Contribute
Register

AMD Radeon Navi GPU

Its looking right.

Try to add boot flags: shikigva=144 and boot with WEG.

Still a no go. Also tried with shikigva=16 which is what is described in the WEG release notes for 1.3.5:
  • Added shikigva=16 (repurposed) property to use AMD hardware DRM decoder in select apps

FWIW - iGPU enabled, same 0.72% CPU usage and still using the AMDRadeonVADrive2
 
Last edited:
Still a no go. Also tried with shikigva=16 which is what is described in the WEG release notes for 1.3.5:
  • Added shikigva=16 (repurposed) property to use AMD hardware DRM decoder in select apps

FWIW - iGPU enabled, same 0.72% CPU usage and still using the AMDRadeonVADrive2
shikigva=16 is recomended when shikigva=80 doesn't works or has freezing problems. Did you tried shikigva=80 first before shikigva=16? If not then you can try it and see if it works.
 
Turns out it was a browser issue, go figure. After clearing all local data for netflix.com it is working fine. I am not using any shikigva bootargs, just WEG and the SSDT-NAVI-RX5700XT.aml.
 
@mattystonnie Does the SSDT for RX 5700 XT work with RX 5700 (non-XT) ? Or is an RX 5700 SSDT in the works? Appreciate all the work that you have done :)
 
Someone knows benchmark apps that uses metal and open cl?

I know luxmark and geekbench.

Edit. Never mind, found a lot of them in the App Store :lol:.
 
Last edited:
I ran several benchmarks. Here are the results :
Capture d’écran 2020-04-22 à 00.58.56.jpg

For each configuration I ran three times the bench'. I averaged and rounded it off. Indeed, the variation of a Metal bench is about 800 and OpenCL : 200.
With the mattystonnie's SSDT the Metal score goes up 70%, ±2% ; and OpenCL score : 63%, ±1%. It's very impressive.

The CineBench R15 OpenGL benchmark shows an increase between 1% and 2%. It may not be significant.

A Novabench benchmark claims to give a result in GFLOPS with OpenCL and show no difference.

Maybe I'll bench with Kishonti's apps later. You will have to specify me some configuration elements, because I know that there, I've been exploring on meaningless criteria because I didn't understand all the advice given here.

the "patch" I'm talking about in the table is the "change PEGP to EGP0" patch in Acpi/DSDT/Patches through Clover Configurator.
VirtualSMC - iMacPro1,1 definition - 5700XT - kexts in /L/E. Clover r5112. agdpmod=pikera was there. Kind regards.
 
I ran several benchmarks. Here are the results :View attachment 464074
For each configuration I ran three times the bench'. I averaged and rounded it off. Indeed, the variation of a Metal bench is about 800 and OpenCL : 200.
With the mattystonnie's SSDT the Metal score goes up 70%, ±2% ; and OpenCL score : 63%, ±1%. It's very impressive.

The CineBench R15 OpenGL benchmark shows an increase between 1% and 2%. It may not be significant.

A Novabench benchmark claims to give a result in GFLOPS with OpenCL and show no difference.

Maybe I'll bench with Kishonti's apps later. You will have to specify me some configuration elements, because I know that there, I've been exploring on meaningless criteria because I didn't understand all the advice given here.

the "patch" I'm talking about in the table is the "change PEGP to EGP0" patch in Acpi/DSDT/Patches through Clover Configurator.
VirtualSMC - iMacPro1,1 definition - 5700XT - kexts in /L/E. Clover r5112. agdpmod=pikera was there. Kind regards.

Dont test Open GL, I think Apple is deprecating OpenGL.

What is really important is the Open CL and Metal performance, apps will use metal.

Future games will use Vulkan, Dota 2 Reborn already uses Vulkan API.
 
The SSDT method works with my setup which consists of an Asus Prime Z370 A ii, 9600k stock clock, XFX RX 5700, and Headless iGPU on the SMBios iMac19,1. Weg is active with ShikiGVA=80 and agdpmod=pikera.
Note: I'm using DisplayPort, not HDMI, the connector furthest from the motherboard. I just tried a film in Apple TV+, and Netflix in Safari and DRM works. @slim.jim you might find that interesting.
I see a notable increase in GeekBench scores (37k to 66k in Metal, and a jump from 40k to 58k in OpenCL) and if I'm not mistaken the BruceX FCPX test went faster by one second. I'm mainly interested in real world performance in Final Cut Pro X and Compressor, so here's some things I discovered:
I noticed that with CMMChris's kext and no Weg, I was able to convert ProRes to h.264 using QuickTime. While not a very sophisticated method, it's very useful me. Using the SSDT with or without Weg, I ran into the -12912 Error and it stops. This doesn't happen in FCP.
The wording is of the error is "The operation could not be completed An unknown error occurred (-12912)"

Exporting my own ProRes file to h.264 using Compressor, Intel Power Gadget indicates a huge spike in the iGPU and the CPU - both seem to max out at 4.5ghz and 1.3gz respectively. Transcoding a 6-minute 1080p file takes 1:21.
The same file, sharing out of FCP, takes 1:37. Not a huge difference, but what I noticed is that in Power Gadget, the iGPU shows similar numbers to just playing the Sony HEVC 4k sample at 0.40mhz - which is low. I've concluded that FCPX doesn't use the GPU for outputting h.264. If selecting an h.265 setting via Compressor, iGPU usage fluctuates between about 0.67 and 0.85mhz usage.
None of these tests change the fact that QuickTime gives me an error trying the "Export As">1080p. This isn't a dealbreaker, but this is a super handy feature for how I use my machine. I'm trying to figure out what @CMMChris's kext does that makes QuickTime exporting work compared to @mattystonnie's SSDT.
Feel free to throw any more tests my way.
 
I am reading through this thread and trying to decide what to do. I have a z390 Designare, 9700k and Sapphire Pulse RX580 8Gb - right now I am using headless platform ID and Lilu/WEG and iMac Pro 1,1 with IGPU activated. What should I do and what are the exact steps to take. I am trying to learn, but I also don't wanna screw up.
 
Back
Top