Linux 5.16's New Cluster Scheduling Is Causing Regression, Further Hurting Alder Lake

[Alex/AT]

It's fun how Intel is repeating the mistake of AMD FX right now.

[mSparks]

You're now claiming i'm saying things I absolutely never did, so this conversation is over.

This wasn't you?

It is true, and there are plenty of benchmarks out there to prove it if you don't believe me. Just go look at some of them. Try anandtech's, for a start.

Or you've just changed your mind about the truth of e-cores giving better multithreading performance than p-cores or anything AMD has to offer?

[smitty3268]

...

You're now claiming i'm saying things I absolutely never did, so this conversation is over.

[mSparks]

Because they're completely different markets?

Not anymore, AMD changed that paradigm 5 years ago, and left Intel with nearly zero market share since.
https://www.youtube.com/watch?v=4Z7EzcADIJM

Again, the entire conversation was about Intel p-cores vs Intel e-cores. You seem to be trying to change the conversation now to a completely different topic. Which is fine if that's what you want to do, but every post I made above was specifically about the Intel side only. I never even mentioned AMD or any other non alder lake chip.

Exactly, So this is competing with the Threadripper range, which competes with zen3 - more cores for people who also game, except Intel don't seem to have grasped that even 5 years later.
You mentioned benchmarks, specifically ones where

The 12900K flagship draws 250+ Watts in part so it can beat AMD in more benchmarks,

Except it doesn't beat AMD in any way sense or shape. Either by drawing 250W, or adding E-cores to target people who are buying Threadrippers.

You also quote a benchmark that clearly shows there is a need for more than one or two p cores on desktop chips, while saying its true there isn't.

[smitty3268]

What good reason?

Because they're completely different markets? It's like comparing it to a laptop chip and determining that the desktop cpu is faster. Well, obviously.

What truth is there in claiming Intel now has the top performing CPU

Again, the entire conversation was about Intel p-cores vs Intel e-cores. You seem to be trying to change the conversation now to a completely different topic. Which is fine if that's what you want to do, but every post I made above was specifically about the Intel side only. I never even mentioned AMD or any other non alder lake chip.

[mSparks]

This entire discussion was about Intel e-cores vs Intel p-cores.

Nobody is comparing a desktop part against servers for good reason. Same with HEDT.

What good reason?

Why not compare an $820 CPU that requires everything else new with a $1,600 CPU that requires everything else new, and is also one of the top selling high end CPUs?

What truth is there in claiming Intel now has the top performing CPU when the top performing CPU is AMD last gen and more than twice as fast as it?

Is that the same good reason they benchmarked an $820 CPU against a $280 CPU then act like its ground breaking that its only a few percent faster than it? (even though both can play all titles at 144fps on a 144hz 4K monitor, and both a long way from being the most powerful CPU you can buy)

[smitty3268]

anandtech compares the 12900K with EPYC 7763 or Threadripper 3960X?

This entire discussion was about Intel e-cores vs Intel p-cores.

Nobody is comparing a desktop part against servers for good reason. Same with HEDT.

[ms178]

AFAIK, there was never a 'performance' regression from the cluster scheduling on the AMD side... IIRC from what came up on the kernel mailing list, it was a regression over spamming of dmesg with topology information due to not being supported.

Indeed, but there could be some performance implications by this post in the LKLM thread: http://lkml.iu.edu/hypermail/linux/k...0.2/05938.html

After all basing scheduling decisions on nonsensical topology information could lead to disaster.

[mSparks]

It is true, and there are plenty of benchmarks out there to prove it if you don't believe me. Just go look at some of them. Try anandtech's, for a start.

anandtech compares the 12900K with EPYC 7763 or Threadripper 3960X?

Or just stick with existing benchmarks.

https://www.cpubenchmark.net/high_end_cpus.html

AMD EPYC 7763
(86%)
87,767
$7,890.00

first Intel

Intel Xeon Platinum 8380 @ 2.30GHz
(61%)
62,317
$8,099.00*

....

AMD Ryzen Threadripper 3960X
(54%)
55,085
$1,599.00

Intel Xeon W-3265M @ 2.70GHz
(39%)
39,765
$6,353.00*

AMD Ryzen 9 5900X
(39%)
39,524
$524.00

Intel Core i9-12900K
(36%)
37,231
$818.91

~twice the price, less performance, ~3 times the power consumption "for the gamers"

You can watch "the market speak" here
https://www.cpubenchmark.net/share30.html

I have a lot less faith than Intel it seems that gamers are that stupid.

[avem]

Theres an easy way to tell when a benchmark graph is paid for by intel/not telling the truth.

TPU paid by Intel? Sorry, didn't bother to read the rest of your comment