Intel Core i9 12900K "Alder Lake" Continues Moving In Right Direction With Linux 5.17

[phoronix]

Phoronix: Intel Core i9 12900K "Alder Lake" Continues Moving In Right Direction With Linux 5.17

Linux 5.16 brought improvements/fixes for Alder Lake S to help the likes of the Core i9 12900K deliver better performance. With the in-development Linux 5.17, the i9-12900K is looking even better. And then on the horizon for Linux 5.18 is the possibility of even better performance thanks to Intel's Hardware Feedback Interface (HFI/EHFI) positioned to land...

https://www.phoronix.com/scan.php?page=news_item&px=Core-i9-12900K-Linux-5.17

[bug77]

The thing is, does it matter that performance improves when workloads can still end up on the wrong core?

[smitty3268]

The thing is, does it matter that performance improves when workloads can still end up on the wrong core?

Is that still an issue in 5.16?

[lamprey]

The thing is, does it matter that performance improves when workloads can still end up on the wrong core?

I'm having difficulties parsing this. The article shows benchmarks of specific tasks where the time taken has decreased as performance has increased for these CPUs with later kernels. What does it matter to you or these tasks what type of core is used?

https://www.phoronix.com/scan.php?pa...For-Linux-5.18 the thread director seems to be on course for 5.18.

[birdie]

Is that still an issue in 5.16?

https://www.phoronix.com/forums/forum/hardware/processors-memory/1307095-intel-hfi-to-premiere-in-linux-5-18-for-improving-hybrid-cpu-performance-efficiency?p=1307124#post1307124

[bug77]

I'm having difficulties parsing this. The article shows benchmarks of specific tasks where the time taken has decreased as performance has increased for these CPUs with later kernels. What does it matter to you or these tasks what type of core is used?

https://www.phoronix.com/scan.php?pa...For-Linux-5.18 the thread director seems to be on course for 5.18.

Some benchmarks are not repeatable. One pass would have a CPU-intensive thread on a P core, sometimes on the E core. It happens when you have cores to spare, And while benchmarks are heavily multi-threaded in order to their job properly (i.e. stress the CPU), most real-world workloads are not.

For the record, I all for optimizations. I was just pointing out, for select AL SKUs, other things are still more important.

[lamprey]

Some benchmarks are not repeatable. One pass would have a CPU-intensive thread on a P core, sometimes on the E core. It happens when you have cores to spare, And while benchmarks are heavily multi-threaded in order to their job properly (i.e. stress the CPU), most real-world workloads are not.

For the record, I all for optimizations. I was just pointing out, for select AL SKUs, other things are still more important.

Ah..sounds somewhat reasonable. I don't know enough about how the current linux scheduler works to know whether it would likely choose the same for a particular benchmark, and maybe that's why it wasn't mentioned in the article.

Anyway, if the same types of cores were used for all runs then I consider these gains significant, and haven't read enough about Alder Lake to have an opinion on whether more or less so than a functioning thread director. (Mostly because there are two mini-itx boards currently in Sweden for socket 1700, expensive - getting more expensive and very expensive, both z690)

[smitty3268]

https://www.phoronix.com/forums/foru...24#post1307124

That has nothing to do with my question. I think we all know 5.16 isn't 100% optimal.

What I was curious about was whether 5.16 is still completely borked in terms of scheduling things extremely dumbly, or it it's generally in decent-ish shape at the moment and just needs polishing. Some of the gains shown in the benchmarks of this article look like it could be that type of thing - however, i haven't checked to see if that's really the case or not.

And the reason I bring this up was that i think 5.16 had some fixes to how the BIOS info was parsed to tell the kernel about the different capabilities between p and e cores. Similar to how it handles hyperthreading.

[bug77]

Ah..sounds somewhat reasonable. I don't know enough about how the current linux scheduler works to know whether it would likely choose the same for a particular benchmark, and maybe that's why it wasn't mentioned in the article.

This was mentioned in the first round of AL testing, when some workloads (superpi or smth like that) scored suspiciously low.

Anyway, if the same types of cores were used for all runs then I consider these gains significant, and haven't read enough about Alder Lake to have an opinion on whether more or less so than a functioning thread director. (Mostly because there are two mini-itx boards currently in Sweden for socket 1700, expensive - getting more expensive and very expensive, both z690)

I feel for you, my 12600k is also looking for a board. Slim picking and pretty bad considering features/$ you get

[smitty3268]

That has nothing to do with my question. I think we all know 5.16 isn't 100% optimal.

What I was curious about was whether 5.16 is still completely borked in terms of scheduling things extremely dumbly, or it it's generally in decent-ish shape at the moment and just needs polishing. Some of the gains shown in the benchmarks of this article look like it could be that type of thing - however, i haven't checked to see if that's really the case or not.

And the reason I bring this up was that i think 5.16 had some fixes to how the BIOS info was parsed to tell the kernel about the different capabilities between p and e cores. Similar to how it handles hyperthreading.

Well, turns out Michael was running these tests already, and the poor behavior of 5.15 is largely resolved in 5.16. So that's good news.