You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The advice to use iMac20,x SMBIOS with frequency vectors results in Apple Comet Lake frequency vectors being used. There you would find QoS values that are heavily limited to at least somewhat fit the TDP of Apple iMac cases. While we mostly tested that on Rocket Lake (11700k) with @vandroiy2013, the results with Comet Lake should be the same. The performance loss due to such a configuration on a fully unlocked setup results in up to 50% performance loss under synthetic and real-world tasks.
The solution is to inject proper frequency vectors as I explained on AppleLife. The supplied file should be useable on Comet Lake, Rocket Lake, and Alder Lake setups. Perhaps it will also be useable on older generations, but I'd recommend testing these through trial and error.
The easy way to determine whether you have performance drains is to disable SSDT-PLUG and run some tasks or benchmarks. You will see noticeably worse results if your system is affected.
The relevant values in question look like this. The first screenshot is the iMac20,1 QoS for the kernel and the second screenshot is the fixed version, based on the Haswell QoS we made some time with @Andrey1970AppleLife. As you can see, Comet Lake values are really badly capped half the way.
I should add that disabling hwp (on → off) caused noticeable temperature decrease during idle times by around 5 degrees (from 40-41° to 36°). While it may be useful for cooling, we expect that it will also decrease overall system responsiveness.
I haven't seen this on Comet Lake(10900k) with imac20,2. I tested cinebench R23 with stock ssdt-plug and no cpufriend, cpufriend with a generated vector and cpufriend with the vectors from Apple Life. All 3 tests scored similarly and were within run to run variation, technically sock ssdt-plug was the best result and all results are similar to R23 runs in windows.
How does one check the QOS kernel like in the screen screenshot?
I understand binary templates are used with 010 editor. I'm not really sure which file to run the template on. I tried /s/l/e/IOPlatformPluginFamily.kext/Contents/Plugins/X86PlatformPlugin.kext/Contents/MacOS/X86PlatformPlugin but the template doesn't return any values. I don't want to waste your time on this since I'm not seeing the issue but will certainly help collecting any data or running tests if you want.
I have a i9 10850K using iMac20,2 and I created my CPUFriendDataProvider.kext with CPUFriendFriend.
The CPU Power management seems to be working fine. I am having a hard time trying to follow the argumentation here. But I will run some tests with kext attached to the first post and see if it makes any difference.