News Single-line Meteor Lake Linux patch boosts performance by 72%

[Admin]

Intel recently released a one-line code patch for its Meteor Lake CPUs, resulting in huge performance wins in Linux. That single line of code fine-tunes the P-State CPU frequency scaling driver.

Single-line Meteor Lake Linux patch boosts performance by 72% : Read more

 

[danwat1234]

Yes, but what does that parameter do? What are the actual changes to clock speeds or pcore versus ecore utilization?

 

[cyrusfox]

Yes, but what does that parameter do? What are the actual changes to clock speeds or pcore versus ecore utilization?

Had to look this up, values close to 0 for EPP prioritize performance, close to 255 prioritize power savings.

The P-State Energy Performance Preference (EPP) is a setting within Intel's P-State (Processor State) power management system that allows the operating system or firmware to balance the trade-off between power consumption and performance. The P-State system is part of Intel's SpeedStep technology, which dynamically adjusts the processor's clock speed and voltage to save energy when full performance is not needed and to provide performance when it is demanded.​

Changing the default from 128 to 115, then ending up 64, this favors performance over power savings in theory... was there an appreciable change to battery life/consumption is the question I am interested in?

I would think their would be a plugged in profile that would set the EPP aggressively(0-64), then it would swing to a different value/mode when on battery depending on user setting(128-255). I don't know how Linux handles this, that has always been one my complaints on running Linux on old laptops. I never seem to get the power settings right and they drain themselves aggressively doing nothing. Apple is king when it comes to mobile computing experiencing in my limited experience, Windows is ok, and Linux is garbage. This change likely reenforces my experience. Why is this not more customer facing setting which can be set in a custom power profile on Linux?

 

[bit_user]

Single-line Intel Meteor Lake Linux patch boosts performance by 72%

What a load of clickbait! I mean that's just factually incorrect. The article states it's "up to 72%", and that was an extreme outlier case! The geomean is only 6.7% faster. Not only that, but highlighting the size of the patch is misleading when it's really just a tuning adjustment to a constant for one of the power plans.

As you might guess from the above, the additional performance comes at a price, which is paid in more power. 4.9% more power, according to the test suite averages that are presented at the end of the phoronix article. So, it's not exactly a free lunch - especially if you care about long battery life or don't like your laptop fan making a ton of noise.

Larabel hopes the patch will be included in the Linux 6.11 Git kernel, if not applied as a maintenance fix for the current Linux 6.10 cycle.

The sad part is that it's compiled directly into the kernel, rather than being a loadable module or maybe read from a config file, better yet!

 

[bit_user]

I don't know how Linux handles this, that has always been one my complaints on running Linux on old laptops. I never seem to get the power settings right and they drain themselves aggressively doing nothing.

The Skylake laptop I got in 2017 never seemed to have abnormally poor battery life, though I mostly used it plugged in. You can indeed have different power plans come into effect, depending on whether it's using A/C or battery power.

This change likely reenforces my experience. Why is this not more customer facing setting which can be set in a custom power profile on Linux?

Those constants are hardware-specific. They're one level below what a user is meant to adjust. I think the idea was that, absent detailed knowledge of the hardware, it wouldn't make sense for a user to adjust those.

Intel's p_state driver exposes an EBP value (Energy/Performance Bias) that ranges from 0-15, which you can tune according to your preference. It's all described here:

https://wiki.archlinux.org/title/CPU_frequency_scaling#Intel_performance_and_energy_bias_hint
​

 

[TerryLaze]

The geomean is only 6.7% faster. Not only that, but highlighting the size of the patch is misleading when it's really just a tuning adjustment to a constant for one of the power plans.

As you might guess from the above, the additional performance comes at a price, which is paid in more power. 4.9% more power, according to the test suite averages, at the end of the phoronix article. So, it's not exactly a free lunch - especially if you care about long battery life or don't like your laptop fan making a ton of noise.

It uses less more power than the more performance it provides, so especially if you care about long battery life it has a positive effect.

 

[bit_user]

It uses less more power than the more performance it provides,

Yes, that's why they can claim it improves efficiency.

so especially if you care about long battery life it has a positive effect.

It only benefits battery life if you're using your laptop to do a set number of fixed-length tasks, like perform a certain number of renders or code compiles. However, if you just start another render or compile pretty much as soon as the previous one finished, you'll run out of battery sooner (but having gotten more done, before then).

If you use it for a task that soaks up the performance, like gaming, then the greater power utilization will simply and directly translate into shorter battery life. Your only benefit would be slightly higher framerates.

 

[TerryLaze]

If you use it for a task that soaks up the performance, like gaming, then the greater power utilization will simply and directly translate into shorter battery life. Your only benefit would be slightly higher framerates.

Extremely doubtful unless we are talking about a laptop with a big enough GPU and in that case you have to game while plugged in anyway.
When gaming on a laptop most of the power draw comes from the GPU, the CPU only works as hard as it has to to reach the FPS the GPU can reach.