Hacker News Reader

Comment by eknkc | Hacker News Reader

So, this got me curious. Assuming a cpu heavy benchmark with no syscalls (calculate primes or something like that) I’d not really expect a major difference between OSs. I guess you have the scheduler possibly fucking things up and the memory management. Maybe?

What else could contribute to a raw cpu benchmark difference?

View on HN

There are such an enormous array of possible influences on this, most likely being the power profile.

By default Windows on a laptop (in this case a Zenbook S 14 with Core Ultra 7 256V) uses a reduced power mode where it limits the max frequency and is less aggressive in scaling up core frequency because of the greatly reduced return per unit of power (which can destroy battery life and make heat profiles unacceptable).

This comparison makes no mention of the power profile, nor does it observe battery usage if applicable.

This looks like a misleading if not completely useless clickbait comparison. Others mentioned schedulers, and while schedulers can account account for a percent or two difference across this sort of wide workload, instead a huge disparity is seen.

Or the more obvious answer: newer binaries/libraries/kernel for Linux. Is it still hard to believe Windows leaves a lot performance on the table? And it does mention powersave governer for Ubuntu.

Coworkers switched from Win11 to Ubuntu, our main Spring Boot project compiled time was cut in half. I guess ntfs is a factor in this specific use case, it's a medium sized project with a few thousands files and classes.

NTFS is 100% the problem here. It really struggles with many small files type use cases. In pretty much all my projects, I notice a huge compile time cut in Linux and Mac as opposed to Windows.

Also a lot of typical Linux pipelines become unbearably slow if you try to replicate them in Windows.

Have you thought about using ccache since you have so many files to compile? Although I assume after the initial compile it’s not that big of a deal.

Of course. Mine is just an example to explain that there are many ways to improve Windows performance

the scheduler has been an issue for windows for quite some time... along with the built-in anti-virus (and ads).

It's almost certainly the scheduler, IMO.

> On average the Ubuntu 24.10 release was about 14% faster than Microsoft Windows 11 out-of-the-box on this Lunar Lake powered ASUS Zenbook S 14 laptop.

What else was the "Out-of-the-box" Windows installation trying to do while they were running the benchmark? Sitting here on my Windows 10 installation with nothing but Firefox and Task Manager open, there are dozens of things clamoring for the attention of the scheduler and using CPU:

https://i.imgur.com/NxBOBXm.png

Yeah, I've installed several of these services on purpose, but I've also uninstalled or inhibited a bunch of things that I didn't need.

Wonder if it is putting things onto p-cores more aggressively. Would need to the power graphs at the same time to see that.

>What else was the "Out-of-the-box" Windows installation trying to do while they were running the benchmark?

It's trying to maximize the performance to power usage ratio out of the box. The default power profile is very conservative, which on a laptop means better battery life and reduced heat. Given that this comparison didn't do even rudimentary power consumption, heat, or CPU frequency scaling measures, it is overwhelmingly likely as the difference.

> there are dozens of things clamoring for the attention of the scheduler and using CPU

Linux machines have hundreds of processes too. It simply doesn't matter in 2024. That isn't a factor and isn't relevant, nor is the scheduler to blame.

This is a silly misleading benchmark that is de facto clickbait. Pretty surprisedto see it doing well here.

You have shitton of custom soft there

Power management maybe.

Power government seems like a likely candidate.

If it hasn't changed since 24.04 the default power governor in Ubuntu 24.10 is "powersave". I'd be curious to see how performance improved if Phoronix tested with the "performance" governor.

It could go either way honestly. The "performance" governor tries to max out the CPU clock rate all the time but because of transient thermal capacity this is not always optimal. The CPU will rip as hard as possible even under powersave, and it might hit a higher transient peak because it starts out cold.

Reply on news.ycombinator.com