Slow compilation time on meteor lake laptop

[pietinger]

[nxe9]

[nxe9]

Hi, I have new insights. I came across the following recently published article.
https://www.notebookcheck.net/Lenovo-ThinkPad-E14-G6-laptop-review-Fixes-lots-of-problems-on-the-E14-G5.927075.0.html#

It's about the same laptop series as mine, only with a slightly worse CPU. They test Intel Core Ultra 7 155U and mine CPU is Intel Core Ultra 155H.

The following test is interesting for me.

[pietinger]

[Anon-E-moose]

You could try using taskset with emerges. (man taskset)

That cpu has 6 performance cores w/hyperthreading giving you 12
8 ecores w/no hypertreading
2 ecores low power with no hypertreading

So I would probably use taskset with 0-11 (from what I've read that should be the performance cores) and adjust to -j12 or so.
(the only way I would know if the 0-11 is correct is to try it)

Without using taskset you will sometimes use either the ecores or the low ecores both of which would impact performance.

Edit to add: lstopo should show which cores are which

Eta2: should look like this https://blog.reds.ch/wp-content/uploads/2024/07/ultra7lstopo.png
_________________
UM780, 6.12 zen kernel, gcc 13, openrc, wayland

[nxe9]

@pietinger: They use windows with performance profile.

[Anon-E-moose]

You can use taskset for any application, a browser, or emerges, or anything you really want performance for.

I use it for my qemu vm's.

Should be as simple as "taskset -c 0-11 emerge -v <whatever>" or "taskset -c 0-3 firefox" to keep firefox on the performance cores
or "taskset -c 12-15" to have it use the ecores.

Does "taskset -c 0-11 7z ..." show any performance difference? As 7z is being forced to stay on the pcores.

Edit to add: taskset just pins an app to a one or more cores/threads. Doesn't matter what the app is.

Do keep in mind when assigning cores/threads it's best to assign threads for the core together.
On my amd cpu, threads 0 and 8 are on core 0.
In your case it's a little nicer as your core 0 threads are "0 1" (so nice linear progression)

So if you wanted to give 8 cores/threads to some app you would do something like this "taskset -c 0-7 <application>"
which uses the first 4 pcores, leaving 2 for other work along with all ecores.
_________________
UM780, 6.12 zen kernel, gcc 13, openrc, wayland

[nxe9]

@pietinger

New information...

Gentoo 6.12.7, mitigations off/on, thermald and laptop-mode-tools services off/on, in all cases the max. value is about 32000.

Arch live usb 6.12.1, mitigation is on (checked with lscpu), max test result about 42000. So the result is comparable to that on Windows (performance profile without PPM).

I can reproduce test results very reliably.

In short, arch linux performs about 30% better than gentoo in multicore. There must be some reason for this, but what? Does gentoo-kernel have some sub-optimal configuration?


One difference, on archlinux