[Success] AMD RX6000 Series working in macOS

[Cyberneticist]

@TheBloke my current GeekBenchmark:

 

[Pavo]

What system did you run this on? Was it your TRX40, or another AMD with PCIe 4.0?

Was ran on my TRX40 system, its the only hackintosh I have now.

 

[Pavo]

I don't believe anyone has found a way to boot macOS with Resize Bar enabled as yet.

I have Resizable-Bar enabled in the BIOS and have Above4G disabled, if I enabled Above4G with Resizable-Bar enabled macOS doesn't boot. But with Resizable-Bar only enabled it boots.

 

[Pavo]

So... I have done some testing and it seems that macOS actually boots when enabling both Above4G and Resizable-Bar in the BIOS, but the GPU seems like its stuck and doesn't allow any display. I can screen share from another MBP and access the OS but the GPU isn't recognized at all. Most likely not a boot loader issue, probably a kernel or driver issue.

 

[imsuperimposed]

Is there anyone can test if "Resize Bar" works or not and if it works is there any performance difference?

Resizable bar requires above 4G decoding to actually be enabled and to function correctly article attached for a better explanation https://www.gameunion.tv/en/games/a...bioss-dark-horse-what-it-and-should-it-be-105
Basically it’s unlikely until apple supports above 4G decoding that resizable bar will work.

 

[TheBloke]

Was ran on my TRX40 system, its the only hackintosh I have now.

OK thanks. I think this confirms what I was saying to @oreoapple yesterday - the Blackmagic Raw benchmark is not testing the full compute power of the card because it is bandwidth limited on a PCIe 3.0 system.

He and I both get 125 FPS even though I have a 6900XT and he has a 5700XT. Then you get 175 FPS with a 6900XT, but you have PCIe 4.0.

And actually the maths seems to work out:

• An 8K video frame (7,680 by 4,320) @ 32 bits per pixel (4 bytes) is 132,710,400 bytes per frame.
• Times 125 FPS is 16,588,800,000 bytes per second (15.44 GByte/second)
  • The bandwidth of a 16x PCIe 3.0 slot is 16GB/s
• Times 175 FPS is 23,224,320,000 bytes per second (21.6 GByte/second)
  • The bandwidth of 16x PCIe 4.0 slot is 32GB/s

Assuming I am doing the correct calculations here, it suggests that at PCIe 3.0, BRAW debayer/decode is bandwidth limited to 125 FPS with 8K frames, but at PCIe 4.0 it is GPU limited and the limit of the 6900XT is 175 FPS.

175 FPS is not that impressive as a top speed for the 6900XT given a 5700XT can do (at least) 125 FPS, but I feel that this BRAW test is probably not a very good indicator of anything other than BRAW. Certainly not representative of 3D rendering, and hopefully not representative of other GPU accelerated tasks like noise reduction (though I guess NR with 8K frames would be similarly PCIe limited; at 4K though, which most people are still using, there'd be plenty of bandwidth.)

 

[byteminer]

Your RX 6900XT is at the following PCI path:

_SB.PC02.BR2A.SL05.PCI-Bridge@0.GFX0

But only first first 4 parts of the path are defined in the DSDT. The two parts in bold must be created by the SSDT in order to reach the correct final endpoint (the GPU). Only then can we change the Device ID.

The process is described in this post:

Gigabyte Z490 Vision D (Thunderbolt 3) + i5-10400 + AMD RX 580

It works. Thanks! So you used this path for construction of SSDT? IOService:/AppleACPIPlatformExpert/PCI0@0/AppleACPIPCI/IOU1@7/IOPP/PXS2@0/IOPP/pci-bridge@4/IOPP/GFX1@0 Glad to hear it. Just curious what is the logic? The thinking was as follows: Your Nvidia GPU was being attached to...

www.tonymacx86.com

I've made the necessary changes and attached both the AML and DSL files. The DSL file should have all the comments intact.

Thanks, this was actually very helpful, now I know how to do this in the future!

In slightly more unfortunate news, the ID spoofing did not lead to immediate success.
I was able to verify that the spoofing worked by spoofing only the device name, which worked fine by itself.
However, spoofing the device ID as well leads to a hangup during the boot process, at which point nothing happens anymore and the PC has to be shut down forcefully.

I assume there's a bit more to the binned cards than just a different device ID, though I haven't been able to turn up any information as to what the secret sauce may be.

I'll continue to experiment some more and maybe some of the subsequent betas will add support anyway, but for now the binned versions are a no-go, as far as I can tell.

 

[valvator]

I can get one RX6900 XT for € 1350, is it a good price? or better to wait ...

 

[msh]

I can get one RX6900 XT for € 1350, is it a good price? or better to wait ...

Wow, I can't get € 1350 for a month.

 

[CaseySJ]

I can get one RX6900 XT for € 1350, is it a good price? or better to wait ...

That’s a bargain in today’s crypto climate.