- Joined
- Apr 12, 2021
- Messages
- 900
- Motherboard
- Asus z590 ROG Maximus XIII Hero
- CPU
- i9-11900K
- Graphics
- RX 6600 XT
- Mac
- Classic Mac
- Mobile Phone
tl;dr:
Cinebench on a desktop SMBIOS is not a sensible way to go looking for the advantages of CPUFriend tuning.
Explanation:
A compute benchmark is a work vs time measurement, which is strongly correlated with power, where in most systems, computer or otherwise, there's a sweet spot for efficiency: this means the least energy costly rate of doing work. IOW whether you are doing work fast or slow, the work may get done, but doing very fast or very slowly is less efficient than in a middle (energy optimal) range. How much less efficient and how much more costly and whether you care about the cost are separable matters. There's also a formula for as to whether doing work rapidly actually helps a device be more useful: doing work slower or faster than needed can matter to overall performance / power trade-offs.
For Apple computers, they offer a range of products which all do similar work, but the mobile ones are more power constrained, due to battery and thermal packing, and so are tuned to run more efficiently to optimize energy balance (which may or may not mean more slowly depending on how you measure work rates). This means that a different formula is used for mobile than for desktop. In turn, even desktops may be tuned to fit the physcal package: for example a iMac20,2 i9-109xx is not max-boosted and doesn't hit the same multicore benchmarks as a same-gen overclocked gaming PC because running the CPU all out next to a fat GPU means getting 400W thermal quietly through those slim iMac cases.
So this all about physics. It's a big balancing act, with different devices having different overall energy profiles to express different design tradeoffs. Think of it like tuning the fuel-air-catalytic balance in a car engine design, where the design refinement helps meet aggregate performance goals for efficiency, emissions, heat and noise. Apple works hard at these tradeoffs but a hacker may tear out the exhaust or put in another EMU chip and the car still rolls making just now it farts flame while melting the polar ice caps. (Cool story bro!)
One of the system energy profile factors is how fast the CPU clock is ramped up / down in response to the workload. This is under OS control in the same way as system sleep / wake etc but a finer granularity. Data called a "power vector" is tailored to the device and expresses a simple curve for how the CPU is marshaled through various power states by a higher-level heuristic coded into the OS energy manager.
CPUFriend lets helps you obtain and manipulate the power vectors from actual Macs and apply them to hacks, so macOS gets access to max the CPU can do. Do you care? Maybe not.
If you are just running Cinebench, you are not likely to see the advantage of a tuned power vector, unless a (mis)configuration has capped top speed because your benchmark is just a pedal-to-metal assessment. The 1% or so change in score could indicate the minor effect of a power vector shift on all-out performance, or it could be just noise in the testing process, such as influence of background system activity, etc.
I think Dortania guide covers this as something that prolly is ok out of the box these days for most hackintosh.
Laptop users have more to gain from CPUFriend than desktop.
CaseySJ may have some Alder Lake specific insights, as 12th gen is the first to overtly separate P/E core which all about these energy tradeoffs, and AppleSi went down this road ahead of Intel. (It wouldn't surprise me to learn that Apple got Intel heading that way as it's the sort of compromise you'd expect from engineering with a lot of experience in mobile, like, you know... the iPhone.
Hth'd and didn't put you to sleep.
