Trying To Make Ubuntu 18.10 Run As Fast As Intel's Clear Linux

[AndrewDB]

Very good article and very interesting benchmarks, thank you very much Michael for all this excellent hard work.
I think what this extensive benchmarking experiment shows is that an entire Linux distribution can be optimized for performance to bring some extra speed for just about any sort of load. It just takes a lot of man*hours of expert Linux software engineers (in other words, you have to throw a lot of money at it, the kind of money Intel can afford).
The next important question of course is, do you really need that extra speed in most cases, versus the swiss-knife versatility, number and variety of packages, and community support for distributions such as Fedora, Ubuntu, Debian, etc? I know I don't want to switch from my current (xxxxxxxxxx) distribution to Clear Linux exactly for those reasons.
But still, Clear Linux remains an interesting distribution and a reference for all developers in the Linux community.

[EarthMind]

Very nice article Michael. Good work!

[Charlie68]

Great article but ... I tried to test Clear Linux on a fairly new notebook with intel processor and graphics, I am not able to give the exact hardware specifications, but what I could notice from the normal user and I did not notice any differences of performance, with Ubuntu or with other distributions.
I would not want all of this in substance to translate into something imperceptible to the user.

[s_j_newbury]

Great article but ... I tried to test Clear Linux on a fairly new notebook with intel processor and graphics, I am not able to give the exact hardware specifications, but what I could notice from the normal user and I did not notice any differences of performance, with Ubuntu or with other distributions.
I would not want all of this in substance to translate into something imperceptible to the user.

What is perceptible is battery usage. The higher performance means the CPU gets the work done with lower utilisation and/or gets the work done more quickly and can get into deeper sleep states.

[F.Ultra]

I think 2.28 they both use now, so that is not a reason of possible diff anymore:

https://packages.ubuntu.com/source/cosmic/glibc

I don't see so much big difference like before, as you can see in these benchmark results it mostly comes from clear docker now

So we can conclude that glibc version was indeed primary reason for big difference before

And of course that every single shared library is compiled with the ricer flags in Clear Linux while only the benchmark applications by themselves (and the kernel in once of the tests) are compiled likewise in these benchmarks. Should be some small percentages that adds up from that as well.

[arjan_intel]

any sort of load. It just takes a lot of man*hours of expert Linux software engineers (in other words, you have to throw a lot of money at it, the kind of money Intel can afford).
The next important question of course is, do you really need that extra speed in most cases, versus the swiss-knife versatility, number and variety of packages, and community support for distributions such as Fedora, Ubuntu, Debian, etc? I know I don't want to switch from my current (xxxxxxxxxx) distribution to Clear Linux exactly for those reasons.
But still, Clear Linux remains an interesting distribution and a reference for all developers in the Linux community.

we're growing Clear Linux to be a lot more versatile, but going from 3000-ish to 15000-ish packages while retaining quality does take some time...

[deant]

i think clear linux demonstrates that linux can be the fastest OS in our time. This just needs to get mainstream into all distros.

[geearf]

we're growing Clear Linux to be a lot more versatile, but going from 3000-ish to 15000-ish packages while retaining quality does take some time...

It'd be nice if you could push some of the cool work you do onto other distributions too, that way everybody wins

[s_j_newbury]

There is an organised effort to learn from Clear Linux for Gentoo here: https://github.com/InBetweenNames/gentooLTO

Of course Gentoo was ahead of game here, no doubt Clear Linux had a lot of groundwork prepared by Gentoo "Ricers"... err I mean users! ;-)

[CochainComplex]

I think what this extensive benchmarking experiment shows is that an entire Linux distribution can be optimized for performance to bring some extra speed for just about any sort of load. It just takes a lot of man*hours of expert Linux software engineers (in other words, you have to throw a lot of money at it, the kind of money Intel can afford).
The next important question of course is, do you really need that extra speed in most cases, versus the swiss-knife versatility, number and variety of packages, and community support for distributions such as Fedora, Ubuntu, Debian, etc?

In my opinion it is really worth the effort. Since small gains usually come with more power consumption or buying the next higher ranked product.
In the last decade instruction sets of x86_64 have developed far beyond the std x86 repertoir. Especially vectorization should be highlighted here.
Why not using it with fmv? A suitable approach would be setting default, avx, avx2 for loops like it is shown here https://clearlinux.org/documentation.../tutorials/fmv

People are moaning that i7xxxx is 5% faster in case x as ryzen yyyy or vice versa. For the future it is foreseeable that the majority of people will use SSD with a lot of capacity in their rig. That said, RAM/SSD Capacity/Speed is growing faster than CPU Computing Power without additional use of specialized instruction sets.
Therefore it is necessary to make use of optimization flags even with branching of multiple versions in one build.
If more people are using this features it is very likely that compiler devs will improve the usability of it.