Qsync Issues RX580/Intel 4600 iGPU Maxed Always

[slipanot]

This has been an ongoing issue.
Im trying to get qsync working correctly using the headless IGPU connector method.
-I use the iMac 15,1 Smbios
-I have XCPM working with my CPU using the plugin-type=1 method through clover "PM shows enabled in IOReg"
-Using the PlatformID of 0x04120004 and also Inject Intel=true
-Whatevergreen and lilu in clover/kexts/other
-GFX0 to iGPU patch in clover
But for some reason my iGPU runs at 1.25Ghz always. "No AGPM"
-Ive tried the dAGPM.kext method and it didn't seem to make a difference either.

Now if I leave PlatformID blank and just use InjectIntel=True and let whatevergreen do the guess work My iGPU idles correctly and and seems to ramp up and down when needed. 0x0D220003 Shows as my platform ID with this method
system profiler shows 2 GPUs "iGPU built in and RX580 PCI"
So Should I just go with the 2nd option? I would like to get the "Correct" way working if possible.
Attached is my current debug and screenshots

 

[jaymonkey]

Now if I leave PlatformID blank and just use InjectIntel=True and let whatevergreen do the guess work My iGPU idles correctly and and seems to ramp up and down when needed. 0x0D220003 Shows as my platform ID with this method

@slipanot,

0x0D220003 does not match a know headless PlatformID.

It's always best to explicitly define the PlatformID as a Device Property when using WhatEverGreen ... very rarely is WEGs's default correct and almost never for a headless ID.

See this guide for more info ... there is a section dedicated to headless PlatformID's.

https://www.tonymacx86.com/threads/an-idiots-guide-to-lilu-and-its-plug-ins.260063/

Make sure you use the correct SMBIOS.

Cheers
Jay

 

[slipanot]

So far I've tried both Smbios iMac 14,2 and 15,1 and also have tried both headless platform IDs of 0x04120004 and 0x0412000B. Both seem to report correctly In IORegistryExplorer once using either of these IDs but no matter what my iGPU runs maxed out all the time. 1.25Ghz no matter what. No AGPM either. Only when I leave PlatformID blank and InjectIntel=True do I seem to get AGPM for my iGPU.

Also attaching my debug with platformID blank and just using InjectIntel=True

 

[jaymonkey]

So far I've tried both Smbios iMac 14,2 and 15,1 and also have tried both headless platform IDs of 0x04120004 and 0x0412000B.

@slipanot,

I have the same motherboard and CPU as you ... ASRock Z98 Extreme 6 with i7 4900K CPU, my dGPU is Vega 64.
I'm using iMac15,1 SMBIOS with 0x0412000B Headless PlatformID.

The side by side example of finding a Headless PlatformID with a recommended SMBIOS in the guide is actually from that system, QucikSync and dGPU are working together extremely well with this config.

As the guide states remove all IGPU patches/fixes in the Devices and Graphics sections of your config.plsit
So remove fake "0x0" for Fake IGPU ID in Devices section and "Inject Intel" from Graphics section of your config.plist.
Inject the headless PlatformID value 0x0412000B for "AAPL,ig-platform-id" as as a Device Property as per the guide.

I don't think the IGPU being pegged is an issue ... i've seen that behaviour on real iMacs with dual GPU's.

You are injecting all of your 3rd party kexts ... as the guide states injected kexts live outside of the MacOS protected memory space and are not cached, injection of that many kexts can make for an unstable system. Install all 3rd party kexts in /L/E as per the guides How to Install Section, keep copy of FakeSMC in /EFI/Clover/Kexts/Other and delete all the others.

Set the Clover Option "Inject Kexts" in System Parameters to "Detect" and set the boot argument "kext-dev-mode=1"

There should be no need for you to have a static patched DSDT in /EFI/Clover/ACPI/Patched ... use SSDT Hot patches rather than a static patched DSDT, any changes you make in the BIOS will causes miss matches to the static patched DSDT. The ASRock Z97 Extreme 6 has an extremely compatible native DSDT. See Rehabmans HotPatch repo for Hot Patch SSDT's, the only really important one you need is SSDT-XOSI.aml to spoof Darwin as Windows 10 and of course SSDT-UIAC to define usb ports correctly which you already have (although i do not see the USBInjectAll.kext so you still have many unused USB ports defined in ioreg)

Once the system is configured correct there should be no need for the NoVPAjpeg.kext and you should remove it, the method/workaround it uses is not a great solution.

Cheers
Jay

 

[slipanot]

I don't think the IGPU being pegged is an issue ... i've seen that behaviour on real iMacs with dual GPU's.

QuicSync IS working when then iGPU is maxed out. But is having the iGPU pegged all the time an issue or say normal? Does yours stay maxed on all the time when using this setup?

 

[shuhung]

Reading bug on Power Gadget 3.5.x, result should divided by ~3, use Intel Power Gadget 3.0.3 to verify GT frequency.

For Haswell IQSV with connector-less ig-platform-id, GT frequency should constant at 0.35 GHz.

 

[jaymonkey]

QuicSync IS working when then iGPU is maxed out. But is having the iGPU pegged all the time an issue or say normal? Does yours stay maxed on all the time when using this setup?

Yes .. its the same on my system, although i do have AGPM for the IGPU in ioreg. Like i said i've seen this behaviour on real iMac's ... also quite a few iMac's keep the CPU at the base clock and only use turbo c states, they do of course use the full range of cpu p states ....

Cheers
Jay

 

[shuhung]

Yes .. its the same on my system, although i do have AGPM for the IGPU in ioreg.

Have you verify with old Intel Power Gadget, clearly software bug, see my GFX frequency as high as 3.02 GHz.

 

[jaymonkey]

Have you verify with old Intel Power Gadget, clearly software bug, see my GFX frequency as high as 3.02 GHz

@shuhung,

I've never seen Power Gadget exhibit that behaviour ... in my case (like @slipanot) the IGPU is pegged at 1.25 which is the top C state for HD4600 IGPU.

Like i already explained many real iMacs keep the CPU and IGPU at high C states and throttle the turbo c states only, these systems tend manage CPU power with just the P states, which is a much better method for desktop systems, it's only really necessary to alter c states through a large frequency range on laptops to help conserve battery power.

A lot depends on which SMBIOS you use ...

Cheers
Jay

 

[shuhung]

I've never seen Power Gadget exhibit that behaviour

Reading bug from Intel Power Gadget 3.5.0 up, results should divided by 3.