[Success] AMD RX6000 Series working in macOS

[baratta930]

I suggest updating to modern GeekBench, ensuring the version you have is the same on both (!) and then redoing your test.

That said, you’re getting 1/2 what you should be getting. (Edit: Nope; I am thinking of metal; can you share your metal score pls?) What happens in 13.2.1 ?

even later edit: 186k in GB6 metal. That’s odd; I’m getting 224k in GB6 metal on a 6900xtxh.

Both Windows and Mac were using Geekbench 5 but not exact same release - Windows 5.4 and Mac 5.5.1. I can re-run tonight when I'm back in front of my rig.

Metal score

Geekbench 6 scores higher than 5. I can re-run all the tests using GB6 but I'm assuming I'll see a similar difference.

Is this before or after first sleep?

I tried before and after sleep. No change.

When I had my Vega56, my scores were nearly identical between Mac and Windows and in line with expectations for the card.

Thanks everyone for the suggestions, keep them coming!

Berardino

 

[zappy39]

 

[dclive]

View attachment 563793

With what GPU?

Separately, I suggest that when you see this GPU issue that you also (in GB6; let’s all flip to the modern version…) run several CPU tests and see if your CPU scores similarly move around, to try to isolate the issue to ‘just GPU’ vs. both.

I’ve not noticed mine varying.

 

[Sabzz]

Geekbench 6 scores higher than 5. I can re-run all the tests using GB6 but I'm assuming I'll see a similar difference.

I just tested both Win11 and Monterey 12.6.2, same result, Windows has a higher score.

run several CPU tests and see if your CPU scores similarly move around

I'll run more tests but the one i did, showed me a slight increase in single core performance but quite a hit in multi-score when compared to GB5.
GB5 Multi - 25k
GB6 Multi - 16k. <--?

 

[dclive]

One cannot compare across versions, the intent is for you to run gb6 in both platforms and compare there.

 

[Sabzz]

One cannot compare across versions, the intent is for you to run gb6 in both platforms and compare there.

Tested GB6 across both platforms, same version, 3 tests on each one after another, Metal x3 , OpenCL x3, Cpu x3 (Mac), Vulcan x3, OpenCL x3, Cpu x3 (Win11), so 18 tests all together.

Results were very similar with little variation to each other (as you stated). More noticeable between OS's.

My Results;

OpenCL

Vulcan/Metal

Note: i updated to the latest AMD driver (released Feb 2023) and now my metal scores are staying around 180k mark. I think there's something going on with my card or config, according to GB6 results, i should be reaching over 200k on GB6. Anyone have any ideas?

Metal Benchmarks - Geekbench Browser

Thanks

 

[root2]

I'm one of the unlucky guys who got a RX 6700 XT, which is not supported by macOS. I dual boot Windows (for gaming) and macOS (for work). Thus, I also bought a secondary GPU just for macOS a RX 560. Back then (in March 2021), it was the only option. Now with the introduction of the 7900 XT in Germany, the GPU prices are significantly dropping:

RX 6600 XT: 265€
RX 6800 XT: 600€
RX 6900 XT: 690€

RX 7900 XT: 849€ - not supported by macOS

So my idea is to replace my dual-GPU setup, for better desktop performance (I use a 2x 4k monitors). My questions are:

1. Should I wait for 7900 XT driver support that might never come?

2. Is there any reason to panic, because sooner or later the RX 6800/6900 XT will be sold out and the only option to get one will be used from eBay? Or should I just wait and chill, the prices will continue to decrease.

3. Should I get a 6900 XT because it is just 90€ more? Is it worth it, considering the higher power consumption?

 

[miliuco]

@root2
I don't expect support for the 7000 serie but who knows, maybe Apple has different opinion.
Only 90€, I prefer 6900 XT, but really both 6800 and 6900 are great cards for macOS (and Windows).

 

[jkbuha]

You worked it out already, but as reference for others:

Always start with \_SB as base ('\': absolute root; leading '_': name reserved by ACPI specification) for External and Scope.
Then every second item is an ACPI entity (4 character name, beginning with a letter from 'A' to 'Z') with a PCI address (after '@'); pick them in order to make the path: PCI0, GPP8, X161 (this one sounds like it comes from an AMD motherboard but these names are just arbitrary). The next object after that, '[email protected]' has an address but no proper ACPI name, and that what is to be fixed.

\_SB.PCI0.GPP8.X161 is our well-formed ACPI path.
First acknowledge it already exists so it can be used: External(_SB.PCI0.GPP8.X161, DeviceObj)
Then, let's put us within the context of this object : Scope(\_SB.PCI0.GPP8.X161)
and define that the child Device at PCI sub-address 0 (Name (_ADR, Zero)) is called BRG0 (or whatever arbitrary name we fancy…) and has child GFX0 at address 0.

Pay attention to addresses! Adjust accordingly if the unnamed bridge, or the GPU behind it, is not at address 0 but at another value
@1: Name (_ADR, One)
@2: Name (_ADR, 0x2)
etc.
If there are multiple unnamed bridges ([email protected]/IOPP/[email protected]/IOPP/[email protected], as seen a few posts before), declare multiple nested devices (as in SSDT-BRG from the above post).

Hi all - I'm having an issue getting my 6650XT to be spoofed/recognised by the system.

My windows path is

ACPI(_SB_)#ACPI(PC00)#ACPI(TRP0)#ACPI(PXSX)#PCI(0100)#PCI(0000)#PCI(0000)#PCI(0000)

So I would expect the path to be \SB.PC00.TRP0.PXSX

However when changing the relative paths in the SSDT-BRG0.aml, I keep getting the ACPI AE_NOT_FOUND error (SB.PC00.TRP0.PXSX) when OpenCore tries to load the file.

 

[starghost1100]

@root2
I don't expect support for the 7000 serie but who knows, maybe Apple has different opinion.
Only 90€, I prefer 6900 XT, but really both 6800 and 6900 are great cards for macOS (and Windows).

I would expect something like the W6800X Duo for the mac pro 2019. with double the memory, two chips. it would be a great addition to the already existing line. otherwise i feel sorry for mac pro 2019 buyers who will be forced to build hackintoshes instead of buying a new super powerful dual card for a super huge $6000-8000.

P.S. Please, I want to be a little Vanga (I hope you know who Vanga is, right?) today and with this prediction