MP All Models AMD Radeon RX 6800 XT / 6900 XT for macOS?

[joevt]

Thanks, I will not get 5k dot per dot but it will be 4K upscale to 5K (I understand).

LG 5K 27MD5KL requires two tiles of 2560 x 2880 to do full 5K. Since USB-C to DP bidirectional cable does not have enough bandwidth, I should not be able to get 2560 x 2880 HiDPI (Pixel Doubling) with the USB-C bidirectional cable to DP directional.

Max resolution should be 1920 x 1080 HiDPI with the 6800XT / 6900XT with the USB-C bidirectional cable to DP directional unless I`m mistaken.

You are not limited to 1920 x 1080 HiDPI even if you decide to go with 4K 60Hz instead of 5K 39Hz. You can choose 2560x2880 HiDPI which is 5120x2880. If the output resolution is 4K then the GPU will scale 5120x2880 down to 4K and the display will scale that up to 5120x2880.

Choosing 1920x1080 HiDPI isn't going to be much of an improvement over 2560x2880 HiDPI since it will need to be scaled from 4K up to 5K anyway.

Basically, the GPU should be able to do up to 16K x 16K (I've tried this with a W5700) and scale that to whatever the output resolution is.

 

[Grumply]

Has anyone come across any Davinci Resolve benchmarks that people have run on Macs with the 6900XT yet? Puget Systems benchmarks on Windows seem fairly average/underwhelming, but would the new card's performance in Metal offer better overall performance than you could get on PC?

I currently have a pair of Radeon VIIs in my 7,1 (which are undeniable beasts). But the Radeon VIIs are selling for such ludicrous prices on ebay at the moment, that I could upgrade to 6900XTs quite painlessly at the moment.

As I understand it, Davinci uses FP32 for video processing - and the 6900XT has a lot more FP32 computing grunt than the Radeon VIIs do. I'd love to know if this will give me a substantial FPS boost when grading (if I upgrade to the newer cards).

Or, is the superior memory bandwidth and speed of the HBM2 in the Radeon VIIs going to give them an edge over the newer cards? If the difference in performance is going to be negligible, I'm certainly not going to bother going to the effort of changing cards.

Real-time playback performance in my Resolve timelines (with both Prores and various raw codecs) is really the only thing I care about (export times don't really bother me).

 

[Grumply]

Thermals have been good so far with two of them

Are the two you have 2.5 slot high models? And if so, do they block the PCIe slots directly above them when mounted (in the 2-slot x16 positions)?

 

[kennyman]

@joevt

Thank you for the detailed info. That`s exactly what I wanted to know.

 

[rondocap]

Are the two you have 2.5 slot high models? And if so, do they block the PCIe slots directly above them when mounted (in the 2-slot x16 positions)?

Yes 2.5 slot models, they do block the PCIE slots just like an MPX module would, have them in the MPX slots and an afterburner in slot 5.

 

[rondocap]

Has anyone come across any Davinci Resolve benchmarks that people have run on Macs with the 6900XT yet? Puget Systems benchmarks on Windows seem fairly average/underwhelming, but would the new card's performance in Metal offer better overall performance than you could get on PC?

I currently have a pair of Radeon VIIs in my 7,1 (which are undeniable beasts). But the Radeon VIIs are selling for such ludicrous prices on ebay at the moment, that I could upgrade to 6900XTs quite painlessly at the moment.

As I understand it, Davinci uses FP32 for video processing - and the 6900XT has a lot more FP32 computing grunt than the Radeon VIIs do. I'd love to know if this will give me a substantial FPS boost when grading (if I upgrade to the newer cards).

Or, is the superior memory bandwidth and speed of the HBM2 in the Radeon VIIs going to give them an edge over the newer cards? If the difference in performance is going to be negligible, I'm certainly not going to bother going to the effort of changing cards.

Real-time playback performance in my Resolve timelines (with both Prores and various raw codecs) is really the only thing I care about (export times don't really bother me).

I've been testing in both FCPX and Resolve. It seems for metal/Resolve, in many cases the single 6900 XT seems to be just as fast or faster than the Vega ii duo, or at least close, which is really impressive considering the price difference.

I have:

Dual W5700X MPX
pro Vega ii MPX
6800 XT and the 6900 XT reference models

An example of the candle test in Resolve with the 6900 XT:

6900 XT
1 NR 2 60 FPS
3 NR 2 48 FPS (Vega ii DUO 43 FPS, Vega ii 21 FPS)
6 NR 2 25 FPS (Vega ii DUO 21 FPS, Vega ii 11 FPS)

04 Blur Nodes 60 FPS
08 Blur Nodes 60 FPS
16 Blur Nodes 60 FPS
32 Blur Nodes 60 FPS
64 Blur Nodes 37 FPS (Vega ii DUO 50.4 FPS, Vega ii 25 FPS)


6900 XT + 6800 XT together, almost perfect
1 NR 2 60 FPS
3 NR 2 60 FPS
6 NR 2 60 FPS

04 Blur Nodes 60 FPS
08 Blur Nodes 60 FPS
16 Blur Nodes 60 FPS
32 Blur Nodes 60 FPS
64 Blur Nodes 49 FPS

 

[Grumply]

I've been testing in both FCPX and Resolve. It seems for metal/Resolve, in many cases the single 6900 XT seems to be just as fast or faster than the Vega ii duo, or at least close, which is really impressive considering the price difference.

I have:

Dual W5700X MPX
pro Vega ii MPX
6800 XT and the 6900 XT reference models

An example of the candle test in Resolve with the 6900 XT:

6900 XT
1 NR 2 60 FPS
3 NR 2 48 FPS (Vega ii DUO 43 FPS, Vega ii 21 FPS)
6 NR 2 25 FPS (Vega ii DUO 21 FPS, Vega ii 11 FPS)

04 Blur Nodes 60 FPS
08 Blur Nodes 60 FPS
16 Blur Nodes 60 FPS
32 Blur Nodes 60 FPS
64 Blur Nodes 37 FPS (Vega ii DUO 50.4 FPS, Vega ii 25 FPS)


6900 XT + 6800 XT together, almost perfect
1 NR 2 60 FPS
3 NR 2 60 FPS
6 NR 2 60 FPS

04 Blur Nodes 60 FPS
08 Blur Nodes 60 FPS
16 Blur Nodes 60 FPS
32 Blur Nodes 60 FPS
64 Blur Nodes 49 FPS

Thanks, I’ve seen the candle benchmarks, and the numbers are very impressive. I wonder about more conventional situations though. 1080p isn’t going to stress the GPU’s slower memory like a 4k/6k timeline would.

I’m wondering if this is the reason Puget’s benchmark results were so underwhelming (compared to the really impressive candle numbers)

 

[diamond.g]

Thanks, I’ve seen the candle benchmarks, and the numbers are very impressive. I wonder about more conventional situations though. 1080p isn’t going to stress the GPU’s slower memory like a 4k/6k timeline would.

I’m wondering if this is the reason Puget’s benchmark results were so underwhelming (compared to the really impressive candle numbers)

The silly part is the 6900 has less TFLOPS than the Vega 2 Duo. I know Apple is being practical, but RDNA really isn't designed for compute tasks, so it doing so well is neat. CDNA absolutely trashes RDNA1/2 in compute performance. In this case RDNA2 Infinity Cache isn't helpful for compute workloads (if we believe reddit) so yeah the lack of bandwidth hurts.

 

[rondocap]

Thanks, I’ve seen the candle benchmarks, and the numbers are very impressive. I wonder about more conventional situations though. 1080p isn’t going to stress the GPU’s slower memory like a 4k/6k timeline would.

I’m wondering if this is the reason Puget’s benchmark results were so underwhelming (compared to the really impressive candle numbers)

I’ve been doing a lot of export tests in resolve and fcpx too, everything from red raw, pro res raw, canon raw, h264, h265, etc, for the most part the 6900xt is fastest, worse case it’s the same as when h264 is more limited by itself and not the hardware.

 

[Grumply]

I’ve been doing a lot of export tests in resolve and fcpx too, everything from red raw, pro res raw, canon raw, h264, h265, etc, for the most part the 6900xt is fastest, worse case it’s the same as when h264 is more limited by itself and not the hardware.

By any significant margin, or just by a bit?

 

[rondocap]

By any significant margin, or just by a bit?

2 6900xt vs 2 w5700x, and single Vega ii:

1. red 6k raw 5 min 45 sec long clip to pro res 422 hq,
2x 6900 xt about 2 min 28 sec to export,
2x w5700x 3 min 5 sec,
1x Vega ii 5 min 17 sec

but with some codecs, there isn’t much difference if the hardware isn’t the limitation, like h264

in general, other gpu affects are faster too on the 6900xts aside from exporting

it’s really fast for pro res Raw too, but I also use an afterburner for that - which brings them closer, but without afterburner fastest times similar to red 6k above with 6900xt still in the lead

I’d like to test with the Vega Ii duo or a second Vega iI, but no way cost to performance will beat the 2x 6900xt if we‘re just talking metal/video speed

also 28 cpu cores really helps, they are constantly in use with a lot of raw files especially.

 

[riggieri]

Want to add a data point on PCI-E card size. Sapphire Nitro+ 6900XT is listed as 310mm and it will not fit. So either the car is longer or I think we have to be under 300mm to fit.

 

[diamond.g]

So who is going to spring for the 6900XT LC?

 

[fireedo]

So who is going to spring for the 6900XT LC?

where to put the radiator?

 

[diamond.g]

where to put the radiator?

Zip tie to the case fan!!

 

[rondocap]

So I have 2 6900xt, and works great. But…I have a Vega ii as well, and for some reason I almost want to use 2 Vega ii, or a duo. The 6900xts are faster for metal, but potentially at higher resolutions the Vega ii duo may still be great too.

not completely logical, but I really like the mpx modules lol

 

[rondocap]

We are in the midst of a weird transition period on the 7.1 Mac Pro. 6000 series AMD drivers are out, but no MPX yet, and we still officially have the w5700x and Vega ii as the best options.

In 2021, I had these questions- they’ve come up, but mostly when the Vega ii came out, so we need an updated view.

from the perspective of a video workflow, raw, pro res, etc.

1. Is the vega ii still worth it? If you got two, would you go for the duo or two separate vega ii?

2. What limitations do you feel the older architecture imposed on your workflow vs what it can bring?

Here are my thoughts, I’m testing these setups:

1. dual w5700x
2. Single Vega ii (need a second to match)
3. Dual 6900 XT

In terms of MPX, the 5700s are really good. Very fast, and modern encoders vs vega ii.
The only time I noticed the Vega ii has an advantage is when the vram can be used more, like in 8K- that’s why I think duo Vega ii may still be interesting.

having said that, the 2 6900xt are next level performance already for the most part. They win in metal performance, and in many areas are superior.
I would imagine a w6800 mpx with 32gb of vram to be ideal

having said that, I’m still interested in adding that second vega ii as I feel it is also very powerful in many situations.
Thoughts?

 

[renatocanton]

Can anyone see the temperature of Radeon RX 6800/6900 XT on Mac OS?

 

[skippermonkey]

Which make and model of the 6900XT fits inside the Mac Pro that people would recommend?

 

[diamond.g]

Which make and model of the 6900XT fits inside the Mac Pro that people would recommend?

IIRC the reference models fit easily.

 

[DFP1989]

Which make and model of the 6900XT fits inside the Mac Pro that people would recommend?

Gigabyte GAMING fitted perfectly in mine, would be more easy to get than a reference card these days I'd expect.

 

[skippermonkey]

I've always been happy with Sapphire cards (currently have two RX 5700XTs) – what about them?

 

[thexash]

This is my list (extracted from my notes) about the card sizes, not a lot that will fit in the Mac Pro unfortunately:

• Card has to be under 300mm to fit.
• Reference RX 6800 / 6900 XT are 267mm and fit easily.
• ASRock RX 6800 Challenger Pro: 305x131x55mm, might just fit
• Gigabyte Radeon RX 6800 XT & RX 6900 XT 16G: L=267 W=120 H=49 mm, should fit
• Gigabyte Radeon RX 6800 XT & RX 6900 XT GAMING OC 16G: L=286 W=118 H=58 mm, should fit
• AORUS Radeon™ RX 6800 XT AORUS MASTER TYPE C 16G: L=324 W=140 H=60 mm, TOO BIG
• PowerColor Radeon RX 6800 XT - Red Devil: 320mm*135mm*62 mm - TOO BIG: https://forums.macrumors.com/threads/amd-radeon-rx-6800-xt-6900-xt-for-macos.2262295/post-29751958
• Sapphire Nitro+ 6900XT ~2500 SGD: 310(L)X 134.3(W)X 55.3 (H)mm - TOO BIG - the 32cm PowerColor RedDevil is too big. From MacRumors forum: "Sapphire Nitro+ 6900XT is listed as 310mm and it will not fit. So either the car is longer or I think we have to be under 300mm to fit."
• Sapphire Nitro+ 6800XT ~1899 SGD: 310(L)X 134.3(W)X 55.3 (H)mm - TOO BIG - the 32cm PowerColor RedDevil is too big
• Asus Tuf Gaming RX 6900 XT: 32 x 14.02 x 5.78 Centimeter: TOO BIG
• Asus Tuf Gaming RX 6800 XT: 32 x 14.02 x 5.78 Centimeter: TOO BIG

The Sapphire cards are too big unfortunately, those are the ones I can buy here

Getting the reference cards - which seem to be the only ones that fit for sure - is almost impossible right now...

 

[johnnymcc]

Trying to update my adrenaline drivers in Boot Camp to the latest version - but it says no AMD GPU detected, even though it's there! Is it because my 5700 MPX module is also installed?

 

[diamond.g]

Trying to update my adrenaline drivers in Boot Camp to the latest version - but it says no AMD GPU detected, even though it's there! Is it because my 5700 MPX module is also installed?

Have you tried unplugging it and seeing if Radeon Software sees what is left over?