The Perfect Customac-Pro: X99-A II, i7-6950X, 128GB G.Skill TridentZ, Aorus GTX 1080 TI Xtreme

[kgp]

I did exactly as expected - entered the patches, entered 2 patches specific to Haswell-E, but after that the boot regularly hangs. The solution to the hang was to just remove those patches and -xcpm flag.

Still, it seems that it works out of the box because my CPU is jumping with the clock from 1,7GHz to 4.0GHz as expected, at least Intel Power Gadget says so.
View attachment 249120
The sleep, of course, doesn't work (auto-starts immediately), but I presume that's expected.
Also, the sysctl machdep.xcpm.mode returns 1 as expected:
View attachment 249123
I presume 5960X xcpm works without the flag.
This is my Clover config (notice flags being turned off):
View attachment 249124
This is all I have and... seems to work. Unfortunately, Pike is not to be bothered now, but he would probably be able to explain more thoroughly.
View attachment 249125
Notice two flags not being in your guide, but present in your downloadable config.

Did you also verify 9 a),b),c) and 12)? The -xcpm flag is obsolete. XCPM is loading by default but needs to be properly configured and subsequently works unless you break it with stuff like Nullcpupowermanagment. I explicitly stated in my guide that I did not use the "xcpm_bootstrap" KernelToPatch entry either, as it broke XCPM in my case. However, the "xcpm_cpuid_set_info" KernelToPatch entry is required to successfully pass 9), 10) and 12) and to properly inject the frequency vector . This is the only "KernelToPatch" entry different from Broadwell-E in your case. Does AppleIntelInfo.kext in your case reveal something similar as depicted in my post? Also study the frequency graph of the Intel power gadget under CPU load, e.g. while running geekbench. Frequencies in the Intel Power Gadget also jump up and down when using the Intel CPU Power Management. This does not mean that XCPM is properly working in your case.

 

[toncij]

A new not yet resolved system installation error, when performing a fresh Sierra 10.12.4 install on a virgin system disk? I did not face any issue when upgrading from 10.12.3 to 10.12.4.... However, it seems to have its solution, which is a second re-installation run. Strange and suspicious in any case...

Well it was a strange occurrence. My initial problems (stuck at 2nd pci init) were gone when I removed the Asus Xonar card and this install problem was not there the second time.

 

[dogmaphobic]

Then, although I created the RAID with the BIOS of my former X79 Gigabyte mobo

You could not have. You used DiskUtility to create it.

My question now is, how disk utility is able to detect three independent disks as a RAID system when the RAID system following your logic is subsequently created by disk utility?

DiskUtility detects that set as a RAID volume because it created it. You can move it to another Mac and it would still be detected as a RAID volume made out of 3 disks. As long as you keep the 3 disks together, Mac OS will recognize it as a RAID set.

Moreover, as the SATA ports 1-6 are set to AHCI in the X99-A II BIOS?

Yes, exactly. If it was a RAID created in BIOS, they would have to be set to RAID instead of AHCI.

P.S. It goes without saying that a RAID 0 is only for transient data and never for storage.

 

[laurentfluttert]

I've installed Sierra 10.12.4 too and had the error at the 6th second being about corrupted install. What I did is I just re-ran the installation and it worked flawlessly; rebooted once, continued installation for 1 minute and that was it. Just re-run it.

I'll try tomorrow, see what happens.

 

[kgp]

You could not have. You used DiskUtility to create it.

Wrong, I used the BIOS of the x79 mobo to create it and the three disks where configured as RAID-0 in the mobo BIOS. I just used Disk Utility to format the RAID-0 compound with NFS.

DiskUtility detects that set as a RAID volume because it created it. You can move it to another Mac and it would still be detected as a RAID volume made out of 3 disks. As long as you keep the 3 disks together, Mac OS will recognize it as a RAID set.

Allright..

Yes, exactly. If it was a RAID created in BIOS, they would have to be set to RAID instead of AHCI.

Exactly and indeed they have been defined as RAID-0 in the x79 mobo BIOS.

P.S. It goes without saying that a RAID 0 is only for transient data and never for storage.

Transient data stored for more than 5 years? Isn't a RAID-0 compound by definition something transient as it does not prevent data-loss in case of subsequent disk errors of any of the disks forming the compound?

 

[toncij]

Did you also verify 9 a),b),c) and 12)? The -xcpm flag is obsolete. XCPM is loading by default but needs to be properly configured and subsequently works unless you break it with stuff like Nullcpupowermanagment. I explicitly stated in my guide that I did not use the "xcpm_bootstrap" KernelToPatch entry either, as it broke XCPM in my case. However, the "xcpm_cpuid_set_info" KernelToPatch entry is required to successfully pass 9), 10) and 12) and to properly inject the frequency vector . This is the only "KernelToPatch" entry different from Broadwell-E in your case. Does AppleIntelInfo.kext in your case reveal something similar as depicted in my post? Also study the frequency graph of the Intel power gadget under CPU load, e.g. while running geekbench. Frequencies in the Intel Power Gadget also jump up and down when using the Intel CPU Power Management. This does not mean that XCPM is properly working in your case.

I see.. Hmm. So, this is the result for 9-12:

This is the log from the kext: cpu.txt

Spot anything wrong?

 

[kgp]

I see.. Hmm. So, this is the result for 9-12:
View attachment 249130
This is the log from the kext: cpu.txt

Spot anything wrong?

Looks pretty fine. Although you miss most of the C and P states! What about 9 a)? XCPM is up in any case. I would try to implement the proper "xcpm_cpuid_set_info" KernelToPatch. Crosscheck the values for Haswell-E with the corresponding table in https://pikeralpha.wordpress.com/2016/07/26/xcpm-for-unsupported-processor/.

In any case, you already managed and successfully passed most of the different steps and your XCPM configuration seems nearly complete already. Good job, man!

 

[toncij]

Looks pretty fine. Although you miss most of the C and P states! What about 9 a)? XCPM is up in any case. I would try to implement the proper "xcpm_cpuid_set_info" KernelToPatch. Crosscheck the values for Haswell-E with the corresponding table in https://pikeralpha.wordpress.com/2016/07/26/xcpm-for-unsupported-processor/.

In any case, you already managed and successfully passed most of the different steps and your XCPM configuration seems nearly complete already. Good job, man!

It's trivial, really, blind usage of ready-made stuff, not really a job of any kind worth mentioning. I've just spent like maybe an hour reading on how it functions, still, I'm unfamiliar with how Darwin works with CPUs and thus not really sure why the log lists multiple sets of multipliers for C and P states, let alone why it frequency and power limit values are so off the charts. C MSR ratios seem wrong too, not sure what applies for Haswells...

 

[AMonney]

Hello @kgp, I just want to thank you a lot for your thread/build I finally got time during Easter to build and install OS Sierra 10.12.4 ! And it works perfectly.

• Mobo Asus X99 Deluxe II
• 64GB G.Skrill Trident Z
  
• i7 6850K
• ZOTAC GTX 980 Ti (ll have to sell this one to get a brand new 1080)

I only get trouble right now to get my NVMe Samsung 960 EVO running as internal drive. Everytime I try what you mentionned and I believe what @beep mentionned in the page 25. When I m booting, I got Kernel panic at the apple logo and nothing else. I have to disable the ACPI patched in order to reboot normally .

It might be because I m not using the good adress for the M.2 port and PCI M.2 (took the one you mentionned and found the second one on an other thread) wasn't able to install windows to get the adress.

I ll try to work on this later on the week.

Have a great night

 

[AMonney]

Ok, I finally got to achieve it without windows ! but it was kind of annoying way to get the right acpi-path. So I used the nickwoodhams technic that allows me to see my two NVMe Samsung 960 EVO but as external drives (so not optimal). Once I got that I run the IORegistryExplorer until I find my two NVMe disk and see the real accurate acpi-path for my motherboard x99 Deluxe II which are :

_SB/PCI0@0/BR2D@20003/H000@0 for the M.2 socket
so :
_SB.PCI0.BR2D.H000

_SB/PCI0@0/BR3A@30000/H000@0 for the additional PCIe card
so :
_SB.PCI0.BR3A.H000

I then edit the file provided for 2 NVMe SSD which give me this :

DefinitionBlock ("", "SSDT", 2, "hack", "NVMe-Pcc", 0x00000000)
{
    External (_SB_.PCI0.BR2D.H000, DeviceObj)    // (from opcode)
    External (_SB_.PCI0.BR3A.H000, DeviceObj)    // (from opcode)

    Method (_SB.PCI0.BR2D.H000._DSM, 4, NotSerialized)  // _DSM: Device-Specific Method
    {
        If (LNot (Arg2))
        {
            Return (Buffer (One)
            {
                 0x03                                        
            })
        }

        Return (Package (0x04)
        {
            "class-code",
            Buffer (0x04)
            {
                 0xFF, 0x08, 0x01, 0x00                      
            },

            "built-in",
            Buffer (One)
            {
                 0x00                                        
            }
        })
    }

    Method (_SB.PCI0.BR3A.H000._DSM, 4, NotSerialized)  // _DSM: Device-Specific Method
    {
        If (LNot (Arg2))
        {
            Return (Buffer (One)
            {
                 0x03                                        
            })
        }

        Return (Package (0x04)
        {
            "class-code",
            Buffer (0x04)
            {
                 0xFF, 0x08, 0x01, 0x00                      
            },

            "built-in",
            Buffer (One)
            {
                 0x00                                        
            }
        })
    }
}

I reboot, and... failed got the white interdiction logo just after apple's logo. Then back to clover, retry and same results.
reboot 2 times and it finally loads So I got no explanation why. But if you run in same trouble as me this trick can maybe helps you out. Thank you once again @kgp, because without your comment of rebooting several times in order to make it work I would have try and retry new ".aml" files without success.

After migrating my OS install on the NVMe I performed a quick test of speed this is a real Mac Pro haha



Cheers