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

[Liber]

@Liber
maybe you should read the previous posts before you write something and open a new thread…

Maybe you are the one who should clearly state your thought, not just ask people read posts and not open a new thread.
Because I don’t see any inappropriate and my thread is just share the real test result in latest release beta MacOS and Bootcamp.
Make your point not just post a garbage message.

 

[s4ru]

Based on the information so far, I have created an interpolated chart for Octane bench. If these results are confirmed then the 6900XT is insanely fast taking account Octane X does not use hardware RT on Radeons at the moment.

View attachment 1762371

These are the numbers I gathered during 4 months wait.
I got pieces from here and there so please take it as a grain of salt.
Numbers might be off and outdated.




OB Score (Octane Bench Score)


=====Octane X @PR7=====
Radeon Pro Vega II Duo - 446 OB (dual ASIC GPU)
Radeon Pro Vega II - 223 OB
Radeon VII - 214 OB
5700 XT - 184 OB
Vega FE - ~174 OB
Vega 64 -~160 OB
Vega 56 - ~126 OB
Vega 48 - 117 OB
5600M - 117 OB
5500M - 83 OB
Vega 20 - ~55 OB
Apple M1 - 40 OB ???
Intel iris 640 - ~17 OB
Intel iris 630 - ~15 OB
Apple A13 - ~15 OB



===== OctaneBench 2020.1 =====
RTX 3090 x2 (RTX ON) - 1341 OB
RTX 3090 (RTX ON) - 659 OB
RTX 3080 (RTX ON) - 552 OB
RTX 3090 (RTX OFF) - 535 OB
RTX 3080 (RTX OFF) - 458 OB
RTX 3070 (RTX ON) - 403 OB
TITAN RTX (RTX ON) - 380 OB
RTX 3060 Ti (RTX ON) - 376 OB
Quadro RTX 6000 (RTX ON) - 376 OB
RTX 2080 Ti (RTX ON) - 349 OB
TITAN RTX (RTX OFF) - 345 OB
RTX 3070 (RTX OFF) - 316 OB
RTX 2080 Ti (RTX OFF) - 314 OB
RTX 2080 (RTX ON) - 256 OB
RTX 2080 (RTX OFF) - 244 OB
GTX 1080 Ti - 192 OB
GTX 980 Ti - 139 OB
GTX 1060 6GB - 81 OB




I tried a benchmark Chess scene on OctaneX PR7 with my Radeon Pro Vega II Duo, and got result of 9 seconds.
As Xenobius pointed out, I am also assuming an OB Score for 6900XT as 400-420-440ish, which in my opinion, is great in terms of cost and performance balance.

 

[The-Bloke]

Beppe has updated his Octane X graph with my 6900XT result:

 

[vslab]

Based on the information so far, I have created an interpolated chart for Octane bench. If these results are confirmed then the 6900XT is insanely fast taking account Octane X does not use hardware RT on Radeons at the moment.

View attachment 1762371

It is probably faster as the current OB benchmarks are very few and we don’t know the exact configurations of the system they are running on. Looking at the metal benchmarks gives a better understanding of the cards capacity I guess, it’s more in the range of the 3080 there.

 

[Xenobius]

It is probably faster as the current OB benchmarks are very few and we don’t know the exact configurations of the system they are running on. Looking at the metal benchmarks gives a better understanding of the cards capacity I guess, it’s more in the range of the 3080 there.

The 6900XT should be on the performance level of the RTX 3080/3090 (or even higher) if Octane X will use Radeons RT cores. Hopefully Apple will introduce support for hardware raytracing in their Metal drivers soon.

The performance difference when using RT cores can be very significant:

Techgage Image - OTOY OctaneRender 2020 - God Rays Project RTX Score (NVIDIA GeForce RTX 3090)

techgage.com

 

[renatocanton]

@DFP1989 Did you want me to try an HEVC test? Upload a file and I'd be glad to give it a go in FCP X.

@vslab I heard back from Beppe of OTOY that Octane Benchmark doesn't support Metal GPUs, and won't until later in the year. So that's out for the time being.

But he did give me an alternative benchmark, in Octane X: a "Chess Set Procedural" orbx file which he's been using to benchmark different GPUs.

I've got the 6900XT in my main system now, and I ran the benchmark as per his instructions.

Here's the competition:
View attachment 1762298

My result...

11 seconds!

View attachment 1762300

Very pleased

I can also report that multi-monitor is working fine with no hiccups noticed yet. I have four screens connected - 2 x DP, 1 x HDMI, 1 x USB-C to DP. One is 3840x2160 @ 60, the other three 1920x1200 @ 60. Sadly I don't have any >60Hz monitors so I can't test that.

Number of outputs is the only thing I don't like about this card. I bought both of my two previous AMD GPUs because they had six outputs. Not that I often used six, but I did sometimes use five. Oh well, can't have everything.

I did cheekily wonder what would happen if I connected a USB-C hub to the GPU, a hub that has multiple HDMI outputs. I'm 99% sure it won't work; best case I expect it will just give a mirrored picture on two screens, as macOS sadly doesn't support MST. But for the sake of £40 I might try it (can always send it back.)

8 seconds here!

 

[The-Bloke]

8 seconds here!

Nice result - but your picture looks different. I think you rendered a different target. Mine looks like this:

The target to render is "RenderTarget PT"

Attached is the file to open in Octane X - unzip it, and then double click it while Octane X is open (I found that if you double click it before opening Octane, it doesn't register it and opens the Demo project, which confusingly also contains a version of this scene.)

Then render "RenderTarget PT" near the bottom of the Scene list in the left pane

 

[renatocanton]

Nice result - but your picture looks different. I think you rendered a different target. Mine looks like this:
View attachment 1762895

The target to render is "RenderTarget PT"

Attached is the file to open in Octane X - unzip it, and then double click it while Octane X is open (I found that if you double click it before opening Octane, it doesn't register it and opens the Demo project, which confusingly also contains a version of this scene.)

Then render "RenderTarget PT" near the bottom of the Scene list in the left pane
View attachment 1762897

you were right, the same 11 seconds ..

 

[s4ru]

OK, I couldn’t resist installing Public Beta 11.4 Beta1 on my production machine, which is something one should avoid especially in the middle of a project…

Anyway, my 6900XTs are fully working now! Wow.

I was lucky enough to get some reference model 6900XTs and one 6800XT when price was lowest. My MacPro is loaded with Radeon Pro Vega II Duo, and two 6900XTs as show in the picture on post #87.

I was concerned about wattage consumption but as long as I monitor with external wattage checker when using Octane X, it seems OK so far.




Tried few tests;


- Octane X PR11 “Chess Scene”
Radeon Pro Vega II Duo (using both) 9sec
6900XT x1 11sec
6900XT x2 5sec
Radeon Pro Vega II Duo + Dual 6900XTs 3sec


Yes. Dual 6900XT is faster than Vega II Duo. Using all of them results 3sec, which I think hitting the ceiling because even if I attach another card via eGPU makes no difference. If I switched to heavier load by rendering with PMC kernel, the speed increase of adding eGPU can be seen.




- Geekbench 5.4.0
6900 XT no.1 189280
6900 XT no.2 187693
Radeon Pro Vega II Duo (single) no.1 81777
Radeon Pro Vega II Duo (single) no.2 101502




- BASEMARK GPU (MacOS Big Sur 11.4 Beta1)
6900 XT no.1 15404
6900 XT no.2 16019
Radeon Pro Vega II Duo (single) no.1 8184
Radeon Pro Vega II Duo (single) no.2 8458


- BASEMARK GPU (Windows 10)
6900 XT 18454






So far so good!
11.4 Beta 1 seems stable too.


I love my MacPro so much. Adding latest GPU can maximize its power further.

 

[thexash]

What model&brand are you using? I can get the Sapphire Nitro+ RX 6800XT or 6900XT where I am but I can find a clear dimension measurement... and even if there was I'm not too sure what kind of clearance/buffer I need to add to the cards' length. Looks like those card are 31cm in length from what I could find.

 

[s4ru]

I’m using reference cards, which is 267mm in length.
You can see my setup on post #87 and the size of space left is 38mm-40mm.
You need to think about how to manage power cables, of course.

 

[thexash]

I’m using reference cards, which is 267mm in length.
You can see my setup on post #87 and the size of space left is 38mm-40mm.
You need to think about how to manage power cables, of course.

Great thanks for the info. Only have the 580X atm and two small PCIE card for SSDs so cable management shouldnt be much a of problem with a 267mm card. The Sapphires (and the Asus I could find) are all 31-32cm though, which won't fit with the cable as far as I can tell. Oh well, I'll try to get one of the shorter one or wait for a MPX module I guess...

 

[Menno87]

Looks like apple is prepping a MPX module?

 

[thexash]

Looks like apple is prepping a MPX module?

We can hope, it might even be one of the first time in history when the Apple GPU price is competitive with the "market" price.

Where I am, getting a MPX 5700 is within a $100 of the second hand or scalped 5700 XT...

 

[tsialex]

Looks like apple is prepping a MPX module?

The photo is not of a MPX module, but from a standard PCIe GPU (seems to be the OEM version of Radeon Pro W6000) and the original Geekbench is from a hackintosh. It's very easy to spot, since no Mac Pro has Core i9 as a CPU, all real Mac Pros identify as Mac Pro (2019), a detail frequently overlocked by hackintoshers, and no other Mac has PCIe slots.

https://twitter.com/i/web/status/1385872837412593671

旧帖归档 - Chiphell - 分享与交流用户体验

旧帖归档Chiphell - 分享与交流用户体验 ,Chiphell - 分享与交流用户体验

www.chiphell.com

 

[Menno87]

Yeah I know, but it look's promising with the product naming. It could be used for a mpx module. Time will tell

 

[chumps]

Yeah I know, but it look's promising with the product naming. It could be used for a mpx module. Time will tell

The naming scheme is definitely for an MPX module. A Radeon's Pro product would be like W6900.

 

[sebh]

I couldn’t resist too. i installed the Beta on a pci M2 SSD. Seems stable so far and sleep is working again.

 

[ArPe]

Would like to see some real time ray tracing demo or game on macOS if there are any.

 

[thexash]

I couldn’t resist too. i installed the Beta on a pci M2 SSD. Seems stable so far and sleep is working again.

Do you use it with your display(s)? or only as a "compute" card for rendering?

 

[sebh]

Would like to see some real time ray tracing demo or game on macOS if there are any.

i just tried the Valley benchmark but its Open CL and old. 105,6 fps with the 6900XT 67,4 fps with W5700X

 

[sebh]

Do you use it with your display(s)? or only as a "compute" card for rendering?

with the displays. one time i connected my LG WQHD+ on the 6900XT via thunderbolt and my Benq SW271 via HDMI on the w5700W.
and the 2th time my LG WQHD+ to the w5700X via thunderbolt and my Benq SW271 via HDMI on the w5700W.

 

[diamond.g]

Do you know if Apples drivers raise the power limit on the reference 6900xt? OOB mine (windows pc) is limited to 255 watts, even though the card is supposed to be good for 300 (technically 375 if you include the PCIe slot).

Seems like the AIB cards would be much faster since they all have higher power limits.

 

[The-Bloke]

Do you know if Apples drivers raise the power limit on the reference 6900xt? OOB mine (windows pc) is limited to 255 watts, even though the card is supposed to be good for 300 (technically 375 if you include the PCIe slot).

Seems like the AIB cards would be much faster since they all have higher power limits.

I haven't looked at this yet, but am planning to try soon with injecting PowerPlay tables from a Windows OC into macOS to see what happens - see if I can spot any benchmark improvements. Same method has worked on previous AMD GPUs I believe.

I'm not quite sure how to monitor GPU power usage in macOS. iStatMenus doesn't seem to have an option for that - at least not with my 6900XT. Maybe because it's a new card? Not sure on that.

 

[iTTT]

my Valley benchmark looks not bad.