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

[birdie]

ARM as well as Apple have been using BIG-little designs for many years now. AMD has been rumored to migrate to a BIG-little uArch as well somewhere in the future. Maybe what Intel has implemented is not that bad after all. Maybe if the Linux kernel properly supported ADL CPUs we wouldn't have such a salty conversation.

[perpetually high]

At the end of the day, Alder Lake is a new platform and you're paying the early adopter penalty and tax in every way. For what? To set some benchmarks? F***ing lol.

Yeah, let me buy this super fast CPU, and then undervolt it right away, tinker with this and that BIOS settings, disable this E core, do this, do that.

Or, buy an established Zen 3. Enable XMP, PBO Auto. Done. No problems.

[birdie]

No, look again. The benchmarks show the opposite. Several of these show clearly that one gains performance by disabling the 8 E cores.

The Linux kernel doesn't understand the ADL CPUs topology. Even the newly released Windows 11 which was supposed to support ADL perfectly struggles with some applications and Windows 10, despite not knowing or understanding ADL CPUs, sometimes works better:

[birdie]

I tried setting that before for someone, and unfortunately it didn't work. I'm wondering if it might just be for private messages or something.

This is a known issue with vBulletin - for some users it works, for others it doesn't. I've asked Michael to migrate to something a little bit better maintained, e.g. XenForo, but he declined.

[sdack]

The article clearly shows the best perf and best perf/watt was achieved ...

No, look again. The benchmarks show the opposite. Several of these show clearly that one gains performance by disabling the 8 E cores.

[smitty3268]

The setting, Ignore List, is at the bottom of the page: https://www.phoronix.com/forums/settings/account

I tried setting that before for someone, and unfortunately it didn't work. I'm wondering if it might just be for private messages or something.

[Markopolo]

From what I can see it looks like a HT'd P-core has about the same performance as a P-core + E-core combo. While there might be some performance left on the table due to suboptimalscheduling, I wonder if a chip with, say, 12 P-cores wouldn't be a better design than this hybrid mess. It probably wouldn't use much more power that 8P+8E chip, it wouldn't need all kinds of software trickery to work properly and AVX-512 would work out of the box. And considering that Sapphire Rapids will have no E-cores, it'd clearly be doable. I get it that Intel needs something to compete against ARM in the mobile segment but this hybrid core idea looks like a step in the Itanium direction. It *might* be more interesting in laptops.

I wonder this as well. The scheduler issues are significant though. Look at the average power consumption for the two scenarios. P only was 50% more power. That’s significant enough to make me hesitant to compare the performance. Also keep in mind intel said that one P core has the same size as 4(!) E cores

[Markopolo]

Sorry, but you are wrong.

What users expect from a CPU is that it runs all of their software well and not just some of it. These benchmarks show one has to look much closer with Alder Lake than before and one is possibly forced to tweak BIOS settings on Alder Lake systems in order to get the most out of them.

This is definitely not what users want. Geeks and nerds may enjoy experimenting with BIOS settings, but the majority of users and developers expect that it just works and that it does so well. So yes, it is a mess Intel has created here, and it is left to others to sort it out.

The article clearly shows the best perf and best perf/watt was achieved using the stock configuration. Doesn’t seem to indicate anything is a mess to me. (If you’re gonna comment on an article the least you can do is read it).

[RedEyed]

Firefox (on Win here but also on Linux) only loads half the page, mostly on brand new articles and says "waiting for TLS hand shake"
Internetexplorer hangs a good while but manages to load the page in the end.
load.jpg
If you give it enough time (10 min maybe?) firefox completes the webpage loading.

It's loading almost instantly here in Kyiv

[mr_marmalade]

Perfect, thank you birdie.