How to build your own iMac Pro [Successful Build/Extended Guide]

[kgp]

System froze just now... after more than 2 days of uptime. So none of the changes I've made are relevant to fix the issue.

Now I'm thinking that it might be GPU related. What I just did before the freeze happened is to put the screens to sleep. The screens are off now, so I cannot turn them back on, I can't ssh to the system, hard drive LED doesn't blink either (meaning the system isn't working).

Sorry for the false hope .

EDIT: The panic report is now different from what I had earlier...

Anonymous UUID:       3016C5F0-081C-315F-2D7E-D03EF4FFBBA3

Wed Jan 31 06:47:13 2018

*** Panic Report ***
panic(cpu 4 caller 0xffffff801816e349): Kernel trap at 0xffffff8018000a2c, type 14=page fault, registers:
CR0: 0x000000008001003b, CR2: 0xffffff832deffaf8, CR3: 0x00000018bdb440d6, CR4: 0x00000000003627e0
RAX: 0xffffff8017d06020, RBX: 0x000000000000a75a, RCX: 0xffffff8129da9a10, RDX: 0x0000000062c3f35b
RSP: 0xfffffd000004e720, RBP: 0xffffff8e2a9b3dd0, RSI: 0x0000000000000008, RDI: 0xffffff801880af78
R8:  0x0000000000000000, R9:  0x0000000000000001, R10: 0x0000000000000000, R11: 0x0000000000000202
R12: 0x0000000000000000, R13: 0x0000000000000000, R14: 0xffffff8018775ff7, R15: 0xfffffd000004e720
RFL: 0x0000000000010002, RIP: 0xffffff8018000a2c, CS:  0x0000000000000008, SS:  0x0000000000000010
Fault CR2: 0xffffff832deffaf8, Error code: 0x0000000000000000, Fault CPU: 0x4, PL: 3, VF: 5

Backtrace (CPU 4), Frame : Return Address
0xfffffd000004dec0 : 0xffffff801804f606 mach_kernel : _handle_debugger_trap + 0x506
0xfffffd000004df10 : 0xffffff801817c744 mach_kernel : _kdp_i386_trap + 0x114
0xfffffd000004df50 : 0xffffff801816e159 mach_kernel : _kernel_trap + 0x5e9
0xfffffd000004dfd0 : 0xffffff8018001120 mach_kernel : _return_from_trap + 0xe0
0xfffffd000004dff0 : 0xffffff801804f03c mach_kernel : _panic_trap_to_debugger + 0x25c
0xfffffd000004e120 : 0xffffff801804edbc mach_kernel : _panic + 0x5c
0xfffffd000004e180 : 0xffffff8018179e08 mach_kernel : _mp_kdp_enter + 0x468
0xfffffd000004e1d0 : 0xffffff801804f16a mach_kernel : _handle_debugger_trap + 0x6a
0xfffffd000004e220 : 0xffffff801817c744 mach_kernel : _kdp_i386_trap + 0x114
0xfffffd000004e260 : 0xffffff801816e159 mach_kernel : _kernel_trap + 0x5e9
0xfffffd000004e2e0 : 0xffffff8018001120 mach_kernel : _return_from_trap + 0xe0
0xfffffd000004e300 : 0xffffff801804f03c mach_kernel : _panic_trap_to_debugger + 0x25c
0xfffffd000004e430 : 0xffffff801804edbc mach_kernel : _panic + 0x5c
0xfffffd000004e490 : 0xffffff801816e349 mach_kernel : _kernel_trap + 0x7d9
0xfffffd000004e610 : 0xffffff8018001120 mach_kernel : _return_from_trap + 0xe0
0xfffffd000004e630 : 0xffffff8018000a2c mach_kernel : _ks_32bit_entry_check + 0x16c
0xffffff8e2a9b3dd0 : 0xffffff80181776f3 mach_kernel : _mp_cpus_call_cpu_init + 0x3e3
0xffffff8e2a9b3e40 : 0xffffff8018177278 mach_kernel : _cpu_signal_handler + 0x3c8
0xffffff8e2a9b3f50 : 0xffffff8018176053 mach_kernel : _lapic_interrupt + 0x43
0xffffff8e2a9b3f70 : 0xffffff801816d856 mach_kernel : _interrupt + 0x136
0xffffff8e2a9b3fd0 : 0xffffff80180012d6 mach_kernel : _hndl_allintrs + 0x126
0xffffffce85b43df0 : 0xffffff801816503f mach_kernel : _usimple_lock + 0x2f
0xffffffce85b43e30 : 0xffffff8018036afd mach_kernel : _ipc_mqueue_copyin + 0x8d
0xffffffce85b43e60 : 0xffffff8018044897 mach_kernel : _mach_msg_overwrite_trap + 0x97
0xffffffce85b43ef0 : 0xffffff801815858d mach_kernel : _mach_call_munger64 + 0x1fd
0xffffffce85b43fa0 : 0xffffff8018001926 mach_kernel : _hndl_mach_scall64 + 0x16

BSD process name corresponding to current thread: ccminer
Boot args: -v dart=0 keepsyms=1 debug=0x100

Mac OS version:
17D2047

Kernel version:
Darwin Kernel Version 17.4.0: Tue Dec 19 10:52:05 PST 2017; root:xnu-4570.47.16~1/RELEASE_X86_64
Kernel UUID: 29EF1F85-10C5-3F25-AED6-5529E5948D42
Kernel slide:     0x0000000017c00000
Kernel text base: 0xffffff8017e00000
__HIB  text base: 0xffffff8017d00000
System model name: iMacPro1,1 (Mac-7BA5B2D9E42DDD94)

System uptime in nanoseconds: 184007461548435
last loaded kext at 92605088619172: com.apple.driver.CoreStorageFsck    546 (addr 0xffffff7f9b863000, size 98304)
last unloaded kext at 92785996583668: com.apple.driver.usb.cdc    5.0.0 (addr 0xffffff7f9b854000, size 28672)

Just something that looks suspicious to me from your boot flags.. Do you really need dart=0? Did you already give it a try to booth with the additional boot flag ncpi=0x2000? I don't need that one but you might ... you have also no dark wake boot flag by the way...

Just some thoughts that came to my mind by inspecting your KP image..

 

[kgp]

@kgp Question: I have a m.2 Samsung 1tb Drive. I just did a clean install on a new system with 10.13.3 imac pro following all your settings and your EFI files.

My question is regarding TRIM, I don't know if my trim is enabled for my M.2 drive, with HFS+ format (not APFS), please see my screenshots:
View attachment 309852 View attachment 309851
View attachment 309850 View attachment 309851

Click on NVMeExpress and then on your NVMe Drive and you will see TRIM Support: Yes

 

[kgp]

On the other hand, my gaming 9 has been outstanding. I don't need Thunderbolt and the sound from the DAC to my super nice old school sound system is perfect.

Sure

Sorry for likely offending you indirectly with my straight blablabla above, Pete.. This was not at all my intention

I do appreciate all your work and contributions!

But I still think that the Designare is a major step ahead What is your opinion? Are you planning to change as well?

 

[Thireus]

Just something that looks suspicious to me from your boot flags.. Do you really need dart=0? Did you already give it a try to booth with the additional boot flag ncpi=0x2000? I don't need that one but you might ... you have also no dark wake boot flag by the way...

Just some thoughts that came to my mind by inspecting your KP image..

I'm running some VMs, I have VT-d enabled, so I believe dart=0 is required (although I am able to boot without this flag).
Regarding ncpi=0x2000, I haven't noticed any difference booting with and without it, so I decided not to use it. I've read on several posts that it's advised not to use it if the system can boot without it.
You're correct about dark wake, that's something I'll have to play with when I'll start tackling system hibernation. But for now since I'm not letting the system hibernate I believe this isn't required.

 

[kgp]

I'm running some VMs, I have VT-d enabled, so I believe dart=0 is required (although I am able to boot without this flag).
Regarding ncpi=0x2000, I haven't noticed any difference booting with and without it, so I decided not to use it. I've read on several posts that it's advised not to use it if the system can boot without it.
You're correct about dark wake, that's something I'll have to play with when I'll start tackling system hibernation. But for now since I'm not letting the system hibernate I believe this isn't required.

I guess dart is obsolete, even with VT-d enabled. My thought was just whether or not npci=0x2000 could help you in solving your KP issues. You are right that darkwake just helps with the hibernation and should not be related with your KP issues. And honestly, I think dart neither...

Anyway, it was just something I wanted to comment on, while inspecting your KP image..

By the way there are several other options you could play with on your Rampage VI.. although with the ASUS Prime X299 Deluxe none of these option needs to be checked in addition..

Just checked the EFI of @AnaktuvGod in addition:

He checked boot flags "npci=0x2000" and "darkwake=1", checked also "KernelPM" and also enables "xcpm_core_scope_msrs © Pike R. Alpha"..

 

[izo1]

On the other hand, my gaming 9 has been outstanding. I don't need Thunderbolt and the sound from the DAC to my super nice old school sound system is perfect.

Yeah I liked the Gaming 9 (Besides all the crazy RGB LEDs everywhere, which I always turn off anyway) but I'm having issues and it has to do with it coming from the factory like that (ie there are some ports that are chipped and broken and there's a hole where the CPU is mounted). It seems Gigabyte is discontinuing the Gaming 9? On Amazon Designare EX is listed as a "New updated version" of Gaming 9.

I mostly use Bluetooth headphones and have very high end DACs for anything serious, so not worried about "losing" the Saber DAC from the Gaming 9. This is more of a workstation for me.

Thunderbolt 3 onboard is very nice though, I was going to buy the Addon card for the Gaming 9 in the future since I have Thunderbolt drives. But having it on board is very impressive and it seems like a good trend by PC makers. It seems like something finally for once the PC industry is taking from Apple (as they usually are early adopters).

I just wish they added 10GbE on this board (Aquantia makes very affordable chipsets and it's used in the new iMac Pro). I think all boards nowadays should come with 1x10GbE and 1x1GbE for legacy. The higher end ASRock X299 comes with 10GbE.

I have to say, though, coming from mostly Asus, Gigabyte BIOS is really boring and doesn't have the options that Asus has in their higher end boards.

Surprisingly it seems like the Designare EX is supposed to be their most top end X299 board...but they still have the Gaming 9 on their landing splash page

Also what's nice is these 2 boards are so similar I'm just going to swap them and go back to being normal without wasting too much time.

There is a nice writeup on the Designare EX

PS. For the TB3 ports on the back, it requires Display port inputs since X299 doesn't have iGPU to enable that.

For example, the iMac Pro TB3 ports support 5k monitors, has anyone tested anything above 4k with TB3 on the X299s?

If not I will try to test on my end. Having TB3 capability is so nice.

 

[kgp]

@Thireus , just updated post #4688.. please check on that.. maybe it helps..

 

[AnaktuvGod]

@kgp Question: I have a m.2 Samsung 1tb Drive. I just did a clean install on a new system with 10.13.3 imac pro following all your settings and your EFI files.

My question is regarding TRIM, I don't know if my trim is enabled for my M.2 drive, with HFS+ format (not APFS), please see my screenshots:
View attachment 309852 View attachment 309851

Check NVMExpress on the left

 

[ccozmo]

Hi @ccozmo,

Did you ever get a response to this question? Or have you tried one on your own?

I'm interested in setting up a RAID and was wondering what the best configuration was.

@Thireus, I think you said you set up a RAID, would you mind sharing what mechanism you used?

Thanks!

Yeah I tested it out for HFS, as RAID is not officially supported for APFS. The write speeds were faster, however I'm going to wait until I find out how iMacPro does it with APFS and then try again

Below is a screen shot of test with single APFS drive, and RAID HFS drive

 

[mgregrs]

@DSM2 and @interferenc ,

The Designare motherboard of @DSM2 has unlocked MSR register. This this is a completely different story and topic...

For me the interesting question is why Sleep/Wake with the Vega worked for @DSM2 with the patched ASUS Prime X299 Deluxe and without the "xcpm_core_scope_msrs" kernel patch.

Error E3 might be indirectly GPU related. Maybe Nvidia GPUs cause the OSX Kernel to write to a MSR Register bit on Wake, which is not yet unlocked. Fact is that the latter causes an S3 related E3 mainboard error and System KP. The system just freezes due to error protection C.11) - boot flags "debug=0x100" and "keepsyms=1", otherwise the system would just reboot. There is no way to access the system via remote access. The system is simply dead after wake.

To prove my theory, somebody else with a Vega and a patched ASUS Prime X299 Deluxe, should try to Sleep/Wake his system with the "xcpm_core_scope_msrs" kernel patch disabled. If this works flawless like in case of @DSM2, the issue might be indirectly GPU related as described above... There is no other way to confirm my suspicion.

Thus guys, we need a crash test dummy with a a Vega and a patched ASUS Prime X299 Deluxe with disabled "xcpm_core_scope_msrs" kernel patch! Anybody with such configuration willing to do the Sleep/Wake test I propose above? Your help and feedback is urgently needed!

Solution: to add another BIOS firmware patch, which unlocks the affected MSR Register Bit addressed by Nvidia GPUs on Wake and related with S3.

... I have VEGA 64, last 13.3 and 13.3 public beta on 2 different SSDs and a patched ASUS Prime X299 Deluxe with disabled "xcpm_core_scope_msrs" kernel patch!

Unfortunately actually (last X hours) I have to finish some part of work on s#1117y windows :/
and unfortunately I'm so tired that my head almost sleep alone on my desk
- I promise to share some feedback as fast as possible about wake/sleep with Vega + our Motherboard
& needed details mentionned by You @kgp....
(I hope.. before my actual sleep/wake problem with my head )

Cheers, Peace, Love and a few hours of rest