Contribute
Register

[Success] AMD RX6000 Series working in macOS

@OOBIDUB Unfortunately I have an update this morning on sleep/wake - having put the system to sleep overnight and woken it this morning, my USB-C connected monitor did not wake. That's a bit annoying.

Before this morning the longest sleep I had done was about 2 hours, and that worked fine. Seems odd that 2 hours would work and 9 hours wouldn't, so probably another factor.

A second difference is that when I went to bed last night, after putting the system to sleep I turned all my monitors off at the mains. Then I turned them back on from the mains this morning, prior to waking the system. Maybe that makes some difference on the USB-C port.

I'll do some more tests and report back. I've not yet tested without WEG, and WEG hasn't been updated for these new GPUs, so it's possible it's causing some problem (as noted with others reporting no >60ghz support with WEG running.)

Update 1: Just now I left my PC for 30 mins and followed this procedure:
  • Put system to sleep
  • Turned monitors off at mains
  • Came back 30 mins later
  • Turned monitors on at mains
  • Woke system up with mouse.
And all monitors connected fine, including USB-C.

Update 2: I did a further six quick sleep tests:
  • Tests 1 - 3: Put system to sleep, waited 30 seconds, woke it up
  • Tests 4 - 6: Put system to sleep, turned monitors off at mains, waited 30s, turned monitors on, waited 30s, woke system up
In all six tests, all four monitors connected.

So maybe it's time based, ie a long sleep causes problems where a short one does not? I don't know what the difference would be between sleeps of 2 and 9 hours, but maybe things can change after a certain period of time.

Or it could just be intermittent - like a 10% chance of it failing or something like that. But then it seems to happen every time for you?

It could also depend on the monitor, and/or on the USB-C to DP cable/adapter. This is what I'm using at the moment:
1619173520006.png


That's plugged into the USB-C port, and from there I have an 8K DP - DP cable to the monitor.
 
Last edited:
Beppe of OTOY has added my 6900XT result to his Octane X benchmark graph:

1619167939404.png


Pretty nice!

@oreoapple I think there must be something wrong with that BRST benchmark. The fact that it's topping out at 125 GPU frames the same on both 5700XT and 6900XT must mean there's some internal bottleneck.

Hopefully that's just a bottleneck in the tester, and not in Resolve as well. Though even if it is in Resolve, it hopefully only affects the BRAW decoding/debayering, which is what BRST tests. As opposed to all GPU accelerated functions.

We can see from the Octane X result that the 6900XT grinds the 5700XT into the dust in 3D rendering.

I'll try and do some Resolve benchmarking, though I only have a Vega 64 to compare it against. I wish Puget would publish their test suite. I said months ago that I had a theory that Pugets poor 6900XT result was down to the AMD cards only using OpenCL versus the NVidia's on CUDA. Given the huge disparity in Metal vs OpenCL figures in our preliminary 6800/6800XT/6900XT benchmarks, I'm hoping to see that the AMD cards on Metal are a match for a 3080 on CUDA. But I'm not yet quite sure how to get a repeatable benchmark in Resolve.
 
The USB-C port on the 6800xt can be connected to a thunderbolt4 dock... you don't get thunderbolt on the downstream thunderbolt ports on your dock; only USB. But you also get displayport/hdmi output working well.

This can be an option for those with a dock and a motherboard with thunderbolt. If you want to use the discrete GPU video output with you dock and your motherboard doesn't have a displayport input for you tunnel your video card's video over thunderbolt, then just plug your dock into the 6800/6900.
This is very interesting, and something I've wanted to test. The only thing that bugs me about the 6000-series cards is that they only have four monitor outputs. Both of my previous GPUs had six. I didn't use six monitors, but I did sometimes use five.

I was thinking maybe I would buy a USB-C dock with 2 x HDMI or DP + HDMI output, and connect it to the USB-C port on the 6900XT, and just maybe I would get five monitor outputs total?

Are you saying that you know this works? Does it have to be a TB4 dock? That'd be rather expensive just to get an extra monitor output.

I just tested my CalDigit TB3 Dock - I plugged it into the USB-C on the 6900XT, then I plugged a monitor into the TB3's DP output. The monitor briefly got a signal, but no picture, and then the signal was lost. Not surprising I suppose. Also I did that with my TB3 cable - I don't seem to have a plain USB-C to USB-C - so maybe that wouldn't work anyway (I don't know if a TB3 cable acts as USB-C when not connected to TB on one end?)

If this is something you've tested I'd love to hear more about what does and doesn't work, and whether you think this could work with a standard USB-C dock rather than TB4. My fear was that it couldn't work because macOS doesn't support MST, so even if a USB-C dock could connect two monitors, they'd both show the same picture.
 
Last edited:
Screen Shot 2021-04-23 at 10.39.31 pm.png


Screen Shot 2021-04-23 at 11.10.06 pm.png


PowerColor 6900XT RedDevil working beautifully with the new Beta
We have Metal in PP
Just tested my worst case editing scenario. Raw CinemaDNG 6k from an external SSD (Samsung T5) at full resolution with Luts and effects. Flawless playback. I am happy
 
Last edited:
Beppe of OTOY has added my 6900XT result to his Octane X benchmark graph:

View attachment 516026

Pretty nice!

@oreoapple I think there must be something wrong with that BRST benchmark. The fact that it's topping out at 125 GPU frames the same on both 5700XT and 6900XT must mean there's some internal bottleneck.

Hopefully that's just a bottleneck in the tester, and not in Resolve as well. Though even if it is in Resolve, it hopefully only affects the BRAW decoding/debayering, which is what BRST tests. As opposed to all GPU accelerated functions.

We can see from the Octane X result that the 6900XT grinds the 5700XT into the dust in 3D rendering.

I'll try and do some Resolve benchmarking, though I only have a Vega 64 to compare it against. I wish Puget would publish their test suite. I said months ago that I had a theory that Pugets poor 6900XT result was down to the AMD cards only using OpenCL versus the NVidia's on CUDA. Given the huge disparity in Metal vs OpenCL figures in our preliminary 6800/6800XT/6900XT benchmarks, I'm hoping to see that the AMD cards on Metal are a match for a 3080 on CUDA. But I'm not yet quite sure how to get a repeatable benchmark in Resolve.
Agree. 6900xt should be about 2x 5700xt in some tasks if not all.
 
11.4 Is now a public Beta. Go get em' boys
So do we now full access to metal and all the goodies on this version ? Have a Sapphire Nitro+ 6800XT. How stable is this version ?
 
So do we now full access to metal and all the goodies on this version ? Have a Sapphire Nitro+ 6800XT. How stable is this version ?
Yes, fully working OpenCL and Metal. Metal is up to 50% faster in GeekBench than OpenCL, a much bigger discrepancy than I've seen on my previous GPUs.

I've been using 11.4 on my main machine for 24 hours and no problems noticed so far. Some people are reporting USB problems, so make sure you have a proper USB Map. And I just learned from someone with an I225-V NIC that they're getting KPs as soon as they plug in an ethernet cable.

Best option for any new macOS release on a Hack - especially a beta that's only been out a couple of days - is to do the upgrade first on a secondary SSD/HDD, or at least to make a bootable backup of your working install before updating the main drive. So you can go back if it becomes unusable.

That said, plenty of people are using 11.4 with no problems so chances are it'll be OK for you too.
 
The only problem I have is the size of the mouse cursor when using HiDPI modes other than 2:1. It's way too big, but besides that, everything works flawlessly. Also, I just discovered a refreshingly new use case: WoW in 8K. I strongly believe and can not think otherwise, that's what Apple had in mind when they provided the drivers for Big Navi. :lol:
 
The only problem I have is the size of the mouse cursor when using HiDPI modes other than 2:1. It's way too big, but besides that, everything works flawlessly. Also, I just discovered a refreshingly new use case: WoW in 8K. I strongly believe and can not think otherwise, that's what Apple had in mind when they provided the drivers for Big Navi. :lol:
What "resolution" & UI "look like" are shown in Sytem info ➧ Display ?
 
Back
Top