Broken Power Management in 10.14.5 Beta?

[shrieken213]

While making some final touches to an SMBIOS switch I noticed strange behavior with the CPU clocks.

 

I'm moving from iMac14,2 to 15,1. Build is 4790K (Overclocked 4.6Ghz), Vega 56, 32GB RAM. I'm on the beta release because of a bug in 10.14.4 that doesn't spin down the Vega fans after they spin up under load.

 

1. Either Intel Power Gadget or power management is broken. According to IPG the clocks almost never fall below the nominal 4 Ghz. At the same time, HWSensors show that the clocks do in fact fall to the idle 800 Mhz clocks.

AppleIntelInfo shows that all P- and C-states have been injected and observed by the kext. This leaves two possibilities: IPG is misreading the clocks, or there is in fact something that's artificially keeping the clocks at 4 Ghz while idle.

 

 

 

2. IPG utilization percentage is bugged. A ton. HWSensors also has a bug where clock speeds occasionally peak at 2^31 Hz (2147.48 THz).

 

3. The headless framebuffer for Intel HD 4600, 0x0b001204, gets the IGPU stuck at full clock. Only by fully enabling IGPU with 0x0300220d is IGPU stepping properly enabled. I don't know if a real iMac15,1, which uses the 0x0b001204 framebuffer natively, shows the same behavior.

 

I'm not very certain about the IGPU since I only recently tried out the headless framebuffer, but I do know for sure that IPG properly read idle clocks in 10.13 with the same config, more or less. I'm wondering whether this is a beta thing or other people have seen the same behavior elsewhere.

AppleIntelInfo.log

config.plist

Edited April 25, 2019 by shrieken213

[hardcorehenry]

1 hour ago, shrieken213 said:

While making some final touches to an SMBIOS switch I noticed strange behavior with the CPU clocks.

 

I'm moving from iMac14,2 to 15,1. Build is 4790K (Overclocked 4.6Ghz), Vega 56, 32GB RAM. I'm on the beta release because of a bug in 10.14.4 that doesn't spin down the Vega fans after they spin up under load.

 

1. Either Intel Power Gadget or power management is broken. According to IPG the clocks almost never fall below the nominal 4 Ghz. At the same time, HWSensors show that the clocks do in fact fall to the idle 800 Mhz clocks.

AppleIntelInfo shows that all P- and C-states have been injected and observed by the kext. This leaves two possibilities: IPG is misreading the clocks, or there is in fact something that's artificially keeping the clocks at 4 Ghz while idle.

 

 

 

2. IPG utilization percentage is bugged. A ton. HWSensors also has a bug where clock speeds occasionally peak at 2^31 Hz (2147.48 THz).

 

3. The headless framebuffer for Intel HD 4600, 0x0b001204, gets the IGPU stuck at full clock. Only by fully enabling IGPU with 0x0300220d is IGPU stepping properly enabled. I don't know if a real iMac15,1, which uses the 0x0b001204 framebuffer natively, shows the same behavior.

 

I'm not very certain about the IGPU since I only recently tried out the headless framebuffer, but I do know for sure that IPG properly read idle clocks in 10.13 with the same config, more or less. I'm wondering whether this is a beta thing or other people have seen the same behavior elsewhere.

AppleIntelInfo.log

config.plist

The same smbios a "little bit" different processor ig platform 0x0412000B. I read somewhere on russian forum that with headles framebuffer it is normal constant frequency for IGPU. I'm not 100% sure about it. Mine stuck as you see at 1100 GHz, whether I watch movies or quick sync it looks like this. Things are different when I look at HWMonitor, when no movies or quick sync CPU GFX show "-" when watching movies jumping 0,12-0,18 W. The highest utilization when converting. As for processor high frequencies pretty much simmilar situation, but in my case I observed it from first Betas. Every update IPG looked almost the same, with high freq until final releases which made normal again, I mean when idle 800-1200GHz. So I'm guessing final release of 10.14.5 will bring my processor back to normal. Frequencies thou improved when I reinstalled 10.14.4 and updated to Beta. Hope that helped a bit.

[shrieken213]

1 hour ago, hardcorehenry said:

The same smbios a "little bit" different processor ig platform 0x0412000B. I read somewhere on russian forum that with headles framebuffer it is normal constant frequency for IGPU. I'm not 100% sure about it. Mine stuck as you see at 1100 GHz, whether I watch movies or quick sync it looks like this. Things are different when I look at HWMonitor, when no movies or quick sync CPU GFX show "-" when watching movies jumping 0,12-0,18 W. The highest utilization when converting. As for processor high frequencies pretty much simmilar situation, but in my case I observed it from first Betas. Every update IPG looked almost the same, with high freq until final releases which made normal again, I mean when idle 800-1200GHz. So I'm guessing final release of 10.14.5 will bring my processor back to normal. Frequencies thou improved when I reinstalled 10.14.4 and updated to Beta. Hope that helped a bit.

 

Thanks. You're seeing exactly what I'm seeing with the IGPU. Idle or not your IGPU is at full clock with headless.

 

My screenshots were with 0x0300220d; the IGPU idles at 0.2 Ghz and HWSensors picks up 0.02~0.05W. It peaks at 10.74W under full load (LuxMark OpenCL stress test). Because the headless connector doesn't allow the IGPU to be used for anything other than HEVC and ME, I can't directly compare load power. Vega has hardware HEVC enabled, so playback's not touching the IGPU at all in either case for me.

 

This all sounds as with headless the IGPU is stuck at full clock but is either on or completely off, while when it's fully enabled it's stepping its clocks and voltages but always on. Which is also kind of weird since if the IGPU is off it shouldn't have a clock speed at all.

 

Were you running 10.14.4 or 10.14.5 when you took that screenshot? I see your CPU is stepping properly.

[hardcorehenry]

19 minutes ago, shrieken213 said:

 

Thanks. You're seeing exactly what I'm seeing with the IGPU. Idle or not your IGPU is at full clock with headless.

 

My screenshots were with 0x0300220d; the IGPU idles at 0.2 Ghz and HWSensors picks up 0.02~0.05W. It peaks at 10.74W under full load (LuxMark OpenCL stress test). Because the headless connector doesn't allow the IGPU to be used for anything other than HEVC and ME, I can't directly compare load power. Vega has hardware HEVC enabled, so playback's not touching the IGPU at all in either case for me.

 

This all sounds as with headless the IGPU is stuck at full clock but is either on or completely off, while when it's fully enabled it's stepping its clocks and voltages but always on. Which is also kind of weird since if the IGPU is off it shouldn't have a clock speed at all.

 

Were you running 10.14.4 or 10.14.5 when you took that screenshot? I see your CPU is stepping properly.

I'm on last 10.14.5 beta. Here is everything very well described. Use translator if you have to, I had to.

[shrieken213]

It seems there's something wrong with WindowServer. Upgraded to 10.14.5r3 just a few hours ago and the high CPU usage is still there. I'm assuming PM is working but the unnecessary load is keeping the clocks up.

 

This is also after reverting to 10.14.4 and reinstalling the beta. Go figure.

Edited April 25, 2019 by shrieken213

[hardcorehenry]

For me pretty mach the same, processor have sometime better or worse moments