Intel i9-12900K Alder Lake Linux Performance In Different P/E Core Configurations

[MadCatX]

I have been running Debian with a custom-built kernel and Mesa on my Apollo Lake Atom laptop for the last three years.
Two weeks ago I threw Debian out and put Windows 11 on it. The difference in performance is immediately noticeable. Web browsers and other heavy applications like productivity suites no longer randomly stall for a minute when scrolling through >20 tabs or multiple pages in a docx file loaded with lots of images, photos and tables.

That sounds like an OOM problem that the multigeneration LRU patches might alleviate. Did you try that?

[mdedetrich]

And yet Alder Lake performs much better on Windows than it does in Linux.

This may be true but it doesn't have anything to do with the scheduler because Windows 10 (which doesn't have the new scheduler specifically for Alder lake) also beats Linux, this seems to be more due to processor support

Intel is not trying to provide an "automagic" solution.

Developers can still choose to directly schedule workloads onto each processor type (or ideally affinity hint the OS as to what the workload type is), see this:
"However, it may be more optimal to run background worker threads on the Efficient-cores. The API references in the next section lists many of the functions available, ranging from those providing OS level guidance through weak affinity hits, such as SetThreadIdealProcessor() and SetThreadPriority(), through stronger control like SetThreadInformation() and SetThreadSelectedCPUSets(), to the strongest control of affinity using SetThreadAffinityMask()."

The idea is that both the software and the hardware (Thread Director) are providing hints to the OS (Windows in this case) and the Windows scheduler matches up the workload to the right core, Intel isn't forcing any sort of automagic scheduling (again Thread Director only gives hints to the OS about the current state of the cores).

See the diagram and description in this section: "Intel® Thread Director and Operating System Vendor (OSV) Optimizations for the Performance Hybrid Architecture"

And again to reiterate, a developer can also still choose hard affinities if they want "through stronger control like SetThreadInformation() and SetThreadSelectedCPUSets(), to the strongest control of affinity using SetThreadAffinityMask()".

All of the quotes are from this intel developer guide: https://www.intel.com/content/www/us...per-guide.html

I actually don't disagree with this, my point is precisely that to get any real measurable result you DO need to use affinity hint for the OS. The issue is that Intel has been pushing the idea that the scheduler will solve this issues without requiring the processor affinity which is what I am calling snake oil.

Its also an obvious difficult problem because programs have to be coded specifically to take advantage of the affinity hints, hence why its not surprising that Intel is pushing this because they don't want to admit its a lot of work to get tangible benefit out of it/

[perpetually high]

Two weeks ago I threw Debian out and put Windows 11 on it. The difference in performance is immediately noticeable.

That's because you are a Linux newbie and have no idea what you are doing. Not a Linux problem. That's a Sonadow/birdie/whatever-your-other-usernames-are problem.

[Sonadow]

Do you have any data to back this up? Because my experience is totally oposit to yours.

I have been running Debian with a custom-built kernel and Mesa on my Apollo Lake Atom laptop for the last three years.
Two weeks ago I threw Debian out and put Windows 11 on it. The difference in performance is immediately noticeable. Web browsers and other heavy applications like productivity suites no longer randomly stall for a minute when scrolling through >20 tabs or multiple pages in a docx file loaded with lots of images, photos and tables.

[MadCatX]

Which is the most basic principal that nobody seems to get.

What about this principle do you believe that people don't get?

If an application is in focus, it means that the user intends to use it right there and then. There is no reason it should get shafted by assigning it to a lower-priority queue or an E core in Alder Lake's case.

What if you are rendering a scene or encoding video in the background while working your report in Word or reading something in the browser? What if your workload consists of multiple processes that exchange data and both require a lot CPU power; like playing and streaming a video game?

This has been the default behavior for Windows since Windows Vista, unlike Linux where every user application gets assigned a nice 0, regardless of the amount of resources and focus time it gets.

Niceness has nothing to do with this. The scheduling problem is about deciding how to allocate CPU resources on the fly. The decision logic must be both accurate, fast and account for the cost of its own calculations. This is not an easy problem and you often run into situations where you'd have to know the future in order to make a sane decision.

Right, a production-use scheduler that Intel and Microsoft worked on for Intel's own hardware on Windows and is currently in widespread deployment right now is inferior to the Linux scheduler,

Are you implying that Linux CPU scheduler is not widely used in production?

so much so that the inferior option actually works properly right now while the Intel developers have to keep the code disabled in the Linux scheduler for various reasons.

What code are you talking about here?

[Anux]

A much better "sweet spot" would be to just buy a 12700K. Mostly the same performance while much better power use and cost.

Not true, i have already debunked this.

Fact remains that for all the benchmarks Michael has done about Linux having better performance over Windows, they simply don't carry forward to real-world computing. Till now nobody can provide a reasonable explanation as to why Windows boots, launches programs and generally respond to application inputs faster than Linux on the same hardware, especially on low-power hardware like Atoms.

Do you have any data to back this up? Because my experience is totally oposit to yours.

[drakonas777]

Birdie is definitely not a Intel/NVIDIA fanboy, he just possess a normal need to defend a certain company and a certain product during 9 page comments thread. Just that - normal need to defend company/product, NOT a fanboy :lol: Anyway.

AL is a decent CPU. The main issue is PL2, which is insane on i9. Totally unpractical and beyond optimal and sane working parameters of such a CPU. I guess they needed aggressive power mode to demonstrate more advantage over ZEN3 in the benchmarks. For me personally, none of the K series, except maybe 12600K, makes sense. Non-K and F SKUs will have all the benefits and will cost less.

As for the hybrid architecture on the desktop itself - it has potential in the future, where E cores will grow in a count significantly. However, for Alder Lake, two more P cores instead 8 E cores would make more sense. It would me more universal architecture with less software magic needed, also, perf would be somewhat the same.

[Sonadow]

Windows 11 largely uses the intel hardware scheduler apparently, but overrides in certain places by always making sure a graphical application that has focus gets put on a p-core, for example.

Which is the most basic principal that nobody seems to get.

If an application is in focus, it means that the user intends to use it right there and then. There is no reason it should get shafted by assigning it to a lower-priority queue or an E core in Alder Lake's case. This has been the default behavior for Windows since Windows Vista, unlike Linux where every user application gets assigned a nice 0, regardless of the amount of resources and focus time it gets.

I'd be surprised if their scheduler is actually even on par with linux, it's just that it's done and working while the linux code is completely turned off at the moment so it's doing something brain dead.

Right, a production-use scheduler that Intel and Microsoft worked on for Intel's own hardware on Windows and is currently in widespread deployment right now is inferior to the Linux scheduler, so much so that the inferior option actually works properly right now while the Intel developers have to keep the code disabled in the Linux scheduler for various reasons.

[smitty3268]

And yet Alder Lake performs much better on Windows than it does in Linux.

There is definitely something going on in the Windows scheduler, that much is for sure. Microsoft is no stranger to BIG.little; they dealt with it before in Windows RT on the Tegra 3 and they recently worked with it again on the SQ1 and SQ2 for the Surface Pro X. It will be hardly a surprise if the lack of performance difference for Alder Lake in Windows 10 and Windows 11 simply comes down to the fact that Microsoft has already worked on the scheduler to the point where the version in Windows 10 is punting jobs between the P cores and the E cores properly.

Linux doesn't currently have the code to tell e-cores and p-cores apart, so the scheduler treats them identically, with terrible results. That is easily fixed and will be fixed soon.

Windows 11 largely uses the intel hardware scheduler apparently, but overrides in certain places by always making sure a graphical application that has focus gets put on a p-core, for example. I'd be surprised if their scheduler is actually even on par with linux, it's just that it's done and working while the linux code is completely turned off at the moment so it's doing something brain dead.

[CardboardTable]

I am getting the impression that the biggest issue appears to be Intel trying to provide a solution for something that from at least my OS studies back at uni is not really solvable, i.e. automagic scheduling on big little design that generally works better than the alternative. Big little designs work best when developers specifically code into their applications how to use cores, i.e. if you are virus scanner you would pretty much always want to use an E core, or for background tasks like checking for emails or index'ing for fuzzy file search.

This is because knowing what should run on an e-core and what should run on a p-core is primarily a subjective thing, its when making something deliberately slower but to save power is acceptable which depends on the context of the application.

Intel is not trying to provide an "automagic" solution.

Developers can still choose to directly schedule workloads onto each processor type (or ideally affinity hint the OS as to what the workload type is), see this:
"However, it may be more optimal to run background worker threads on the Efficient-cores. The API references in the next section lists many of the functions available, ranging from those providing OS level guidance through weak affinity hits, such as SetThreadIdealProcessor() and SetThreadPriority(), through stronger control like SetThreadInformation() and SetThreadSelectedCPUSets(), to the strongest control of affinity using SetThreadAffinityMask()."

The idea is that both the software and the hardware (Thread Director) are providing hints to the OS (Windows in this case) and the Windows scheduler matches up the workload to the right core, Intel isn't forcing any sort of automagic scheduling (again Thread Director only gives hints to the OS about the current state of the cores).

See the diagram and description in this section: "Intel® Thread Director and Operating System Vendor (OSV) Optimizations for the Performance Hybrid Architecture"

And again to reiterate, a developer can also still choose hard affinities if they want "through stronger control like SetThreadInformation() and SetThreadSelectedCPUSets(), to the strongest control of affinity using SetThreadAffinityMask()".

All of the quotes are from this intel developer guide: https://www.intel.com/content/www/us...per-guide.html