One-Line Patch For Intel Meteor Lake Yields Up To 72% Better Performance

Michael Larabel reports via Phoronix: Covered last week on Phoronix was a new patch from Intel that with tuning to the P-State CPU frequency scaling driver was showing big wins for Intel Core Ultra "Meteor Lake" performance and power efficiency. I was curious with the Intel claims posted for a couple benchmarks and thus over the weekend set out to run many Intel Meteor Lake benchmarks on this one-line kernel patch... The results are great for boosting the Linux performance of Intel Core ultra laptops with as much as 72% better performance. [...]

When looking at the CPU power consumption overall, for the wide variety of workloads tested it was just a slight uptick in power use and thus overall leading to slightly better power efficiency too. See all the data here. So this is quite a nice one-line Linux kernel patch for Meteor Lake and will hopefully be mainlined to the Linux kernel for Linux 6.11 if not squeezing it in as a "fix" for the current Linux 6.10 cycle. It's just too bad though that it took six months after launch for this tuned EPP value to be determined. Fresh benchmarks between Intel Core Ultra and AMD Ryzen on the latest Linux software will be coming up soon on Phoronix.

What about Windows?

[DrMrLordX]

Is Meteor Lake having the same problem in other operating systems?

Re:What about Windows?

[retchdog]

Nah, Intel probably informed their industry partner (it's called "wintel" for a reason) and left the open source dweebs to twist in the wind.

Re:

[TechyImmigrant]

Nah, Intel probably informed their industry partner (it's called "wintel" for a reason) and left the open source dweebs to twist in the wind.

Dunno about that. Linux is pretty popular amongst the techiest of the techies within Intel. But wrangling all the power states and policies with heterogeneous cores is beyond my ken. I'll stick to cryptography thanks.

Let's hope that this "EPP value"...

[ffkom]

... is not lowering the stability as much as it improves performance. Would not be the first time Intel power settings where involved with platform stability [guru3d.com].

Re:

[DamnOregonian]

To be fair, Intel's fix for the 13th and 14th gen problem is "to use Intel settings, and not those of the Motherboard manufacturer."
Motherboard manufacturers have a history of doing bad things [theverge.com] to CPUs.

In this particular case (EPP value), it was Intel that selected the new value.

Re:

[alvinrod]

Intel is as much to blame here as they're not enforcing any kind of standards, which means that the motherboard manufacturers are naturally going to go as hard as they can or harder than the next company so that they've got the bigger bars in the product reviews. Naturally Intel doesn't discourage this either as it makes them look more competitive against AMD even if their CPUs are drawing over 200W just to do so. When the excessive voltage/current cause the silicon to degrade Intel just points their finger

Re:

[DamnOregonian]

Intel is as much to blame here as they're not enforcing any kind of standards

Sure, then AMD is to blame for AMD motherboard manufacturers turning AMD chips into smoke machines.

which means that the motherboard manufacturers are naturally going to go as hard as they can or harder than the next company so that they've got the bigger bars in the product reviews.

Kind of like allowing voltage adjustments on the 7800X3D and 7950X3D, even though AMD said not to?

Naturally Intel doesn't discourage this either as it makes them look more competitive against AMD even if their CPUs are drawing over 200W just to do so.

Are you fucking stupid?
Naturally, AMD doesn't discourage this either as it makes them look more competitive against Intel even if their CPUs need to draw 400W to beat them.

When the excessive voltage/current cause the silicon to degrade Intel just points their fingers at the board manufacturers as though it's somehow not their problem.

Uh, that's exactly what AMD did.

What the fuck is broken in your brain, dude?

"up to" doing a lot of heavy lifting

[Anonymous Coward]

On one specific benchmark it was 72%, on average it was 7%.

Re:

[DamnOregonian]

Yes.... that is what "up to" means. Is there an alternative formulation of that sentence you'd like to see?

Re:

[Cyrano de Maniac]

"typically 7%, and in some limited scenarios up to 72%"

The hard part

[Tony Isaac]

is knowing which *one line* to patch.

Wish described what that one line was better.

[OneOfMany07]

What is an EPP value? What does it change? Why is that one line like black magic with an opaque sounding 'magic number'?

Oh well. Point AI at it soon. Profit... right?

Re:Wish described what that one line was better.

[DamnOregonian]

It's a bias in the intel power management driver/hardware for selection of frequency vs. power consumption.
It very much is a black magic number, because the effect it has is not straight-forward to predict.

For example, if you set the EPP to 0 (biased toward frequency/performance) may vastly increase power for very little actual gain.
It will also reduce the power available to the iGPU to near zero, so if you're using it, you're going to be unhappy about its new 150Mhz clock rate.

Awesome

[Mes]

Does this optimization also come with another built in Spectre type vulnerability?