KernelAndKextPatches 10.13x,10.14.x,10.15.x X99/X299

[williamhx]

31 minutes ago, latze said:

 

Yes, now enable them one by one, boot and see what happens. You changed the fakeCPUID to 0x040674 also right? 

I suggest you start with the one that starts with 42060000  DC330600

 

 

Yes, i use 0x040674.

I tried them one by one, and also many combinations, looks like the bootstrap patch will always keeps my CPU at 4.00GHz.

[latze]

1 minute ago, williamhx said:

Yes, i use 0x040674.

I tried them one by one, and also many combinations, looks like the bootstrap patch will always keeps my CPU at 4.00GHz.

 

Don't enable the bootstrap patch, for your CPU it should be enough just to set CPUID to 0x040674

[vector sigma]

Guys, can you try this Clover for xcpm (or compile the last from source) and see if still require the CPU fake id or it works with out it? New patch location in 10.15.4 + for the _xcpm_bootstrap function. Thanks in advance.

 

P.S. I mean without additional patch related to xcpm, just KernelXCPM=true

Edited April 27, 2020 by vector sigma

[latze]

13 hours ago, vector sigma said:

Guys, can you try this Clover for xcpm (or compile the last from source) and see if still require the CPU fake id or it works with out it? New patch location in 10.15.4 + for the _xcpm_bootstrap function. Thanks in advance.

 

P.S. I mean without additional patch related to xcpm, just KernelXCPM=true

 

@vector sigma it works without patches and any cpu type / fakeid set, the only patch that I still have to enable is the performance one otherwise the cpu is not used at maximum capacity although xcpm is enabled without it.

[vector sigma]

4 hours ago, latze said:

 

@vector sigma it works without patches and any cpu type / fakeid set, the only patch that I still have to enable is the performance one otherwise the cpu is not used at maximum capacity although xcpm is enabled without it.

Thanks for the confirmation. About performances I see if I can do something but I need to study it. Good to know it doesn't require the cpu fake id as suspected. 

[latze]

2 minutes ago, vector sigma said:

Thanks for the confirmation. About performances I see if I can do something but I need to study it. Good to know it doesn't require the cpu fake id as suspected. 

 

Yes, I was thinking about the issue and how it can be fixed properly also, because I doubt real macs need this kind of patch to run at peak performance. What the performance patch does basically is to force CPU to high frequency. I looked through the kernel disassembled code also regarding this but my power management knowledge is not so advanced at the moment.

[lucke]

On 4/27/2020 at 8:29 AM, latze said:

 

It is at 2550 because the performance patch ( 41C1E608 B800FF00 0031D290 90 ) sets it like that. See the FF in the code, that sets the frequency. To set it to your max frequency ( 4300 ) change it to 2B so the replace code will be 41C1E608 B8002B00 0031D290 90

 

Thanks latze, I made the change, now the frequency is expressed correctly. I noticed something that perhaps I had never considered ... the pset performance patch maintains the required work frequency: corereq, almost always at the maximum, but by deactivating the various patches to test the xcpm according to the new vector clover project sigma, overall performance decreases by 30/40%, however the core frequencies generally work more optimally. Where the problem could be ...? In the pstate performance patch? if you have any idea?

[latze]

16 hours ago, lucke said:

Thanks latze, I made the change, now the frequency is expressed correctly. I noticed something that perhaps I had never considered ... the pset performance patch maintains the required work frequency: corereq, almost always at the maximum, but by deactivating the various patches to test the xcpm according to the new vector clover project sigma, overall performance decreases by 30/40%, however the core frequencies generally work more optimally. Where the problem could be ...? In the pstate performance patch? if you have any idea?

 

The performance patch changes the frequency requested to be always maximum by changing the value sent to IA32_PERF_CTL MSR (0x199). Why the requested value is not always optimum that is a question I would like to see the answer too also  but it is beyond my knowledge at the moment. What I noticed is that although the frequency stay usually higher ( not always to the maximum though ) the reported power usage on my CPU is pretty much the same with or without the performance patch. So until someone comes with a better solutions I guess we have to stick with it. 

[vector sigma]

17 hours ago, lucke said:

however the core frequencies generally work more optimally

I would call it "efficency" . For a laptop it's what matter, but I think desktop's is not so different as cpu cores can't run at max frequency all the time together by design.

[latze]

16 hours ago, vector sigma said:

I would call it "efficency" . For a laptop it's what matter, but I think desktop's is not so different as cpu cores can't run at max frequency all the time together by design.

 

It doesn't work optimal though, if you check some videos with benchmark running on mac pro with intel power gadget running you'll see that under load the req stays at maximum. For us it doesn't happen so there must be something missing that affects that. 

[vector sigma]

4 hours ago, latze said:

 

It doesn't work optimal though, if you check some videos with benchmark running on mac pro with intel power gadget running you'll see that under load the req stays at maximum. For us it doesn't happen so there must be something missing that affects that. 

please make a screenshot with HWMonitorSMC2.app in full screen with these settings:

 

to enable charts right click on each cpu sensor and press on .

launch a benchmark and make the screen shot.

 

P.S. after enabling intel power gadget and pmu the app needs a restart.

Edited April 30, 2020 by vector sigma

[lucke]

8 hours ago, vector sigma said:

please make a screenshot with HWMonitorSMC2.app in full screen with these settings:

 

to enable charts right click on each cpu sensor and press on .

launch a benchmark and make the screen shot.

 

P.S. after enabling intel power gadget and pmu the app needs a restart.

I have decided to give my help...

 

[vector sigma]

Thanks @lucke, I realized only now you are Italinan, ciao !

If you look at the "Frequency core avg" (media) and requested one by OS, Frequency core req ("riq" as requested in Intalian) both charts are very similar indicating that the CPU really follow the request of the OS. Also I see cores do their job reaching 4.3 GHz from 1.1 GHz. What is wrong?

[nmano]

#6 This patch for X99(Haswell-E)

Comment    String  <->xcpm_bootstrap HASwell-E (c) Pike R. Alpha Modify By N.Mano
Disabled   Boolean <-> No
Find       Data<->   80C3C480 FB42
MatchOS    String <-> 10.15.x
Replace    Data<->   80C3C180 FB42

#7 This patch for X299(Skylake-X/Cascade Lake-X)


Comment    String  <->xcpm_bootstrap Skylake-X/Cascade Lake-X (c) Pike R. Alpha Modify By N.Mano
Disabled   Boolean <-> No
Find       Data<->   80C3C480 FB42
MatchOS    String <-> 10.15.x
Replace    Data<->   80C3BD80 FB42

#8This patch for X99(Broadwell-E)

Comment    String  <->xcpm_bootstrap Broadwell-E (c) Pike R. Alpha Modify By N.Mano
Disabled   Boolean <-> No
Find       Data<->   80C3C480 FB42
MatchOS    String <-> 10.15.x
Replace    Data<->   80C3BC80 FB42

#9 This patch for (IvyBridge)

Comment    String  <->xcpm_bootstrap IvyBridge (c) Pike R. Alpha Modify By N.Mano
Disabled   Boolean <-> No
Find       Data<->   80C3C480 FB42
MatchOS    String <-> 10.15.x
Replace    Data<->   80C3C680 FB42

10.15.4 & 10.15.5 xcpm_bootstrap

[lucke]

3 hours ago, vector sigma said:

Thanks @lucke, I realized only now you are Italinan, ciao !

If you look at the "Frequency core avg" (media) and requested one by OS, Frequency core req ("riq" as requested in Intalian) both charts are very similar indicating that the CPU really follow the request of the OS. Also I see cores do their job reaching 4.3 GHz from 1.1 GHz. What is wrong?

@vector sigma Ciao! I think there is nothing wrong, at least in the management of the frequencies and therefore of the processor load, now the visualization and the work of the single cores as the frequencies is correctly indicated at least in my case ... But I believe that "wrong" there is a loss of performance: the first image is with all the patches active before the update to "Clover xcpm":

and this after (now) with or without pstate performance patch:

It is clear that there is a loss of about 35% of the performances...

Edited April 30, 2020 by lucke

[vector sigma]

On 4/30/2020 at 1:11 PM, vector sigma said:

It doesn't work optimal though, if you check some videos with benchmark running on mac pro with intel power gadget running you'll see that under load the req stays at maximum. For us it doesn't happen so there must be something missing that affects that. 

@lucke I was referring to the above statement, which clearly is not the case . 

[lucke]

36 minutes ago, vector sigma said:

@lucke I was referring to the above statement, which clearly is not the case . 

Ah ok..., sorry!

[jlcdgd]

Hi Lucke,vector sigma

 

Clover 5115 makes everything better, much better, but not perfect.

Adding the Brumbauer kext that I use for Mojave, geekbench are better and protools in my case works better. I don´t know exactly  what is the kext patching, but it works.

Please Lucke, double check it. I am not using any other patch. And I have tried a lot of combinations.

 

Maybe vector sigma, can analize what is the kext doing.

Without kext

With kext

 

It is a very old kext by the way, still working...

Please let me know if it works for you Lucke...

 

 

 

PMDrvr.kext.zip

2 hours ago, vector sigma said:

@lucke I was referring to the above statement, which clearly is not the case . 

 

 

 

 

[jlcdgd]

2 hours ago, lucke said:

Ah ok..., sorry!

 

3 hours ago, vector sigma said:

@lucke I was referring to the above statement, which clearly is not the case . 

 

Check my previous post about Brumbauer

 

[lucke]

26 minutes ago, jlcdgd said:

 

 

Check my previous post about Brumbauer

 

@jlcdgd I'll test and report...

[vector sigma]

31 minutes ago, jlcdgd said:

 

 

Check my previous post about Brumbauer

 

It access smrs, read them, modify them and contains a lot of code as a complete and dynamic power management for Haswell +. The kext is built as a sub class of AppleIntelCPUPowerManagement. Him or xcpm, not sure together can work. 

Edited May 1, 2020 by vector sigma

[jlcdgd]

1 minute ago, vector sigma said:

It access smrs, read them, modify them and contains a lot of code as a complete power management for Haswell +. The kext is built as a sub class of AppleIntelCPUPowerManagement. Him or xcpm, not sure together can work. 

They work together perfectly with xcpm and really smooth in my case...

Thanks vector

[vector sigma]

11 minutes ago, jlcdgd said:

They work together perfectly with xcpm and really smooth in my case...

Thanks vector

mmmh, what happen if you disable xcpm?  (if you can)

[jlcdgd]

In Clover? Or where....

8 minutes ago, vector sigma said:

mmmh, what happen if you disable xcpm?  (if you can)

 

 

[vector sigma]

11 minutes ago, jlcdgd said:

In Clover? Or where....

yes, Clover enables xcpm (for unsupported CPUs) only if you set KernelAndKextPatches->KernelXCPM = true. For supported CPUs it just will work.

If your is an unsupported one, KernelAndKextPatches->KernelXCPM = false (or just remove the key) will work as well, I guess. Otherwise you will have two programs that will interefere each other. 

 

P.S. you can disable it also in Clover GUI at boot (binaries patching menu). 

Edited May 1, 2020 by vector sigma