MP 7,1 Need your input! What benchmarks and tests would you like to see on Mac Pro with 6900XT?

[rondocap]

I would like your input, I am testing a Mac Pro with a few new GPU configurations, and I would like for you to tell me what you would like me to test.

I’ve already done a lot of testing in final cut pro, as well as da Vinci resolve, but I can do more specific ones and whatever resolution or codecs that you would like to see as well. Raw, different cameras, etc.

I would also love suggestions for other non-video editing type of benchmarks I can do, stuff like 3-D rendering, machine learning, or anything else that GPUs can be used for.

Let me know specific requests and I would love to do it!

Final goal is to put together all of the benchmarks in one place, that way people can have an idea of performance as the years go on since I think a lot of people will start to get into this Mac Pro as it gets cheaper.

Specs:
28 core Xeon w3275m
96gb ram
Afterburner

Available gpus to compare:

2x w5700x mpx
2x 6900 XT
1 Vega ii mpx
(580x mpx too if anyone wants numbers)

 

[hifimac]

RedShift and Octane test scenes and benches would be great! Curious how 2x 6900xt match up against Nvidia cards.

 

[Grumply]

Would be fantastic to see the numbers you get for this updated (6k) "candle" benchmark in Davinci Resolve:

Standard Candle UHD: Rocket Science

Ladies and Genitals, I've been curious about this for some time. Instead of doing a complete restart, I just wanted to see the Standard Candle benchmark in UHD. For those curious and likeminded, you can download the project called 'Rocket Science' below. Similar to the previous project, it's...

liftgammagain.com

Ideally from each one of your GPU combos:

1x 6900xt
2x 6900xt
1x Vega II MPX
1x W5700X
2x W5700X

Some people have already posted various numbers up, but from wildly different specced 7,1s, so it's a bit hard to draw many clear conclusions from them thus far.

 

[rondocap]

Would be fantastic to see the numbers you get for this updated (6k) "candle" benchmark in Davinci Resolve:

Standard Candle UHD: Rocket Science

Ladies and Genitals, I've been curious about this for some time. Instead of doing a complete restart, I just wanted to see the Standard Candle benchmark in UHD. For those curious and likeminded, you can download the project called 'Rocket Science' below. Similar to the previous project, it's...

liftgammagain.com

Ideally from each one of your GPU combos:

1x 6900xt
2x 6900xt
1x Vega II MPX
1x W5700X
2x W5700X

Some people have already posted various numbers up, but from wildly different specced 7,1s, so it's a bit hard to draw many clear conclusions from them thus far.

I’ve done a few of those if you see on that page already, but I’ll try to do a few more

 

[riggieri]

NeatBench! And 8K R3D performance…

 

[rondocap]

NeatBench! And 8K R3D performance…

Sure, and right now I am doing read 6K komodo to pro res 422 HQ.

In Resolve it almost maxes out the 28 core cpu, and of course the gpu too.
20 min 6k clip took about 13 min 28 sec to pro res 422 hq from red raw

 

[rondocap]

RedShift and Octane test scenes and benches would be great! Curious how 2x 6900xt match up against Nvidia cards.

In the chess set procedural, render target PT (2), with two w5700x the time is about 10 seconds, that seem about right? Just want to make sure I'm doing the correct benchmark

 

[hifimac]

In the chess set procedural, render target PT (2), with two w5700x the time is about 10 seconds, that seem about right? Just want to make sure I'm doing the correct benchmark

This is the graph floating around on the OTOY forums. With a single 5700xt clocking in a 21 secs, 11 secs for 2x sounds about right.

 

[rondocap]

This is the graph floating around on the OTOY forums. With a single 5700xt clocking in a 21 secs, 11 secs for 2x sounds about right.
View attachment 1794442

Yep that’s what I figured, saw that graph. I will run more tests as well. The usual numbers I see for octane are something like 406 for Vega ii duo, is that in a different program? How do I get those numbers or is it not possible on Mac? Windows?

 

[riggieri]

Sure, and right now I am doing read 6K komodo to pro res 422 HQ.

In Resolve it almost maxes out the 28 core cpu, and of course the gpu too.
20 min 6k clip took about 13 min 28 sec to pro res 422 hq from red raw

You need to enable the GPU debayer and decompression in preferences to test R3D Decode.
You shouldn’t use any CPU.
For me, on a 16 core I use the CPU for everything 5K and below, GPU for 6K and above.

 

[rondocap]

You need to enable the GPU debayer and decompression in preferences to test R3D Decode.
You shouldn’t use any CPU.
For me, on a 16 core I use the CPU for everything 5K and below, GPU for 6K and above.

I just did that with 6k Red footage - it was faster with debayer only. With debayer and decompression, was slower.

But this is on a 28 core CPU too - so that may make a difference as in debayer only the CPU itself will be used more.

I read that in some cases, like on a 9900k cpu, it was faster with both on - and in other cases, just with debayer on more powerful cpus

What do you think?

 

[rondocap]

You need to enable the GPU debayer and decompression in preferences to test R3D Decode.
You shouldn’t use any CPU.
For me, on a 16 core I use the CPU for everything 5K and below, GPU for 6K and above.

I tried those settings again, and seems like I got some improvement from 6k red raw to pro res 422 HQ - 13 min down to 9 min with debayer and decompression

not sure why it went slower the first time, possibly my camera raw settings weren’t set to red default

 

[DFP1989]

@rondocap I’d be really interested in your experience with dual W5700X in Final Cut.

So far I’m not really getting a useful improvement for my workflow with the 6900 XT editing HECV 4:2:2 10 bit.

I assume with the matching cards both work together in FCP like the dual cards in a 6,1?

 

[rondocap]

@rondocap I’d be really interested in your experience with dual W5700X in Final Cut.

So far I’m not really getting a useful improvement for my workflow with the 6900 XT editing HECV 4:2:2 10 bit.

I assume with the matching cards both work together in FCP like the dual cards in a 6,1?

Yes the matching cards both work together, right now i also have matching 6900xt as well and they both work.

is your 10 bit from a canon r5 or which camera? Let me know the exact settings and size of clip, resolution, and maybe I can run some tests with it

what’s the biggest problem you’re having, playback or export time?

 

[DFP1989]

@rondocap Yeah R5 and 1DX III. Playback is my issue, it’s just rubbish unless I transcode everything.

4096x2160 at 25p is my typical shooting mode on those cameras. Shot or long clip doesn’t matter, it basically stutters the moment you hit play.

My M1 Mac can play it perfectly, though it has other limitations once a project gets more complex (especially come export time).

At this point in time the M1 is not ready to replace Intel for me for my main machine, but I’m not getting the performance I’d be expecting for from the 7,1 for my workflow either (with or without the 6900 XT).

 

[diamond.g]

This is going to be a dumb question, but does macOS use the hardware HEVC decoder on the 6900XT?

 

[DFP1989]

This is going to be a dumb question, but does macOS use the hardware HEVC decoder on the 6900XT?

My understanding is the issue is the card doesn’t support HEVC files encoded in 10-bit 4:2:2 (the 4:2:2 being the problem).

I’m returning from a trip tomorrow and plan to give macOS 12 beta a try to see if there are any improvements, though I’m doubtful.

 

[rondocap]

My understanding is the issue is the card doesn’t support HEVC files encoded in 10-bit 4:2:2 (the 4:2:2 being the problem).

I’m returning from a trip tomorrow and plan to give macOS 12 beta a try to see if there are any improvements, though I’m doubtful.

I don’t think any GPU is going to help the playback unfortunately, like you said the M1 is really the only one that can do it For 10 bit 422

 

[DFP1989]

That’s my expectation as well.

 

[rondocap]

That’s my expectation as well.

I tried some red 6K footage on the M1, and when you have to export a large file, it really struggles and eats up all of the ram, so I know where you’re coming from with the limitations as well

I think the trick with the Mac Pro is that you really have to take advantage of the things it’s good at. Maybe consider getting the ninja recorder, that way you can do pro res on the R5- that will make everything super fast

 

[DFP1989]

Yeah I’ve got a Ninja, just never cared for building up a compact camera like that.

I use the 1DX III mostly when using a slider which is great as it’s lighter, and C200 in Cinema Raw Light for everything else.

It’s not too big of a deal, I set the MP transcoding and it’s well and truely done when I sit down to edit the following day, would just love to not have to do it (and double up on disk space).

Yes I had a project that simply would not export on my M1 Mac due to ram limitations, was exceptionally frustrating. Obviously never have such issues with the 7,1.

 

[rondocap]

Yeah I’ve got a Ninja, just never cared for building up a compact camera like that.

I use the 1DX III mostly when using a slider which is great as it’s lighter, and C200 in Cinema Raw Light for everything else.

It’s not too big of a deal, I set the MP transcoding and it’s well and truely done when I sit down to edit the following day, would just love to not have to do it (and double up on disk space).

Yes I had a project that simply would not export on my M1 Mac due to ram limitations, was exceptionally frustrating. Obviously never have such issues with the 7,1.

Yeah for now that has to be like that, that’s why I’ve moved to using only raw formats or pro res when i can, the Mac Pro works a lot better for that

 

[DFP1989]

Yeah for now that has to be like that, that’s why I’ve moved to using only raw formats or pro res when i can, the Mac Pro works a lot better for that

Call definitely no problems with red for Canon raw light.

 

[rondocap]

You need to enable the GPU debayer and decompression in preferences to test R3D Decode.
You shouldn’t use any CPU.
For me, on a 16 core I use the CPU for everything 5K and below, GPU for 6K and above.

Some weird results - seems like debayer and decompression is definitely faster when using the 6900xt, but when I try it with the same export with the Vega ii, seems like debayer only is faster. I wonder why that is?

In resolve, Red 6k raw to Pro res 422 HQ, 20 min clip:

2x 6900XT: 6 min 15 sec
2x w5700x: 9 min 8 sec
1x Vega ii: 12 min 8 sec

 

[thexash]

Can you benchmark a game like "Shadow of the Tomb Raider" to test the 3d rendering performance?

I make 3d applications so I'd be interested to know the difference between the W5700, 6900XT and the Vega II for that use cases - and games are pretty close equivalent.