Huananzhi F8D + x2 E5 2678v3 + 128Gb Samsung DDR4 Reg ECC + OpenCore + Monterey

[tarkh]

Huananzhi F8D + x2 E5 2678v3 + 128Gb Samsung DDR4 Reg ECC + OpenCore + Monterey
(assembled partly from used components)

​

Components:
(I ordered most of the components directly from AliExpress, but tried to find analogs on Newegg)

Motherboard: Huananzhi F8D (AliExpress | Newegg)
CPU: x2 E5 2678v3 (AliExpress | Newegg) (used)
Memory: x8 16Gb Samsung DDR4 Reg ECC (AliExpress | Newegg) (used)
CPU cooler: Huananzhi A500 (AliExpress | Newegg)
WiFi+BT PCIe: Fenvi T919 (AliExpress | Newegg)
Case: Deepcool MATREXX 55 (Newegg)

Bios:
There are some limitations in the stock bios, for example, closed ME-regions, not precise BCLK frequency, the lack of memory timing settings and the limitation of Turbo mode on all processor cores.

Despite the fact that macOS will work perfectly with the stock bios, enthusiasts can flash the bios to a custom one, in which all restrictions have been removed and some improvements have been added. To do this, you will need a CH341A programmer with a test clip (Newegg) and a modified bios. It is also possible to flash the bios using the Afudos program, but the firmware will be incomplete and the ME-regions will remain closed. I strongly advise you to buy a programmer, because the bios flash process may not go according to the plan and the motherboard may turn into a "brick". It is possible to restore the operability of such board only with the help of a programmer.

A separate topic is the unlocking of turbo mode on all processor cores and its subsequent undervolting. This can be done in several ways: through flashing the values directly into the BIOS, or through pre-assembled EFI modules. I've tried both options, but working through EFI modules is faster and more convenient for me.

You can find detailed manual on how to flash bios on Huananzhi F8D, unlock Turbo mode on all E5 2678v3 cores, all modified bios firmware and all proper bios settings for macOS in my GitHub repository (see below).

How to Create a macOS Monterey Installation USB:
Use Tonymacx86 Guide to create USB installation media, then mount it's EFI partition and copy EFI folder from my GitHub repository (see below) to the root of the mounted EFI patition. Also do not forget to alter OpenCore config.plist by generating your own serial numbers for iMac17,1.

Common notes:

• My custom EFI already preconfigured to boot up installation media and installed macOS Monterey itself. It contains all necessary precompiled ACPI modules, kexts and assets. System is stable, power manegement works fine, no KP and other unpleasant stuff. Also this EFI should work with macOS Mojave, Catalina and Big Sur.
• Even that Huananzhi F8D is x99 platform, it contains server chipsets, and because of this NVRAM works fine.
• I own specific WiFi+BT PCIe card, which is supported by macOS natively, so everything related to iMessage, AirDrop and Handoff works normally.
• System memory: this motherboard supports 8 channel DDR4 RAM (x2 CPUs with 4 channel each), so if you utilize all 8 memory slots, this WILL give you performance boost up to 15% comparing to only 4 utilized slots (2 for each CPU).
• My Workstation was made for sound production in Logic Pro X, so video adapter is not important on this rig. I use Nvidia GT 740 2GB, which is perfect for my needs and supports two 2K 31" displays. However, starting from macOS Monterey Apple dropped support for Nvidia Kepler, but it is still possible to revert Kepler drivers with Geforce-Kepler-patcher V2.
• Built-in sound tuned and works fine, but because system is for sound production, I have second external MOTU Ultralite-mk3 Hybrid card.
• USB mapping was done by Dortania instruction. It seems like all USB ports on Huananzhi F8D located on one controller, so there is under 15 ports limit and I've shrinked some USB3 ports to USB3-only mode. Also I own WiFi+BT PCIe card (plugged in the second PCIe port from top) and Bluetooth device takes one USB port. My USB mapping is related to my hardware setup and it is possible, that you need to alter my USB-Map.kext with your specific options.
• Turbo Mode unlock on all CPU cores at the same time works well, system shows 10-15% productivity gain in benchmarks. But in production I've disabled Turbo Mode unlock, because I simply don't need this amount of power. 24 cores and 48 threads for smooth DAW operation doing pretty well, so I stick with 2.5 GHz base clock and stock Turbo Boost to 3.3 GHz on max 3-4 cores at the same time.

What do not work:
For now only system sleep doesn't work. Sadly, but because of 2 CPU sockets and macOS design, system wake from sleep is broken. If anyone will find solution to this problem, please let me know.

My GitHub repository for Huananzhi F8D + E5 2678v3:
My open source GitHub repository is specially dedicated to this build. The reason I do not re-post all information and assets here is because I frequently update them in upstream and it is really hard to maintain synchronization on multiple resources. So if you have the same setup, feel free to clone my repo.

Conclusion:
I've ordered listed above parts from China (to Russia) and total cost of this rig was under $ 1.000 (excluding SSD's and PSU that I already had), which is pretty good considering its performance. The price quality ratio suits me quite well for my current tasks. I builded this setup in February 2021 and have been working on it for a year now.

Benchmarks:

(Xcode benchmark, compare to Mac results)

(Cinebench R23)

(Geekbench 4 & 5)

 

[WhenMusicAttacks]

Thanks for the interesting journey but..dual Xeon with low frequency for audio production? That's a huge waist of components and time for something that will perform below i7 6700k in 99% of projects. And that includes even huge orchestral template wich are supposed to make use of that many cores. As you start doing soem real work with groups and buses, you will chocke even at 1024. Sure the logic pro benchmark will show 240 tracks but that's a dummy load.

 

[tarkh]

Thanks for the interesting journey but..dual Xeon with low frequency for audio production? That's a huge waist of components and time for something that will perform below i7 6700k in 99% of projects. And that includes even huge orchestral template wich are supposed to make use of that many cores. As you start doing soem real work with groups and buses, you will chocke even at 1024. Sure the logic pro benchmark will show 240 tracks but that's a dummy load.

Well I can't agree with that. Previously I've had studio setup with i7 5960X 8/16 @ 4.2 GHz and obviously dual Xeons perform way better on heavy load with a LOT of sends/insets/channels and plugins. Logic Pro X designed to utilize all cores and their number is more important then CPU clock itself. So 2.5-3.3 GHz in my case works just fine with 48 threads enabled. All heavy instruments as Omnisphere, Nexus 3, some of Arturia, MPC, Native Instruments and NI Komplete with bunch of chains - I simply do not have any troubles with them. It's good with doing soundtracks/soundfx for TV shows and recording/making music, both classic or modern. Buffer size usually stays between 64-256. Largest projects usually do not exceed 120-150 channels in my case, but number of channels heavily depends on workflow. And yes I do use groups and buses, here is one of my templates:

The main reason why such setup (dual Xeon) can perform really bad on Hackintosh is wrong PM setup, lack of necessary kernel patches and ACPI modules. If you tune your Hack in a proper way, then Logic Pro X performs smooth and without any problems (crashes, glitches etc.).

 

[WhenMusicAttacks]

Well I can't agree with that. Previously I've had studio setup with i7 5960X 8/16 @ 4.2 GHz and obviously dual Xeons perform way better on heavy load with a LOT of sends/insets/channels and plugins. Logic Pro X designed to utilize all cores and their number is more important then CPU clock itself. So 2.5-3.3 GHz in my case works just fine with 48 threads enabled. All heavy instruments as Omnisphere, Nexus 3, some of Arturia, MPC, Native Instruments and NI Komplete with bunch of chains - I simply do not have any troubles with them. It's good with doing soundtracks/soundfx for TV shows and recording/making music, both classic or modern. Buffer size usually stays between 64-256. Largest projects usually do not exceed 120-150 channels in my case, but number of channels heavily depends on workflow. And yes I do use groups and buses, here is one of my templates:

View attachment 540588​

The main reason why such setup (dual Xeon) can perform really bad on Hackintosh is wrong PM setup, lack of necessary kernel patches and ACPI modules. If you tune your Hack in a proper way, then Logic Pro X performs smooth and without any problems (crashes, glitches etc.).

We have many more tracks than that in our orchestra templates, and the 6700k perfromed above even 7960x, we had tried 2678 v3 before - we got very nice logic benchmark scores but projects would not load as they did on the simple 4 core

Always on Logic Pro. I was able to replicate the same issues in different OS and DAWs like Cubase. Projects are limited by single core and architecture, not number of cores

 

[tarkh]

We have many more tracks than that in our orchestra templates, and the 6700k perfromed above even 7960x, we had tried 2678 v3 before - we got very nice logic benchmark scores but projects would not load as they did on the simple 4 core

Always on Logic Pro. I was able to replicate the same issues in different OS and DAWs like Cubase. Projects are limited by single core and architecture, not number of cores

Well then there was some issue with CPU setup within Hackintosh ecosystem. Bad vectors, interrupts, latency etc... Cant say more, because I've had opposite experience. If you can give me any project or point some plugin/chains to test on my machine - that will be interesting.

 

[WhenMusicAttacks]

Well then there was some issue with CPU setup within Hackintosh ecosystem. Bad vectors, interrupts, latency etc... Cant say more, because I've had opposite experience. If you can give me any project or point some plugin/chains to test on my machine - that will be interesting.

xcpm was implemented correctly, and also my experience is the same on Windows, we have a slave machine running eastwest play as that engine is not efficient on macOS, connected using VEP.

 

[tarkh]

xcpm was implemented correctly, and also my experience is the same on Windows, we have a slave machine running eastwest play as that engine is not efficient on macOS, connected using VEP.

Well IDK how correctly xcpm was implemented, but that's strange. Also can't say anything about Xeon + Windows, and it's DAW software. However, in my case on macOS LPX utilize CPU just fine. Here is 126 track project (old and not optimized) with bunch of plugins on almost all channels, buses, etc. As you can see, the CPU usage while playing/seeking is around 25-30%. Rarely spikes go up to 50-75% in my practice. Only with some VST in large strip.

It's really important to properly tune the CPU for macOS. I've never heard, that CPU clock is more important for LPX then core count. Well, maybe if cores work below 2GHz - yes, but in my case it is 3-3.3GHz doing job well. BIOS settings, it's firmware and MB chipset important too. So can you please suggest me any Virtual Instrument for LPX (with any channel strip setup) on which you faced stuttering and glitches while using Xeon? I'm really interesting to dive in to this problem and test it on my hardware. Thanx!

 

[theroadw]

@WhenMusicAttacks I also thought CPU speed played a bigger role in DAW performance, but I guess it's the balance of Memory speed, bus speed, cpu speed, as well as number of cores.
My main rig was a z820 with dual E5-2667 and happily chugged along playing 400+ track mix sessions with lot's of plugins and automation and only in rare instances some plugin/track count combo made it strain a bit.
Last year while mixing a movie, I had to bounce some heavily plugin processed tracks or those sections would get playback errors once every 10 plays or so.
Because of friends advise I bought a M1Pro MBP to test if it was capable of helping out, and to my surprise it basically replaced the monster, and with 30%+ cpu headroom.
The clock speed on the MBP is lower and it has less cores, but the ram/bus is so much faster that sessions run perfectly at 64 sample buffer. All while running in Rosetta and while browsing the web.
So long story short, it may be that other factors are at play, os simply that the session or DAW needs vary from person to person.

 

[tarkh]

theroadw, especially latest versions of LPX really improved multi-core support. On Mojave and Logic 10.4 I've had some glitches with some third-party plugins, on 10.5 mostly. Everything was OK and now with 10.7 version - no issues at all.

Returning to the topic of my build - the main point here is the price quality ratio - MB=250$, 2xCPUs 24/48 threads total @ 3.3GHz with Turbo Unlock hack=250$, 4x16 Gb DDR4 Samsung 2133 which you can overclock to 2400-2666/decrease timings=200$... So for 700$ you'll get pretty powerful rig core.

 

[WhenMusicAttacks]

@WhenMusicAttacks I also thought CPU speed played a bigger role in DAW performance, but I guess it's the balance of Memory speed, bus speed, cpu speed, as well as number of cores.
My main rig was a z820 with dual E5-2667 and happily chugged along playing 400+ track mix sessions with lot's of plugins and automation and only in rare instances some plugin/track count combo made it strain a bit.
Last year while mixing a movie, I had to bounce some heavily plugin processed tracks or those sections would get playback errors once every 10 plays or so.
Because of friends advise I bought a M1Pro MBP to test if it was capable of helping out, and to my surprise it basically replaced the monster, and with 30%+ cpu headroom.
The clock speed on the MBP is lower and it has less cores, but the ram/bus is so much faster that sessions run perfectly at 64 sample buffer. All while running in Rosetta and while browsing the web.
So long story short, it may be that other factors are at play, os simply that the session or DAW needs vary from person to person.

That is exactly the issue with dual Xeons compared to single socket consumer Intel. Dual Xeons have so much inter core and then inter socket latencies, that's why a single socket consumer Intel CPU with a very short ringbus will outperform it in real world tasks that require low latency such as audio.

The M1 has much better single core performance; the clock speed means nothing when comparing different architectures. I am talking about comparing frequency on Intel Haswell to Sky (Coffee) Lake wich have very similar IPC.