The "Official" C612 (aka lets built the machine Apple never gave us)

[volition]

Happy to report some initial success @volition!

Same here, much thanks to you. Following your exact instructions, apart from adding info about the memory in Clover configurator, El Capitan 10.11.6 installed just fine. Maybe it's needed since system report only lists 8 slots, so only registers 32 of the 64 GB. I'll try with it and see. If not, I'll use the same slots you are (I guess I'm using 0, 1 (CPU1) and 9, 10 (CPU2) now). Attaching the EFI.zip here that worked for me.

All the RAM registers and the RX 480 works in Windows 10.

...Then I realized my mistake was that I was trying to emulate @metacollin's end result, where as what I should have been doing was trying to emulate his journey to get there...

Seems that was my mistake too. Trying to get OS X installed with native power management in one go. As well as trying to do it with Sierra 10.12 GM (because of the RX 480). (It probably doesn't matter but I was using Clover at v3726 instead of v3751 too.)

Looks like this base install should work with a wide array of CPU's, GPU's and RAM for 10.11.6. I tried the same with 10.12 GM, but couldn't get it to work. I'll try again and come back with some more info on that.

I am now currently where metacollin was in this post. A working el capitan install, no CPU power management, working nvidia web driver, geekbench 3 score of 55k (I assume bios tweaks may explain the difference between my 55k and his 62k, I reverted bios and used minimum changes). Full networking works out of the box. Currently USB 3,0 is disabled. Full step-by-step instructions below.

Same here, except lack of GPU, which in addition to crappier CPU's and only half the RAM, maybe explains much lower scores.

Integer Processor integer performance 1814 24715
Floating Point Processor floating point performance 1884 34702
Memory Memory performance 1495 1870
Geekbench 3.3.4 for Mac OS X x86 (64-bit)

It's nice to hear your results were more in line with @metacollin's, even without all the deeper BIOS changes (disabling virtualization, power management, etc), messing with ACPI tables and Clover settings (patches, CPUFakeID's).

Oh, and no onboard audio (seems it's in system report, but not selectable as core audio in system preferences).

Also suddenly had the Thunderbolt AiC working. I disabled it when installing 10.11.6, then enabled it (hotplugged, enabled and in legacy mode), but couldn't get it to work. Then found some threads on here which lead me to install the Intel drivers in Windows 10. Rebooting the machine with the device on (Universal Audio Apollo Quad), it still didn't work. Then rebooted again, but changed SMBIOS to MacPro5,1 in Clover, and booted into OS X. Suddenly it worked. I've rebooted a few times since then, with SMBIOS being reverted to MacPro6,1, and it's still working. System report (still) says there's no Thunderbolt drivers loaded, and the Apollo is recognized as a PCI device under audio in systems report (common, from what I understand), but it works.

Btw, under PCI in systems report, it says "There was an error while gathering PCI device information". Not sure if that's significant.

Skipping ahead, what's next? By reading between the posts I believe the next steps metacollins took was to generate the SSDT (by "just making stuff up" ). I believe this step needs to be done first BEFORE trying to use a fakecpuid of 0x0306F0/2 and removing NullCPUPowerManagement.kext from my build.

That's where I got seriously stuck. Using ssdtPRGen, there was no definition for these CPU's. I added that but didn't follow through after that. I'll try to do it, following this.

Also, would like to figure out why my windows geekbench is so low. Some possibilities: geekbench is reporting my memory weirdly "63.9 GB - 1 Mhz", and CPU-Z is reading the bus speed for the CPU is 100 Mhz when on mac os x reports it as 400 Mhz.

Both system report and Geekbench says 32.0 GB, 2400 MHz, DDR4 here in 10.11.6. Same with Windows 10 (task manager, CPU-Z, Speccy) except it reports all the RAM. I'll try Geekbench in Windows. The BIOS reports ~1867 MHz, which is max for these CPU's.

- Check OsxAptioFix2Drv-64 to CLOVER/kexts/10.11/ (not 100% OsxAptioFix2Drv-64 is required instead of recommended OsxAptioFixDrv-64).

Used OsXAptioFix2Drv-64. Where/why is OsxAptioFixDrv-64 recommended though?

- Check npci=0x2000 (fixed issue I experienced where it would hang after PCI configuration began)

Confirmed!

Once I got here, this got me past the +++++++++++++++++. I then experienced a hang on mbinit line, which prompted the following supermicro BIOS change.

For supermicro bios:
Under the BIOS on your machine, disable USB 3.0 (see this post)

Tried without too, so can confirm this as well.

...I then got the GPU working by...

Sounds great. All I can see is warped aspect ratio, lag and most of the screen flickering white each time I type a letter. I'll probably get a GTX unless Sierra installs. Looks like it'll take some time to get the RX 480 working in El Capitan.

Thanks again for adding a step-by-step. Saved me a lot of time and frustration!

 

[catswift]

@volition Nice, I'm glad that it helped.

As a follow up, I've come to the determination that getting native power management working will be critical for me. Without it I have two options:

Option #1: Use the the fake CPUID of 0x0306E4 (all other fake CPUID's kernel panic for me); however with this method my max geekbench score is 53k-ish, and I know that this machine is capable of much more.

Option #2: Use KernelCPU clover option; with this I get 67k-ish geekbench (not the best I think it could be), however audio playback become choppy and the OS X clock will drift (from my research this is possibly caused by wrong front-side bus speed).

Neither option is ideal, and I have the feeling that with working power management the CPUs will work at optimal levels and OS X will be able to manage its clock correctly. What I don't know is why the fake cuid used by @metacolin works for him and not for me. I did end up generating (correct?) ssdts for these cpus, but even with those the fake cpuid never worked. I ended up scrapping those and running with the ones clover generates automatically. I think I might try removing a CPU and doing one CPU at a time, or maybe try doing a Sierra install, or maybe try different SMBIOS combinations with those other fake cpuids. I do think getting on the right fake cpuid is the key though, just wish I knew what to tweak to get that to work.

[edit]
I should note I tried but never got the motherboard audio working, but that doesn't matter to me because USB audio devices do work. So a $5 USB-to-stereo plug solves hooking this up to my speakers and my USB headset already works fine.
[/edit]

 

[catswift]

Progress has been made. I tried a fresh install with sierra; I used latest clover and this time used OsxAptioFixDrv-64 (not "2"). I also found the clover kext patch (see attachment). With this new usb installer in place I set the FakeCPUID to 0x0306F0 (the last number is the stepping and 0 matches my cpu stepping), and lo and behold the sierra installer booted right up! So I finally passed the hurdle of not being able to use the same FakeCUID as metacollin.

Ok, now I installed sierra (same GPU fix worked), I turned my attention to native power management. I generated a ssdt using pike's script, I removed null cpu kext, and I made sure to use -xcpm_ipi boot args. First time through it kernel panics, so I look up the error "Setting P-Limit is unsupported on this system" which leads me to a 6-day-old issue on pike's ssdt generator github, which suggests removing the _DSM method from the ssdt. I did this and sierra now boots without null cpu kext, with a custom ssdt and with the -xcpm_ipi arg.

Unfortunately, when I pay close attention to the startup logs I see "this is an unknown CPU model 0x3f -- power management may be incomplete or unsupported"; more googling reveals pike's statement that this means xcpm is not actually being loaded :/. The machine runs but it only geekbenches at around 57k (what I interpret as not running at turbo speeds). Unfortunately I am having the hardest time finding a cpu monitoring program that actually works on this system; Intel Power Gadget refuses to install saying it can't handle 2 cpu systems, and hwmonitor CPUSensor.kext kernel panics for lack of broadwell support. And AppleIntelInfo.kext has always crashed on this machine, so its not much help either.

So right now I'm at something of an impasse; I'm on the edge of having working native power management. I have no good method for confirming what its actually doing. And I have some wishy-washy evidence that its not actually reaching turbo speeds.

 

[volition]

@catswift, great stuff. With Sierra, I had issues but can now get in with just the base BIOS and EFI settings from before and your additions, with a few small changes. Last digit of FakeCPUId to 1 (reflects stepping Geekbench reports for the CPU's I'm using), and using OsxAptioFix2Drv-64 (like with 10.11.6). Not using '2' yielded "OsxAptioFixDrv: Error - requested memory exceeds our allocated relocation block" after the +++'s. Had to be the kext patch and/or FakeCPUId that made the difference. So thanks again!

Had to to add GraphicsEnabler=No too, to get past display loss. Probably related to the RX 480 and Sierra nearly having native support for it. With that, it's unaccelerated like 10.11.6, but in contrast has full resolution.

(I'd already tried everything apart from (ACPI, kext) patches. The most extreme being to install and use 10.12 to and from a physical disk (linked through a raw VMDK) in VirtualBox on 10.11.6. On the actual X10DAi hardware, both trying to boot that disk (which I now can, adding the changes above) and installing 10.12 with Clover, would get stuck or reboot just after mbinit (at PCI configuration if using -x).

Adding the RAM to config.plist worked too. Both 10.11.6 and 10.12 now report 64 GB. Added 4 entries for slots 0, 1, 7 and 8. All with Kingston DDR4 at 1867 MHz (leaving the model and make info blank). So thanks yet again.

With all of that, I'm not seeing any difference in Geekbench 3 and 4 scores (consistently lower in 4) in Sierra. Incidentally, I added the Kernel CPU patch, but can't see any impact.

For native CPU power management, I see that message about 0x3f just before the login screen too. Isn't there some way to save the logs? Like this?

I tried Intel Power Gadget with the same results a couple of weeks back, and had crashes when trying to load AppleIntelInfo.kext too. I tried a few monitoring apps last night, but I guess if they won't work for you now, they won't work when using NullCPUPowerManagement like me. I'll try generating and patching the ACPI stuff again, after cloning what I have now to a new disk.

Lastly, TB still works (daisy chaining a Drobo 5D and UAD Apollo. I have a display at the end of it which doesn't though).

 

[catswift]

With all of that, I'm not seeing any difference in Geekbench 3 and 4 scores (consistently lower in 4) in Sierra. Incidentally, I added the Kernel CPU patch, but can't see any impact. For native CPU power management, I see that message about 0x3f just before the login screen too. Isn't there some way to save the logs? Like this?

@volition It looks like there should be, but I don't know how to do it (this is still my first hackintosh). I've been taking a lot of videos of the boot seq with my phone to catch all of the logs :/

Geekbench 3 & 4 use completely different metrics so don't cross-compare the numbers (took me a bit to figure this out as well).

In other news, I spent all weekend and still came up short of a solution to get xnu native power management working. Still stuck in the same spot; either I get ssdt pstates loading and the P-Limit kernel panic, or I get the invalid cpu and no xcpm. I've found 4 other instances of people hitting the P-Limit kp, but none of them figured it out as far as I can find.

What does seem to work though (and I'm hesitant to mention it cause it just feels wrong ), is a kext from okrasit (msr199.kext) + NullCPUPowerManage.kext (no ssdts or anything required). If I understand the theory correctly, his kext basically tells the cpus to turbo, and then relies on a bios setting which then keeps it from actually turbo-ing all of the time (in our bios that setting is "Enhanced Halt State" under CPU C State Control under power management (this is wrong, see below)). I'm a bit uncomfortable with it right now, simply because I still have no mechanism for displaying CPU temperatures in os x; I do observe the system spin up fans during geekbench and then spinning them back down afterward, so I'm 99% positive its not all-turbo-all-the-time. However, with this combination I'm finally pulling the speed I was expecting (more than I was getting with KernelCPU which had the clock drift and usb audio issues). So for now, I'm happy. I'll report back if the cpus end up melting on me :/


[edit]
I pulled the "Enhanced Halt State" from the manual for the mb; it was incorrect. In my bios the setting is called "Enable CPU Autonomous Cstate" under "CPU HWPM State Control".
[/edit]

 

[volition]

@volition It looks like there should be, but I don't know how to do it (this is still my first hackintosh). I've been taking a lot of videos of the boot seq with my phone to catch all of the logs :/

Same here. Including a few screenshots from my last Sierra test install where I guess I checked something in Clover to add a system preferences pane, where you have log options. A bit short on time so haven't tried them yet.

In other news, I spent all weekend and still came up short of a solution to get xnu native power management working.

Thanks I'll try that other kext and BIOS changes ASAP. Not worried about these CPU's. In even other news, it seems metacollin is still into it, has Sierra going and is working on audio. Maybe just not getting notifications for this thread.

 

[catswift]

Thanks I'll try that other kext and BIOS changes ASAP. Not worried about these CPU's. In even other news, it seems metacollin is still into it, has Sierra going and is working on audio. Maybe just not getting notifications for this thread.

@volition Nice! I had tried the AppleALC kext but that was probably a day or two before @metacollin patch, will have to revisit.

I'm happy to report I used the hackintosh for my entire work day yesterday; super fast machine, very very pleased. No cpus melted. Got a geekbench3 of 73k once, highest yet for me. I did encounter one issue where my usb audio would get crackly after 4-6 minutes of use, so I made the below changes to the BIOS and that fixed it; listened to music for hours perfectly fine after that.

If @metacollin does revisit his thread, I want to thank him in advance for paving the way for us. Doing this hack has been the funnest project I've done in a long time, and I probably wouldn't have done it if he hadn't posted about it.




For the BIOS settings for the USB audio fix, under Chipset Config -> South Bridge:

Legacy USB Support - Enabled
XCHI Hand-Off - Enabled
ECHI Hand-Off - Disabled
Port 60/64 - Disabled
USB 3.0 Support - Disabled
EHCI1 - Enabled
EHCI2 - Enabled
Azalia - Disabled

 

[volition]

@catswift, I have AppleALC working now. I haven't checked it extensively, since I'm using the UAD Apollo for audio, but I had headphone output at least. Just built the latest master from the AppleALC repo, which includes @metacollin's patch. Copied .kext over to EFI partition's ../11.12 and ../Other, changed config.plist Devices/Audio to just read

        <key>Audio</key>
        <dict>
            <key>Inject</key>
            <integer>4</integer>
        </dict>

and it worked. No BIOS changes. Attaching the .kext here in case. A few screenshots with layout-id 4



For xnu PM, I tried the .kext approach and it seems to have made some marginal difference. Ran GB3 a few times and multi-score went from ~24k to ~25k with these low-end v4's. There's a bit more fan noise too.

I wasn't aware of this thread until after I got the X10DAi and TB AIC. But incredibly glad I found it, and for @metacollin, yours and other ppl's help, because after a week of attempting I still wasn't getting anywhere with macOS. Now it's nearly all there. Just missing xnu PM (and some better CPU's to take advantage of it) and accelerated GPU. USB 3 would be nice, but I can go without. It's been a great first experience considering we have something probably better than a maxed-out Broadwell-EP Mac Pro if Apple ever introduce them, at a fraction of the cost. Not my last Hackintosh.

 

[catswift]

@catswift, I have AppleALC working now. I haven't checked it extensively, since I'm using the UAD Apollo for audio, but I had headphone output at least. Just built the latest master from the AppleALC repo, which includes @metacollin's patch. Copied .kext over to EFI partition's ../11.12 and ../Other, changed config.plist Devices/Audio to just read ... and it worked. No BIOS changes. Attaching the .kext here in case.

Dang... I installed it and nothing new showed up in my audio input / outputs. Built fresh from the repository, only difference in the config.plist is mine has ResetHDA to true. Not going to worry too much about it though, this little usb adaptor is working fine.

 

[volition]

@catswift, just a heads up I contacted metacollin about power management. He solved it for 10.12 using multiple kernel patches unique to the generation of CPU's. Should see a guide in the near distant future.