Mavericks: Native CPU/IGPU Power Management

[CybermaniaK]

It is likely you have your system so busy producing graphs, that it will spend quite a bit of time with a non-idle multiplier. If AICPUPMI is showing that you're reaching 8, then you're reaching 8.

CPU p-states are one of those things difficult to observe without effect.

Could be.

But I have another question then why in AICPUPMI is not registering something between 8 and 35 ? When clearly Power Gadget have those states registered..

 

[RehabMan]

Could be.

But I have another question then why in AICPUPMI is not registering something between 8 and 35 ? When clearly Power Gadget have those states registered..

The tool could be looking at different things or it is just the presentation.

We do know that you're using iMac14,x. And we know that iMac profiles do not use states between idle and nominal... not on real Macs and not on hacks either. Your CPU is nominal at 35 and idle at 8.

 

[CybermaniaK]

The tool could be looking at different things or it is just the presentation.

We do know that you're using iMac14,x. And we know that iMac profiles do not use states between idle and nominal... not on real Macs and not on hacks either. Your CPU is nominal at 35 and idle at 8.

Cool, so then according to my files provided on post #117 we can say that Power Management is working properly ?

 

[RehabMan]

Cool, so then according to my files provided on post #117 we can say that Power Management is working properly ?

I think so. Continued observation will reveal animals out of normal cloud formations.

 

[toleda]

Cool, so then according to my files provided on post #117 we can say that Power Management is working properly ?

I do not conclude PM is working properly on your machine nor do I know what is wrong.. AICPUPMI and IPG info are significantly different from the same info in Post #1 for the same processor. A clean install may be necessary.

 

[RehabMan]

I do not conclude PM is working properly on your machine nor do I know what is wrong.. AICPUPMI and IPG info are significantly different from the same info in Post #1 for the same processor. A clean install may be necessary.

From ioreg, what do you see is wrong with CPU PM?

 

[RehabMan]

What changed between Mountain Lion and Mavericks power management? The existing processor frequency/power state reporting tools (i.e., MSRDumper, HWMonitor) stopped reporting power states between idle and max non turbo.

...

The highlighted section is probably not accurate.

I recorded p-state 17 on a Haswell HP Envy laptop with an i7-4700mq (nominal is 24) running 10.9.1.

See here where I reported it, along with the details: http://www.insanelymac.com/forum/to...-locked-msrs-and-hp-envy-15-j063cl-i7-4700mq/

It could be that something has changed in 10.9.2 where the system uses less p-states. But to make this statement generally about 'Mavericks' is definitely not accurate.

In addition, these tools simply read the CPU MSRs directly. So I don't think it is that the tool is not reporting it, or that somehow Mavericks has disabled reading the MSRs, but rather more likely that OS X is not using those states (or perhaps not as often) as it once was.

 

[CybermaniaK]

I do not conclude PM is working properly on your machine nor do I know what is wrong.. AICPUPMI and IPG info are significantly different from the same info in Post #1 for the same processor. A clean install may be necessary.

Well this is a clean install, because I was trying to update my 10.9.0 to 10.9.2 but I decide to make a clean install.

Now I did OC to 4.4ghz and I found something weird.

HWMonitor completely stop reporting it's stuck. see this screenshot
https://farm8.staticflickr.com/7407/13979604820_77561d235e_o.png

And now AICPUMPI it's reporting only AICPUPMI: CPU P-States [ 8 35 44 ] which sound "normal" because my OC was on the turbo multiplier for the four cores. And obviously I had to generate a new SSDT with the -turbo flag and it's loading. I can see X86PlatformPlugin and AGPMController on ioreg...

So now i'm very confused..

 

[toleda]

From ioreg, what do you see is wrong with CPU PM?

The AICPUPMI report is incomplete: no registers, missing pstates, etc.
Attached previously

Last login: Sun May 11 15:39:29 on console
...$ cat /var/log/system.log | grep "AICPUPMI:"
May 11 15:37:27 localhost kernel[0]: AICPUPMI: CPU P-States [ 35 37 ]
May 11 15:37:30 ... kernel[0]: AICPUPMI: CPU P-States [ 35 37 39 ]
May 11 15:37:30 ... kernel[0]: AICPUPMI: CPU P-States [ 8 35 37 39 ]
May 11 15:37:31 ... kernel[0]: AICPUPMI: CPU P-States [ 8 35 37 38 39 ]

 

[toleda]

The highlighted section is probably not accurate.

The tools are accurate for what they are reporting. The issue is the resolution of the tools. HWMonitor is 1 sec or more. IGP default is 200 ms; reduce reporting interval (increase time) and IGP looks like HWMonitor. Run the IGP command line tool and see the same measurements as HWMonitor. The formula; for more pstates, read msr more often.

The conclusion that imac14 PM is not working because HWMonitor did not show P states between lowest power and highest non turbo is not correct. Pstate 17 with macbookpro11 is simply a HWMonitor reporting anomaly. imac14/macbookpro11 PM always worked in both native and non native (with the correct CPU ssdtPRGen ssdt) configurations. As Post #1 says, HWMonitor cannot be used to make correct power management decisions.

All test results in Post #1 are 10.9.2/C64/C1021. No changes noted in subsequent testing with developer releases.