MP 7,1 What tests would you like to see vs 2019 Mac Pro and M2 Ultra?

[rondocap]

I wanted to see what type of tests you'd think would best show the difference between the 2019 Mac Pro, and the new M2 Ultra 76 core GPU.

2019 Mac Pro will be:
28 Core Xeon
192GB Ram
AMD W6800x Duo GPU

M2 Ultra:
24 core cpu, 76 core GPU
128GB Ram

I want to test some R3d Raw files, that always shows the GPU power nicely. Pro Res the Ultra will obviously crush anything.

It's really hard to benchmark or test macs sometimes - on PCs you can just throw a game on and look at the FPS and see a lot more info. Mac has many use cases, like video, 3D, audio, etc - that sometimes one area can be much stronger than the other on different systems.

And lots of quirky behavior too - when I tested 4 GPUs in the Mac Pro, dual W6800x Duos - performance was worse sometimes in Final Cut or Resolve because the software was not really optimized to take advantage of all 4 GPUs, 2 worked much better with a single 6800x Duo.

 

[rubberducker]

I have that exact Mac Pro set up and would love to see how the M2 Ultra handles multiple layers of 8k. r3d.

At the moment 2 mins or so of three layers of that bogs down the system (in Premiere at least) especially if you add in some upscaled h264s as well. It’s border line unusable.

Would love to see M2 ultras take on that.

 

[m1maverick]

It's really hard to benchmark or test macs sometimes - on PCs you can just throw a game on and look at the FPS and see a lot more info. Mac has many use cases, like video, 3D, audio, etc - that sometimes one area can be much stronger than the other on different systems.

Interesting...I didn't realize PCs couldn't be used for video, 3D, audio, etc.

 

[rondocap]

Interesting...I didn't realize PCs couldn't be used for video, 3D, audio, etc.

Haha, yeah didn’t mean it like that.

 

[m1maverick]

Haha, yeah didn’t mean it like that.

Whew! I was surprised to read that!

 

[ZombiePhysicist]

Obviously Geekbench and metal. But do run. Game! A good one is no man’s sky as I believe it’s native on both Intel Mac and ASi.

 

[MisterAndrew]

Have you received your M2 Ultra Mac Pro yet? I'm excited to see some comparisons. I would test the things that Apple claims are faster on the M2 Ultra plus everything else you can throw at it. The Studio reviews were testing Cinebench, Handbrake, and Blender so those could be thrown in.

I suggest to run workstation benchmarks such as SPECworkstation 3.1. You might need to install Windows to do that. e.g. Boot Camp on the Intel Mac Pro and Parallels VM on the M2 Ultra Mac Pro. Other benchmarks include UL PCMark 10, BAPCo CrossMark, 7-Zip 21.7, JetStream, Speedometer, Principled Technologies WebXPRT4, GFXBench, UL 3DMark.

You should be able to find some games that run on macOS to test the FPS. If not, test them in Windows.

There's something called Phoronix Test Suite, a collection of benchmarks, that is suppose to be able to run on macOS. I haven't tried it.

 

[chfilm]

I’d also run some r3d tests and almost more importantly some denoising nodes in resolve! Maybe try that good old candle benchmark?

 

[randy85]

Idea for a Resolve benchmark:

1. create 120 fps 4k project
2. re-conform a 4k (or 8k) clip to 120 fps in 'clip settings'
3. add temporal and spacial noise reduction to one node and a lut to a second node
4. watch it cry and observe the playback frame rate

You can also test export times.

I just tried this on my base M1 Pro and with a 4k prores clip I get 6 fps with the NR settings I put in - so plenty of headroom to test the better systems.

 

[chfilm]

Idea for a Resolve benchmark:

1. create 120 fps 4k project
2. re-conform a 4k (or 8k) clip to 120 fps in 'clip settings'
3. add temporal and spacial noise reduction to one node and a lut to a second node
4. watch it cry and observe the playback frame rate

You can also test export times.

I just tried this on my base M1 Pro and with a 4k prores clip I get 6 fps with the NR settings I put in - so plenty of headroom to test the better systems.

That sounds like a great test. Retiming in general would be interesting!

 

[rondocap]

Here are some quick tests I ran on Final Cut so far:

M2 Ultra 76 core GPU is impressive, here’s how it did in exporting R3D Raw 6k 20 minute clip to ProRes 422 HQ in Final Cut vs a Mac Pro.

Mac Pro 28 core Xeon, 192GB ram, various GPUs tested

Mac Studio M2 Ultra 24 core cpu, 76 core GPU, 128GB ram


Export of 20 min 6K R3d Raw to 422 HQ:

2x 6900 XT 8 min 52 sec
M2 Ultra 10 min 11 sec
W6800w duo 10 min 14 sec
W5700x 17 min 47 seconds
Vega ii 18 min 34 sec

Did slightly worse at 8K vs 6K but still impressive:

Export of 20 min 8K R3d Raw to 422 HQ:

W6800w duo 18 min 24 sec
M2 Ultra 24 min 47 sec
W5700x 33 min 30 seconds
Vega ii 36 min 19 sec

 

[AndreeOnline]

I wanted to see what type of tests you'd think would best show the difference between the 2019 Mac Pro, and the new M2 Ultra 76 core GPU.

I think you were already one of the contributors on the Rocket Science Resolve benchmark?

It would be nice to see how the AS M2 Ultra performs there. It's kind of GPU heavy and we already have quite a few configurations to compare against. Always nice with as "standardized" tests as possible.

 

[rondocap]

I think you were already one of the contributors on the Rocket Science Resolve benchmark?

It would be nice to see how the AS M2 Ultra performs there. It's kind of GPU heavy and we already have quite a few configurations to compare against. Always nice with as "standardized" tests as possible.

M2 Ultra 76 core GPU, 128 GB of Ram

Rocket Science Benchmark, UHD ProRes

09 Blur: 64 fps
18 Blur: 33 fps
30 Blur: 20 fps
66 Blur: 9 fps

1 TNR: 100 fps
2 TNR: 53 fps
4 TNR: 25 fps
6 TNR: 17 fps

 

[MisterAndrew]

That's interesting so far. It looks like it's having trouble keeping up with the 6800 and 6900 GPUs.

 

[avro707]

Here are some quick tests I ran on Final Cut so far:

M2 Ultra 76 core GPU is impressive, here’s how it did in exporting R3D Raw 6k 20 minute clip to ProRes 422 HQ in Final Cut vs a Mac Pro.

Mac Pro 28 core Xeon, 192GB ram, various GPUs tested

Mac Studio M2 Ultra 24 core cpu, 76 core GPU, 128GB ram


Export of 20 min 6K R3d Raw to 422 HQ:

2x 6900 XT 8 min 52 sec
M2 Ultra 10 min 11 sec
W6800w duo 10 min 14 sec
W5700x 17 min 47 seconds
Vega ii 18 min 34 sec

Did slightly worse at 8K vs 6K but still impressive:

Export of 20 min 8K R3d Raw to 422 HQ:

W6800w duo 18 min 24 sec
M2 Ultra 24 min 47 sec
W5700x 33 min 30 seconds
Vega ii 36 min 19 sec

That 2x 6900XT combination is very, very fast.

 

[Jethro!]

The real test is which machine will have the greatest useful longevity.
I'll put my bets on the 2019 over the 2023.

 

[MisterAndrew]

The real test is which machine will have the greatest useful longevity.
I'll put my bets on the 2019 over the 2023.

Definitely, if you consider Windows and Linux also. Presumably native Linux will be able to be installed on the M2 Ultra Mac Pro, but of course GPU upgrades that have kept the 4,1/5,1 going for so long aren't possible.

 

[rondocap]

From using it a little more now, the Mac Studio is certainly very fast. It feels in general a bit more responsive than my Mac Pro, for normal tasks - makes sense since the single core performance is so much better on it.

Video workflow has been very fast too - with R3d raw. I even hooked up the Mac Studio to my Mac Pro via Thunderbolt, and can use the many storage drives that are in my Mac Pro - that's funny to me, ha.

If you consider the fact that the W6800W Duo didn't cost too much less than the entire Mac Studio with 76 core GPU, that is certainly pretty impressive. It's also quieter and uses much less power.

The 128GB of ram has been holding strongly too, most I've used is maybe under 90GB.

 

[MisterAndrew]

Ohh, you got the new Studio. I thought you got the new Mac Pro. I guess nobody bought one of those yet lol.

 

[rondocap]

Can someone give some insight as to what might be going on here? I tested numerous times with no background rendering, clean library in Final Cut Pro. Davinci Resolve gave similar results.

The Mac Pro for this task, R3D raw to H.264 consistently does around 3 min 16 sec, while the M2 Ultra is 7 min 18 sec.

4k R3d raw on a 4k timeline, no background render on either, and cleared everything.

Or something else possibly wrong?

Exporting it to ProRes 422 HQ gave the same result, so maybe it's nothing to do with the H264.

Other similar tests I ran either have them closer, or the M2 ultra can even win sometimes - like if test 6K r3d raw I get different results.

Why would the W6800x duo be this much faster with H264 or pro res as the final export?



10 min 4k red raw on 4k 24p timeline, export to to h.264 (Similar results when exported to 422 HQ)

Mac Pro W6800W Duo 3 min 16 sec
Mac Studio M2 Ultra : 7 min 18 sec
MacBook Pro M2 max 8 min 36

And then different 4k file, but now 60p - and the M2 Ultra is faster below, but same H264 export.

10 min 4k60p red file on 4k timeline to h.264

Mac Pro W6800W Duo 3 min 9 sec
Mac Studio M2 Ultra : 2 min 45 sec

Any ideas?

 

[AndreeOnline]

Can someone give some insight as to what might be going on here? Any ideas?

The first thing that came to mind was that .r3d debayering was optimized for AMD cards quite a while back (after having been exclusive to RED Rocket or CPU). Maybe there's something there that's missing on Apple Silicon.

But that doesn't add up with the last results—because it looks like you're using RED raw with the same type of debayer quality in all tests, right?

 

[rondocap]

The first thing that came to mind was that .r3d debayering was optimized for AMD cards quite a while back (after having been exclusive to RED Rocket or CPU). Maybe there's something there that's missing on Apple Silicon.

But that doesn't add up with the last results—because it looks like you're using RED raw with the same type of debayer quality in all tests, right?

Yeah, the same. The results are from different Red cameras, so maybe there is something in there causing this to happen.

I am going to narrow it down to just one file type, that may be easier.

I even tried it with both decompression and debayer settings mixed around in Resolve, but same results

 

[Dopemaster]

Can someone give some insight as to what might be going on here? I tested numerous times with no background rendering, clean library in Final Cut Pro. Davinci Resolve gave similar results.

The Mac Pro for this task, R3D raw to H.264 consistently does around 3 min 16 sec, while the M2 Ultra is 7 min 18 sec.

4k R3d raw on a 4k timeline, no background render on either, and cleared everything.

Or something else possibly wrong?

Exporting it to ProRes 422 HQ gave the same result, so maybe it's nothing to do with the H264.

Other similar tests I ran either have them closer, or the M2 ultra can even win sometimes - like if test 6K r3d raw I get different results.

Why would the W6800x duo be this much faster with H264 or pro res as the final export?



10 min 4k red raw on 4k 24p timeline, export to to h.264 (Similar results when exported to 422 HQ)

Mac Pro W6800W Duo 3 min 16 sec
Mac Studio M2 Ultra : 7 min 18 sec
MacBook Pro M2 max 8 min 36

And then different 4k file, but now 60p - and the M2 Ultra is faster below, but same H264 export.

10 min 4k60p red file on 4k timeline to h.264

Mac Pro W6800W Duo 3 min 9 sec
Mac Studio M2 Ultra : 2 min 45 sec

Any ideas?

Weird. Perhaps Resolve needs an iterative update for optimising specifically with the M2 Ultra? Inconsistencies like that to me point more to software than hardware. No expert in this field though.

 

[DrEGPU]

I’d be interested in finding out the deep-learning capabilities. I have doubts the M2 could compete against 2x 6800XT’s, or even 2x RTX3090’s (running windows or Linux, but wouldn’t be directly comparable obviously).

 

[Apple Knowledge Navigator]

Export of 20 min 6K R3d Raw to 422 HQ:

2x 6900 XT 8 min 52 sec
M2 Ultra 10 min 11 sec
W6800w duo 10 min 14 sec
W5700x 17 min 47 seconds
Vega ii 18 min 34 sec

Did slightly worse at 8K vs 6K but still impressive:

Export of 20 min 8K R3d Raw to 422 HQ:

W6800w duo 18 min 24 sec
M2 Ultra 24 min 47 sec
W5700x 33 min 30 seconds
Vega ii 36 min 19 sec

This is what we need more of, real world testing.

I'm so sick of people harping on about how the Ultra isn't a replacement for other GPUs based purely off of benchmarks. I think the results above are extremely impressive and only confirm that the Studio and Pro are great value when compared to the other cards that were tested. Some of those cards were approaching $5,000 alone.