108 comments

[ 2.9 ms ] story [ 168 ms ] thread
I'm really surprised that the OS makes so much difference to CPU bound tasks.
Not to discredit Clear's work at all, but it's not so much the 'OS' in standard parlance, it's still just another Linux distro. It's down to compiler flags and optimizations, which generally anyone can do given they have the time and patience.
I was just thinking about this this morning. I feel like any time I update the OS on a device it gets slower, and was wondering if there was some OS focused on being fast (and hopefully battery-efficient). Maybe I'll have to try throwing Clear Linux on my old 2011 notebook and seeing how it does!
What processor does it have?

I have Clear Linux running on a Haswell Core i7 and it's great, but it doesn't run on a Core2Duo I've got laying around.

Also, it needs EFI booting - not sure yours is new enough for that.

Interesting, good to know re the Core2Duo. I have an ancient Thinkpad T61p that I was going to try this out on as speed differences on hardware that old become obvious. But I guess I'll stick with Fedora.
That's what I was also wondering. I've got an old Lenovo SL500 with a C2D/3GB RAM sitting behind my TV for streaming/web browsing. Currently got Mint/MATE running on it (I know it's not the fastest thing around, it was just a quick setup for something basic and not too visually "offputting" for others in the house).

The hardware is old, sure, but I don't need it badly enough to justify building/buying something else and I had it laying around from the mid-2000's when I was testing WiMAX rollout in my city. Figured it would get a bit more use instead of living on the "old hardware" shelf (shelves) in the basement.

> I feel like any time I update the OS on a device it gets slower

Never had this problem on Debian Linux, FWIW - even when upgrading to a new release. Of course sometimes stuff gets marginally slower due to new security mitigations or whatever, but not in a way that's even perceptible, let alone catastrophic.

Never felt this on Fedora either, to the contrary actually.
For battery life, don't bother with linux.
Things have actually improved a lot. I'm running Ubuntu right now and drawing ~5W from my 51Wh battery on a new Ice Lake Dell XPS 13. Highly recommended.
What makes Clear Linux so much faster then other distributions?
There's a LWN article on it: https://lwn.net/Articles/644675/
That article describes why containers start up more quickly on Clear Linux.

I believe that the main reason applications run faster is that it determines which exact CPU you have and compiles everything with the flags that enable every optimization that your CPU can handle.

It has all Intel specific hardware acceleration / performance settings on by default. So if you're on Intel it's perfect. Sort of wish AMD had their own Distro to see which one really is the best on their respective processor architectures.

Edit:

Look here:

https://clearlinux.org/

Under 'features' it says the following:

> Intel Optimized > Highly tuned for Intel platforms where all optimization is turned on by default.

Note that the tests in the linked article were done on an AMD CPU.
Essentially I don't see why AMD wouldn't support some of the optimizations, but I could see why it wouldn't support ALL of them. That is noteworthy though, I missed that bit of the article.
Compiler flags you can just add yourself to your makepkg.conf on Arch (and BSDs with their ports trees).
My understanding was that this is also why Gentoo is popular, since you are compiling it all anyway, you may as well compile it all for your specific processor.
Ah, but Gentoo also supports binary releases. Useflags are obnoxious to set up.
I may be wrong here, but as far as I understand one of the major advantages is simply that they have many versions of certain important libraries in your system and are able to select the one that better matches your hardware. Most distros are super generic and none of the system libraries are making use of instructions that are not present in all of the x86-64 machines ever released.

It makes me really sad that other distros are not following Clear's lead here.

Well, Stampede Linux had this more than a decade ago and it wasn't a big enough differentiator.
Other (binary) distros are solving for a different problem: providing a good-enough generic base that runs on most systems.

If you want everything micro-optimized to your particular system's hardware configuration, you know where to find Gentoo.

Although Clear does it with prebuilt binary libraries. No need to compile your own.
I'm pretty sure Clear only has a single build target. For example, their RPM repository[1], only has a single arch (x86_64) and a single RPM for each package. I think it is mostly that they (1) invested effort in maxing out the optimization flags for all the libraries (2) have a higher minimum CPU requirements (must support SSE 4.2, which was first released ~10 years ago).

But yes, I agree it would be nice if other distros had two amd64 releases; one that targeted newer hardware, and another for lowest-common denominator.

[1] https://cdn.download.clearlinux.org/releases/32250/clear/x86...

They optimise some stuff but you'll generally see phoronix benchmarks score fedora workstation the same on a lot of workloads. Its package manager is horrendeous and there are no obvious upsides to just compiling yourself from the AUR automatically.
It would be interesting to see how it fares against Peppermint, which is known for working really well on low-end hardware.

Also, I'm curious if Clear Linux is also fast on hardware that is slow not because it is cheap but because it is old. Would it speed up my Core2Duo laptop?

It's more aout getting rid of what slows you down (bloatware/fat distros) than it is about linux (the kernel) itself.

Who else finds that animations/smooth scrolling just breaks the flow/gets in the way?

Smooth animation can be critically important when dragging, swiping or performing "fine" scrolling motions, either with a mouse/touchpad or on a touchscreen. Because then lags and freezes are what "gets in the way" and breaks the flow. Otherwise, yes, most animations just add latency for no good reason.
Interesting point to discuss, because I find animations etc slow things down and for me that breaks flow (if I'm using sharp tools, I want instant response).
I do. I disable animations on my Android phone and my desktop OS is animation-free.
Does bloat even apply nowadays to Linux? On arch you can compile and optimise your software well and ryzens have upwards of 256KB instruction cache.
> Who else finds that animations/smooth scrolling just breaks the flow/gets in the way?

It improves ux by hiding the latency behind visual transformations.

I am entirely the opposite. I find that a lack of animations breaks the flow. I also tend to prefer a "nicer" UX over pure speed and efficiency.
It's all context dependent really. It can go either way. For example:

- Smooth scrolling that's actually smooth and fast gives you immediate feedback on where the last line you were reading is. There's no guesswork in "how much did that page jump down exactly" anymore.

- Smooth scrolling which is laggy and bouncy doesn't remove the confusion and may add to it instead.

For animation example:

- A quick (even 3-frame) animation for minimising a window doesn't delay you, but gives you information: where is your window now on the taskbar, and the action indeed happened (if you have 2 similar documents opened and the window just disappears, it may not be clear)

- A slow, fancy animation that rolls up your window just introduces delay and distracts from the task, even if it provides the benefits above.

Finally, useless animation, especially connected to the accelerometer, can induce motion sickness, so sometimes it's worth making it less smooth/much faster. We can still register 2-3 frame blinks as something specific happening.

As usual, there's a lot of grey area in between having animation, or not.

This might sound stupid but does that also affect gaming performance, or simply Wine and/or Proton?
This is their second recent article about linux laptops, and I have seen no mention of battery life. While its nice to have a 15% faster OS, I would also like to know how the battery life differs between various distros (and Windows).
I should note that power and performance are not opposites. Good power management ensures you don't waste power, which helps you draw all you need when you need it by the component that needs it. Good power management often helps overall performance.

One thing that it often works against is latency, because it takes time to wake up things that are sleeping. But that is really not something you should care about if you're using a laptop. Your laptop is not servicing Google search results.

Performance is a solved problem nowadays, latency isn't. Especially with cloud-based software being pushed. The benefits of a tweaked linux kernel, such as liquorix are easy to spot. Yes, slightly reduced brute performance but large leaps in smoothness.
Yes, it often does in a broad sense. But there is a lot more to power management. Things like putting busses and devices into low power states when idle, and batching up I/O so that devices can spend most of their time in low-power states. Scheduler tweaks to idle most of the cores, etc.
My anecdotal evidence after searching for the "perfect" laptop OS across 2 laptops and various BSDs, a dozen Linux distros, etc.: the best power tuning you can do will still only get you around 85-90% of what you can in Windows (with no tweaks).
I think this is generally true until you discover Windows Defender or the Logitech Updater or whatever has lost its mind and been spinning on CPU for the past three hours and now your battery is dead. I have a surface with pretty great battery life, but it's also rather fickle and you can lose hours of runtime because it slept on a pea.
Anecdotal as well; on my X220 and GPD Pocket 1, I get over 40% more battery life under Linux than under Windows. Especially on the Pocket the difference is extreme. I do not use virus scanners (including win defender) on either platform and I check task manager/top etc regularly to optimise. On my Surface Go battery life on Linux is also better than under Win, not as extreme as on the other 2 devices, but significant.
I've been running native Linux on a $299 2015 Chromebook[0] for years and I get about 5 hours of casual programming, Youtube, browsing, etc..

In practice it works really nicely for what I use it for. A temporary portable computing device that isn't my main workstation but still gets a lot of use.

[0]: https://nickjanetakis.com/blog/transform-a-toshiba-chromeboo...

I get excellent 8-10 hr lifetime on a Lenovo X1 Carbon and a Dell Inspiron 7590 with Ubuntu.
Too me the X1 Carbon is a 16:10 aspect ratio screen away from being the perfect laptop.
Want to get a linux notebook for running Ubuntu 18 Lts for my projects on fpga. Wonder carbon is ok as I like their keyboard but keyboard only. Need thunderbolt for my AI part as well.
Is 4GB enough with the current web browsers and web pages? It feels like 8GB is the minimum for normal users, and 16GB is the minimum for software developers. I have a 4GB Chromebook and a 4GB MacBookAir, and they are both somewhat painful to use these days, even with ad blockers.
Which browser? I've had zero problems with 4GB laptops when avoiding Chrome
Which Chrome-avoidant browser?
Firefox, IE, Safari, or Edge (but not Edge Beta) to name a few. IIRC, there's Pale Moon or something else based off Firefox as well.
Can't avoid Chrome on a Chromebook. On the MBA, I use Chrome as well. I've migrated to Firefox on my Windows and Linux laptops, but I thought Firefox had battery issues on MacOS.

I usually keep about 5-7 pinned tabs open (Gmail, Contact, Calendar, Keep, Drive, HackerNews, etc) and cycle through ~3-7 transient tabs, and I find that the browsing experience is not enjoyable on 4GB machines anymore.

I don't have any distinct cpu/memory/battery profiling datasets to provide anything but a personal anecdote. I dropped Chrome due to general laptop responsiveness in favour of a combo of Firefox with AdNausem and Privacy Badger. Which, has worked fine for my use (not noticed any adverse battery effects but I've also not been actively monitoring), though I don't use nearly the number of google services you've listed. Your results may differ, I've heard mixed reviews (again, anecdotally, so take with a grain of salt) of google services in firefox and safari.
> but I thought Firefox had battery issues on MacOS.

Does it still, are you referring to 70+? Firefox 70 in October came with big improvements[1], I didn't do any testing though. I think I saw some benchmarks showing it's slightly worse than Chrome, but it was minor. Can't find them right now unfortunately.

[1]: https://arstechnica.com/gadgets/2019/10/firefox-70-brings-en...

> Can't avoid Chrome on a Chromebook

It is deeply psychologically jarring, but Firefox runs through the Android and I think Linux compat layers on a Chromebook just fine. I have no idea how much this actually buys you when the OS is by Google for running Chrome, but it does work.

4G or 8G to me the most impactful is cpu/gpu. hardware rendering lets the cpu cool off. Without it webpages will peg the cpu, heating or throttling. Very annoying no matter the ram.
Xiaomi Mi Book 12.5 also has 4 GB and it's really snappy and good enough for me with a tiling WM and Emacs. Sadly not sold outside China anymore.

Any other cheap machines worth considering? This was fantastic given build quality, fanless m3 Intel CPU and all Intel components.

Acer usually has something comparable.
I have chrome and firefox running, 20 tabs between the two, some heavy ones (google doc, gmail), iterm and PDFs with preview on an MBA 4GB, no issues, memory pressure in activity monitor is still green. One little nitpick, my dev enviroment is on a remote 4GB VM ;)

What applications do you run that make it so bad for you?

i run intelij idea, firefox, dbeaver, spotify, and a regular python stack and that's enough to consume the 8gb ram that i have on my matebook 13. and when all ram is consumed and it hits swap it's so slow that i get hicups, and that with ssd. i tweaked idea to use less ram, and if and i can keep it below 7.x it works ok. i have no idea how u can use anything with 4gb. on my macbook it's the same but i dont get hickups if i hit more then 16gb on my machine.
I have quite a few more tabs open at any point in time including some large spreadsheets so 8GB tends to be my minimum before I start noticing on Windows/Firefox even with an M2 SSD. Firefox alone is at 4GB for me. This is also my fault for not snoozing/bookmarking aggressively.
This laptop's bigger brother, the Motile M142, has 8GB and a faster CPU. I got one last month for $299 and I'm very happy so far. Runs Ubuntu great.
I have never owned a computer with more than 4GB of RAM. It's fine as long as you don't embed separate copies of chromium in your text editor, IRC-substitute, and music player.
More RAM -- and using the ECC kind -- definitely improved my compilation speeds.

But maybe your workflow doesn't include compilation?

Was that a non sequitur or are you saying that ECC ram had an impact on compilation speed?
I'm impressed that you can be productive with 4GB. On my dev laptop, I am often hitting 14-15GB usage. Chrome, Docker, NodeJS, and VSCode are the main culprits. I'm upgrading the machine to 32GB this weekend.
2gb is for the gpu so it’s really 2gb verses 6gb
Intel really hit the jackpot with Clear Linux, it's the best performing distro for Machine Learning/Deep Learning now, often being 30% faster than regular distros like Ubuntu. Not sure how they did it, but I am enjoying it very much! Thanks!
Consensus above is performance governor and compiler flags.
I have no idea why phoronix isn't highlighting the reason for Clear Linux superior performance, it isn't magic:

Clear Linux OS sets the CPU governor to performance which calls for the CPU to operate at maximum clock frequency. In other words, P-state P0. While this may sound wasteful at first, it is important to remember that power utilization does not increase significantly simply because of a locked clock frequency without a workload.

from: https://docs.01.org/clearlinux/latest/guides/maintenance/cpu...

The only reference in the article for this is in the compiler details box, where you can see that all the distributions run the ondemand governor, and Clear runs with performance.

So, what's the catch? Why doesn't everybody do that?
Had to guess: Perhaps Linux users are more comfortable keeping an eye on the CPU meter, and generally knowing what their apps are requiring of the computer. If you can be bothered with that there's no reason to mind the CPU maximums automatically, and maybe CPU capacity is even more important to you than battery life.
Most people doesn't care for maxing the performance of their CPU most of the time, and they care more about power consumption. This is why that's the default on Windows too. Also, if everybody would do it it will have impact on global power consumption. Anyone can change the governor if they want, it isn't complicated.

Phoronix also could have change the governor and benchmark using the same governor, but for some reason he choose not to. This isn't the first time he does it, the comparison is a repeated theme in a series of articles he made. Unless I miss something, which very well could be, this doesn't reflect well on the site professional credibility.

(comment deleted)
Super crappy battery life, fans are loud, and heat.
> thermald is a Linux thermal management daemon used to prevent the overheating of platforms.

> By default, thermald is disabled in Clear Linux OS.

It sounds like the catch is that your computer will happily fry itself.

thermald is good for ordinary users I guess.

I hate those things though. I disabled all of those thermal and laptop mode management daemons on my Fedora laptop. I ran a CPU stress while manually tweaking the Intel RAPL settings until I found the 80C limit and just set that on boot. I also wrote into the boot scripts the powertop settings that I liked.

Even without RAPL settings the laptop won't fry itself. You can see the machine check events in dmesg as the CPU hits 90-95C and throttles.

I thought CPUs and GPUs throttled themselves to keep the thermals safe?
You're right.

I had a W520 Thinkpad that would shut off if it got too hot and wouldn't enter to BIOS until it cooled.

(Probably is true of many laptops; that one seemed to frequently overheat on me during intensive, prolonged compilations.)

Open the laptop up and reapply the thermal paste. It dries up over time as it gets exposed to higher temperatures, and then you get constant overheating problems as the heatsink can no longer cool the CPU effectively.
Safe for the CPU is a pretty low bar: the Tjunction of their latest mobile CPUs is 100°C, and I'm pretty sure most users would not be very happy with using a laptop at that temperature.
Some laptop models (particularly the more "fruity" ones) get awfully close to that Tjunction number. Close enough that there's most likely a decent case that the hardware is taking quite a bit of actual wear and tear from those temperatures.
Have you used a Macbook? Because they will happily get to 99C before their fans are even at 100%.
Thermald is relatively new (Intel made) fan modulation system designed mostly to

(1) keep boost frequencies running as long as possible.

(2) keep your fan from constantly turning off and on.

Your system has its own fan speed curve & power envelope that works no matter what the system is doing. Not having thermald does not somehow cause your system to overheat (for example many AMD systems aren't support)

Turning thermald off is Intel signally that they have an on-silicon/in-kernel/in-EFI/partner guidance method that now does as equal or better job.

There is a lot more that Clear Linux does than just set the CPU frequency governor, as shown here among other articles: https://www.phoronix.com/scan.php?page=article&item=ubuntu18...

There is also various compiler tuning, a number of patches to different components, using PGO + LTO compiler optimizations, and much more.

Right, many of those changes make it much less suitable distro for your average Linux poweruser. To me it make sense to highlight those facts in such articles, instead of presenting it as the best distro that knocks out all the rest.

The reason that this irritates me is that after stumbling on one of those comparisons I was puzzled how could this be and was thinking maybe I should switch. I went to Clear's websites and read through the whole thing until I understood that thank you, this is nice but not for me. While this was educating, it was also waste of my time. My view is that good tech journalism should save me time, not the opposite.

for example?
Running the CPU at 100 percent and disabling thermald are probably enough for rendering it unsuitable for most people.
If I disable speedstep on my i7 8700K and lock it at 5.0GHz, CPUZ shows it running at a steady 5.0Ghz. But my temps stay relatively low when little is running. Starting Prime 95 Torture Test though, it quickly heats up as you'd expect and cools back down when I stop the test. But the whole time, the frequency never moves from 5.0Ghz. Do you or anyone know what's going on there?
How does the lack of thermald make it unsuitable? I run arch Linux and I've never installed nor used it (there's not even a package for arch in the main repos). Your firmware already knows when it needs to throttle.

AFAICT, what thermald gains you is a smoother degradation of performance; it starts throttling earlier and more gradually than would happen otherwise. It's not obvious to me that there's any actual benefit of this; thermal limits are what they are, and you pay the price eventually if you really need all those cycles. Whether my system is slightly slower for a longer period of time or much slower for a shorter period of time doesn't really seem terribly significant.

Oh, I'm preety sure that Clear is unsuitable for most Linux users, thermald included or not, for many reasons. Specifically regarding thermald, I believe most people want their fans to kick early on low RPM and wait for the CPU to heat up and then max the fans out. The reason I believe so is that the whole industry have invested and still investing huge amount of money and complexity in making that possible and that is the default behavior of all hardware. Almost. There is right now top post regarding 2019's macbooks not behaving that way and the frustration it cause users.

It's not that I think that Clear is a stupid unnecessary project or anything like that, it is actually quite interesting experimental work. But it an experimental project and most people doesn't want something that avant-garde as their daily driver.

what kind of poweruser doesn’t run as fast as possible?
The essence from the linked article:

> - Switching to the P-State "performance" governor on Ubuntu 18.10 only allowed it to win over Clear Linux in an extra 5 benchmarks... Clear Linux still came out ahead 66% of the time against Ubuntu either out-of-the-box or with the performance governor.

That's a wonderful insight and I didn't even know you could do that. After some searching found this(1). Definitely trying this on my desktop today.

(1) https://askubuntu.com/questions/1021748/set-cpu-governor-to-...

Edit: I tried it. Didn't make any noticable difference to me except fan is making a constant loud noise. FWIW, Webpack still takes the same time to compile (not even difference of 100ms before and after).

Intel probably just compiled it with their own compiler, purpose built for their cpus. But with distros I dont care about speed (unless something insane is going on like thermald on thinkpads) but rather how much hardware works well (giving a sharp stare to lenovo and their fingerprint readers in thinkpad series).
That laptop has an AMD processor.
And it does that largely by disabling all thermal management. Thanks, but...
(comment deleted)
(comment deleted)
(comment deleted)
It's worth noting that the $199 price tag for this laptop may be inaccurate or temporary. I looked it up on Wal-Mart's website just a moment ago and it shows $279, marked down from $500:

https://www.walmart.com/ip/Vega-HDMI-Front-RAM-Rose-FHD-4GB-...

Not saying it wasn't $200 at the time he bought it, just saying that he might've bought it on sale (there was probably a bunch of Linux enthusiasts clamoring, "There's a laptop on sale for $200 at Wal-Mart that can run Linux!"); it doesn't appear to be its retail price anymore.

It will be interesting to see how these benchmarks look with more memory.

Could it be that Windows just eats more base memory than Linux, and the benchmark apps do not have enough memory to run under Windows?

Yeah that's great and all until you want to play any decent game then its not supported or chews 1 million gigawatts of power. Yay for performance mode!