , I really don't understand the point in those tests.
When Michael compares same system with different Kernel then it makes sense as he tests only one variable.
Here we would like to test the OS variable hence the compiler must be constant value.

The funny thing is it is not hard to do given what's already there in the benchmark framework.
Just download CLnag on all platform and run.

In this specific case we don't even know the compilation configuration in Windows which is ridiculous.
We know the HD IO in Windows is worse than Linux / macOS so those test make sense.
Everything else, makes no sense.

[carewolf]

Not sure if that is actually true. Last year, I switched to vanilla LLVM since Apple's fork had OpenMP disabled.

Isn't Apple's fork something ancient like 3.x?