How to Enable AMD GPU Encoding instead of Intel's?

[shaveblog]

@jaymonkey , just want to give a quick update on this issue. After setting the dGPU as the primary display port and the iGPU became headless, the FCPX stopped working. My PlatformID is set to Macmini8,1 so the FCPX does not expect the iGPU to be headless. I had to reenable the iGFX as the primary display port in BIOS in order to get FCPX working again (even though my only monitor is actually connected to the dGPU)

It seems that I have to change my PlatformID to iMac18,3 as suggested above, but I'm afraid FaceTime will stop working if I do that! I've already spent a long time playing with serial numbers and codes till I got it working and really don't want to do it again! So, will changing the PlatformID from Macmini8,1 to iMac18,3 affect the FaceTime operation?

Try changing your SMBIOS to iMac19,2 and creating an SSDT with ssdtPRGen to enable your BIOS overclock. It's always best to match the SMBIOS as closely as you can with your CPU, and that iMac happens to have your exact CPU albeit the unlocked version (no real Macs are overclocked, even the ones that have K-version CPUs). First you get the SMBIOS right in Clover, then you make a custom SSDT to get your BIOS overclock. When you have done these things, and disable your Intel graphics in both BIOS and Clover, I am confident you will get GPU transcoding happening in FCPX.

Disabling Intel graphics may affect Facetime or it may not. But taking the steps I listed above will get FCPX working again for you with your hardware.

 

[justinhe]

to use dGPU only, the only option is to use SMBIOS iMacpro1,1 + Mojave 10.14.5 + disable iGPU in BIOS, you don't need to set "defaults"

 

[shaveblog]

to use dGPU only, the only option is to use SMBIOS iMacpro1,1 + Mojave 10.14.5 + disable iGPU in BIOS, you don't need to set "defaults"

No, iMacPro SMBIOS is not the “only option”. Nobody needs to spoof an iMac Pro just to use dGPU only. I’ve been doing it for years with an iMac12,2 and later an iMac13,2 SMBIOS (see my post above).

Why do you think forcing a Skylake Xeon SMBIOS on a Coffee Lake CPU is the “only option”? It isn’t, and it may cause more issues than the one issue you seem to believe it alone solves.

FWIW in trying to eke out the most performance from my i7 Ivy/280X system, I tried that iMac Pro SMBIOS. Didn’t improve anything and messed with Speedstep. I went back to the correct SMBIOS for my CPU and everything was better. This is Hackintosh 101. Unless your system has similar hardware there is no benefit to using the iMac Pro SMBIOS. It doesn’t magically boost anything.

 

[justinhe]

if you follow this post: Radeon RX 4XX/5XX standalone system, AMDRadeonX4250.kext (GVA support H264) does not support HEVC HW

you will know why.

PS: Coffee Lake has no difference to Skylake, except for iGPU capability.

 

[shaveblog]

Honestly, if your advice helps the OP, good for both you. I am not an expert here, only another end user offering his experience.

We all have different priorities for our machines. My highest priority is FCPX stability and performance for both ProRes and H.264/x264 transcoding. I am wholly uninterested in H.265 hardware transcoding, whether it’s handled by integrated Intel or discrete GPU. Far from being the new standard, H.265/HEVC was a step backward from H.264, lowering video quality while moving all the expensive energy-intensive processing burden from the content provider to the end user. It has caused problems across the board from gamers who’ve seen system performance plummet to low-powered puck-sized set top boxes which can’t cope with the processing needs of playing glitch-free 4K H.265.

I am fortunate in that nothing I do with my Hackintosh either personally or professionally requires or would even marginally benefit from H.265 hardware transcoding. I don’t see this capability as being a bedrock Hackintosh box that needs ticking for anyone. Maybe I’m missing something here — is GPU H.265 transcoding needed for better Bitcoin mining etc.? If you can tell me what I am missing out on by not having hardware HEVC transcoding on my Hackintosh, please let me know. Until then, I’ll continue to optimize my systems for the highest quality transcoding, which still means software, and still means H/x264.

 

[themacmeister]

I have somehow lost H.264 Radeon RX 580 accelerated encoding/decoding under 10.14.5 with iMacPro1,1 SMBIOS and Dummy141.kext

??!!

I just tested transcode with Handbrake 1.2.2 and it is using CPU only?!

Any help appreciated.

 

[mitch_de]

Handybrake can use the CPU or GPU (by selecting video codec name ends with ...videotoolbox). VideoToolbox its Apples GVA ecoding "box", but uses only iGPUs (=INTEL) and NOT AMDS / NVIDIAs.

 

[oli.mathieu]

Hello

VideoToolbox its Apples GVA ecoding "box", but uses only iGPUs (=INTEL) and NOT AMDS / NVIDIAs.

I've realised that if there isn't any iGPU, it uses CPU
Thanks

 

[macnb]

No, iMacPro SMBIOS is not the “only option”. Nobody needs to spoof an iMac Pro just to use dGPU only. I’ve been doing it for years with an iMac12,2 and later an iMac13,2 SMBIOS (see my post above).

Why do you think forcing a Skylake Xeon SMBIOS on a Coffee Lake CPU is the “only option”? It isn’t, and it may cause more issues than the one issue you seem to believe it alone solves.

FWIW in trying to eke out the most performance from my i7 Ivy/280X system, I tried that iMac Pro SMBIOS. Didn’t improve anything and messed with Speedstep. I went back to the correct SMBIOS for my CPU and everything was better. This is Hackintosh 101. Unless your system has similar hardware there is no benefit to using the iMac Pro SMBIOS. It doesn’t magically boost anything.

I have an i7-3770K OC'ed to 4.8Ghz + RX580.
I disabled the IGPU.
I use iMacPro1,1 model ID

..and you are right, speedstep gets messed up (even though I had an SSDT for CPU freq).
...so the solution was to use CPUFriend.kext that injects the "native" (iMac13,2) freq vectors into an "unnatural" model ID (iMacPro1,1).
I.E. Inject the iMac13,2 freq vectors into the iMacPro1,1 model id.
Now I get the whole range of P-States for the Ivy CPU - all the way up to 4.8Ghz.
See my results here.

..and RX580 is fully utilised for encode/decode on Mojave when using FCPX.

 

[shaveblog]

I have an i7-3770K OC'ed to 4.8Ghz + RX580.
I disabled the IGPU.
I use iMacPro1,1 model ID

..and you are right, speedstep gets messed up (even though I had an SSDT for CPU freq).
...so the solution was to use CPUFriend.kext that injects the "native" (iMac13,2) freq vectors into an "unnatural" model ID (iMacPro1,1).
I.E. Inject the iMac13,2 freq vectors into the iMacPro1,1 model id.
Now I get the whole range of P-States for the Ivy CPU - all the way up to 4.8Ghz.
See my results here.

..and RX580 is fully utilised for encode/decode on Mojave when using FCPX.

Glad you worked things out with your build. I’ll have to check out that CPUFriend kext, sounds interesting, although at this point my various Hacks are so reliable I rarely check back in here anymore, especially since switching them all from MultiBeast to vanilla installations.

FWIW since I posted that message about SMBIOS, I traded up (down?) (sideways?) from the i7/Z77/Ivy system mentioned in that post to a newer Skylake system that somehow feels faster in everyday operation and renders FCPX even faster than the i7 Ivy rig, despite having a lowly i3-6100 CPU. Paired with an AMD 280X GPU, the new system Geekbenches higher in single-core despite being an unoverclockable i3 (at least on this Gigabyte H110M-A) vs. an overclocked Ivy i7. The Ivy scores higher in multi-core but this doesn’t seem to matter with anything I actually do with this system. I mostly use it for FCPX and Adobe Lightroom/Photoshop work and somehow the little power-sipping i3-6100 makes Mojave fly faster than the OC’d i7-3770K ever did. It’s essentially a cheap Hackinmouse build with an added 280X for graphics grunt and it’s by far the fastest and most reliable Hack I’ve built yet. The significantly lower power consumption is a nice bonus as well.

On paper an i3-6100 shouldn’t be this fast and powerful running Mojave, even with that 280X helping out. But it shows how benchmarks don’t always tell the whole story. As someone who spent many hours trying to eke the highest OC out of that 3770K to get another 100 Geekbench points and maybe a few seconds faster BruceX, I’m happy leaving this puny i3 and outdated AMD alone while they fly through whatever I‘m doing.