These states don't work with my processor.
These states don't work with my processor.
I'm working on an i5-8350U under Linux GNU/Linux, though I suspect it should function similarly on Windows. This setup is being used for testing across platforms, and I plan to stick with the GNU/Linux configuration regardless. Performance was observed via turbostat and confirmed by other tools. Initially, I noticed no C-state settings on my machine. Turbostat lists various C-states, including hardware ones starting with "CPU," which show core residency percentages like C7%, C6%, down to C1%. The kernel seems to request these residency values when following the turbostat manual, and it confirms that the actual hardware states are what matter. This explains why idle states I disabled with cpupower still appear without significant impact, possibly because they map to fewer available hardware states. Through testing, power consumption rises sharply once reaching C1%, with PKG usage around 6 watts and core usage near 15 watts when running in C0. One or two cores are offloaded due to constraints, but overall performance remains questionable. RAPL behavior at such low power appears unreliable unless the hardware supports it. If this holds true, disabling other states and relying solely on C1 might still yield some benefit for latency. Interestingly, a quick latency test in C0 with normal settings showed better results than expected, though likely due to minor inaccuracies or unmet power limits.
It's a fairly specialized subject, and I haven't encountered many discussions about it. It seems like few people have explored it deeply, which might explain the lack of widespread information. Personally, my experience with the 5950X showed that disabling software C-States prevented it from rebooting when idle. I wasn't sure how that mechanism functioned, especially since Ryzen Master still indicated the cores entered sleep mode. While I grasp the general idea of power gating, the full picture—especially in laptops for maximum efficiency—feels complex and not fully understood to me. There are likely many other considerations, like WiFi power management, that play a role.
C-States aren't mysterious powers. They play a key role in CPU speed. I've seen this often—having to post late at night because the rest of the community is still asleep. I usually write around 3 a.m. with my sleeplessness. Maybe I should have sent a quick message to @Eigenvektor and @Kilrah. It's just a small ping, barely noticeable.
Stay aligned with European timing, that’s what I need to do. I seem unfamiliar with C-states as well. Keep this in mind: sensor and software readings aren’t always accurate or very precise, particularly when changes happen rapidly. The energy saved could be just a few milliwatt-hours per millisecond. The system probably can’t monitor them effectively. The real benefit usually shows up only after longer periods, like extending a laptop’s battery by ten minutes. I also recall reading that data taken while sleeping might be less trustworthy, since parts of the device are inactive then.
It seems like you're trying to figure out how to improve performance while saving battery. The situation involves using a laptop with maximum settings but wanting to reduce latency by disabling C-states. You might want to check your device's settings or manufacturer tools to see the current C-state and its impact on speed.