[Success] AMD RX6000 Series working in macOS

[Usal]

Here is something for you... Alleged AMD Radeon Pro W6000 Series RDNA 2 Graphics Cards For Apple Mac Pro Workstation Pictured & Benchmarked

 

[byteminer]

Please try this modified kext. Just need to replace the file and reboot.

No change, unfortunately.

I also tried the device-id override via DeviceProperties, but that, too, did not deviate. The result is the same as the FakePCIID approach.
Strangely, for the DeviceProperties approach, even though I specified the correct device-id in config.plist (I think), it still shows the original one in IOReg once booted up.

<key>PciRoot(0x2)/Pci(0x0,0x0)/Pci(0x0,0x0)/Pci(0x0,0x0)/Pci(0x0,0x0)</key>
<dict>
    <key>device-id</key>
    <data>v3MAAA==</data>
    <key>model</key>
    <string>AMD Radeon 6900 XT</string>
</dict>

 

[TheBloke]

Are you testing this with or without WhateverGreen, @byteminer ? If with WEG, I would think you'd need to have the FakePCIID entries before WEG in config.plist Kexts -> Add; is that the case?

If that's been tested then if it were me I'd test without WEG, just to be sure it wasn't interfering in some way. (I don't have a specific reason to think it will, just that's something I'd test to be sure.) That would require using MacPro 7,1 SMBIOS though.

 

[macdz]

finally got it working. did took a long cut of downloading and installing the recovery of CATALINE 10.5.7 and then updating via beta profile. but its here, finally.

People should remember to leave their system do the work if they have not added the BOOT ARGS: "agdpmod=pikera" the installation happens on its own. when you don't see the mac Installation and only see mac as boot option in the open core menu, you know that its good to go. now you can either add the option and you will start getting the display as mentioned before.

The only issue I have is that I find the brigthness low even after 100 % at hardware level. what can the be issue here ?

 

[CaseySJ]

No change, unfortunately.

I also tried the device-id override via DeviceProperties, but that, too, did not deviate. The result is the same as the FakePCIID approach.
Strangely, for the DeviceProperties approach, even though I specified the correct device-id in config.plist (I think), it still shows the original one in IOReg once booted up.

<key>PciRoot(0x2)/Pci(0x0,0x0)/Pci(0x0,0x0)/Pci(0x0,0x0)/Pci(0x0,0x0)</key>
<dict>
    <key>device-id</key>
    <data>v3MAAA==</data>
    <key>model</key>
    <string>AMD Radeon 6900 XT</string>
</dict>

macOS 11.4 actually supports three NAVI 21 device IDs:

• 73A2 1002 = Navi 21 GL ?
• 73AB 1002 = ??
• 73BF 1002 = Navi 21 [Radeon RX 6800/6800 XT / 6900 XT]

Source: TechPowerUp

We can try spoofing to 0x73AB, which might be a better match. The attached ZIP contains the changes:

• Modified SSDT (SSDT-Change-GPU-FakePCIID.aml)
• Modified kext (FakePCIID_RX6000_Graphics.kext)

After this we'll have to go back to the proverbial drawing board...

 

[byteminer]

Are you testing this with or without WhateverGreen, @byteminer ? If with WEG, I would think you'd need to have the FakePCIID entries before WEG in config.plist Kexts -> Add; is that the case?

If that's been tested then if it were me I'd test without WEG, just to be sure it wasn't interfering in some way. (I don't have a specific reason to think it will, just that's something I'd test to be sure.) That would require using MacPro 7,1 SMBIOS though.

I've been testing on iMacPro1,1 with WEG, but I also just did a test run on MacPro7,1 without WEG. The result is the same between the two configurations. FakePCIID was loaded before WEG, right after Lilu.

 

[CaseySJ]

I've been testing on iMacPro1,1 with WEG, but I also just did a test run on MacPro7,1 without WEG. The result is the same between the two configurations. FakePCIID was loaded before WEG, right after Lilu.

We can even move FakePCIID kexts above Lilu, but not sure if that by itself will make any difference.

 

[macdz]

Here's the results of a couple of Metal GeekBench 5 scores on my:

Asus AREZ 8GB Dual OC RX580:
View attachment 516160

And my Sapphire Nitro+ RX6800:
View attachment 516162

Quite an improvement

How am I getting more on the same GPU ?

 

[byteminer]

Well, I've moved it up just in case.

I did test the 0x73AB just now and things have indeed changed slightly.
Instead of booting to the OS normally (but with no driver loaded), I'm now running into a black screen right as it would normally go to the login screen. I also saw the well-known purple lines appear for a split-second.
Will test the same setup with MacPro and no WEG now, since the black-screen thing may be related to that.

 

[TheBloke]

How am I getting more on the same GPU ?

GeekBench is a pretty poor benchmark tool. It doesn't test long or hard enough, so it's very susceptible to background activity and variations in circumstance. You could say it's got a poor signal to noise ratio.

My best GB 5.4 Metal score is 183k. My lowest is 160k. That's a 15% difference. I've seen the same in CPU tests on multiple systems, both in macOS and Windows.

GB gives a general indication, and is fine for quick tests and bragging rights, but it shouldn't be relied upon for accurate data. Best to assume any result has +/- 20% margin of error