MP All Models DaVinci Resolve 'Standard Candle UHD': Rocket Science

[pattielipp]

I have some new results from my Mac Studio M1 Ultra 64c, 128gb :
Interesting that the mac pro beat the m1 ultra in blur tests but in TNR, the studio smoked the mac pro.
I also ran an after effects test comp on both machines.. Mac pro took 5:40min to render, the M1 needed only 3:30min. I'm a bit in shock!

Mac Pro 7.1 2019 | 16 core | two separate Vega Pro II 32GB:



09 Blur: 60 fps
18 Blur: 34 fps
30 Blur: 20 fps
66 Blur: 10 fps



1 TNR: 49 fps
2 TNR: 24 fps
4 TNR: 14 fps
6 TNR: 10 fps


Mac Studio M1 Ultra 64GPU


09 Blur: 54 fps
18 Blur: 28 fps
30 Blur: 16 fps
66 Blur: 8 fps

1 TNR: 60 fps
2 TNR: 40 fps
4 TNR: 20 fps
6 TNR: 14 fps

Any chance you could re-run with the 18 beta?

 

[chfilm]

Any chance you could re-run with the 18 beta?

This was on the 18 beta.
Interestingly enough, when I took an actual HD project with a simple noise reduction on it, the mac pro beat the mac studio WAYY off, by factor 4x. I have no idea why, but the gpu on the studio was completely maxed out, while the mac pro still had overhead and finished within 30 secs where the mac studio took 1:18.
I can test other stuff if you tell me what you wanna know.

 

[AndreeOnline]

This was on the 18 beta.
Interestingly enough, when I took an actual HD project with a simple noise reduction on it, the mac pro beat the mac studio WAYY off, by factor 4x.

This is interesting and one of the things that I don't find 'all that great' about the Rocket Science test (and similar tests).

While I set up 'Rocket Science' myself, it is based on, and a strict carryover, of the 'Standard Candle' HD benchmark that we have a lot of data from.

Way before I finally provided the clip and the node setup for Rocket Science, I tried to convince some of the more experienced colorists over at Lift Gamma Gain, to take part and agree on a more realistic, but still demanding, node setup; perhaps a few stages of grades, qualifiers, noise reduction and maybe two or three popular effects.

As always, people were pulling in a few different directions and it ran out in the sand. I think I did such a setup for myself but didn't include it in Rocket Science, since I wanted it to be familiar to all who already knew Standard Candle.

Ultimately, I think it would make more sense now to have a Resolve benchmark with realistic grade setups, varying between just primary grades, something in the middle, and a heavy setup. Then you could add a render test since that is separate from the real-time grading and we know that the new Macs might struggle more in real-time (compared to multi RTX setups), but be very quick on export. That should be reflected as well.

This way people could look at that and realize "oh, I mostly do assembly cuts and primary grades—Resolve would be a great option for me".

 

[thirdhonk]

Not sure if someone posted a 2080 TI before so here is mine. Mac Pro 5,1 with RTX 2080 TI with stock clocking in Windows 10. Resolve 17.

6K
09 blur - 30
18 blur - 17
30 blur - 11
66 blur - 5

1 tnr - 29
2 tnr - 18
4 tnr - 10

UHD
09 blur - 32
18 blur - 18
30 blur - 11
66 blur - 5

1 tnr - 35
2 tnr - 20
4 tnr - 11

HEVC
09 blur - 31
18 blur - 18
30 blur - 11
66 blur - 5

1 tnr - 34
2 tnr - 20
4 tnr - 11

 

[thirdhonk]

Just installed my new 6900 XT, here are my results. Expected more, maybe I need to tweak something. A little bump over the 2080 TI. This was in Windows 10, will test it in Monterey as well.

6K
09 blur - 30
18 blur - 18
30 blur - 12
66 blur - 6

1 tnr - 31
2 tnr - 22
4 tnr - 14

UHD
09 blur - 34
18 blur - 20
30 blur - 13
66 blur - 6

1 tnr - 42
2 tnr - 26
4 tnr - 15

HEVC
09 blur - 33
18 blur - 20
30 blur - 13
66 blur - 6

1 tnr - 36
2 tnr - 24
4 tnr - 15

 

[mattfezz]

Mac Pro 2019 16 Core
1x Pro Vega II Duo
1x 6900 XT
(3GPU total)

UHD
9 blur - 108
18 blur - 56
30 blur - 33
66 blur - 16

1 tnr - 80
2 tnr - 38
4 tnr - 18
6 tnr - 12

Note I noticed a big difference in performance when:
• Scopes Disabled
• Playback settings - Hide UI Overlays on
• Playback settings - Minimize interface updates during playback on
• Performance mode - Disable

These settings are not saved with the project unfortunately, so data could be a bit hard to compare as it does affect the FPS a fair bit.

 

[AndreeOnline]

These settings are not saved with the project unfortunately, so data could be a bit hard to compare as it does affect the FPS a fair bit.

And this is the reason why the project is setup the way it is: I felt I had to keep everything the same for backwards compatibility. The actual numbers aren't THAT interesting (it's a very unrealistic node setup anyways)—but relative performance can be compared as long as everyone is testing the same thing.

Thanks for adding the data! 👍🏻

 

[pattielipp]

Re-doing this test with new hardware..4,1 flashed to 5,1 MacPro with dual X5690's for 12-core/24 thread, 96GB@1333, Radeon Pro W5700 8GB/PIXLAS, Monterey/Opencore 9, Resolve 18.1.4 Studio. timeline/proxy media off. Safari open with 1 active tab.

6K/UHD/HEVC
01 TNR 27/34/32.5
02 TNR 17.5/19/20
04 TNR 10/11/10.5
06 TNR 7.5/7.5/7.5

09 Blur 25/27/26
18 Blur 14/14/14
30 Blur 9/9/9
66 Blur 4.5/4/4

 

[pattielipp]

Ran this on a base M1 Mac mini.. Didn't even bother finishing these tests as the performance just isn't there. Just one more reason I'm sticking to my 13 year old Mac Pro.


6K/UHD/HEVC
01 TNR 8/9/9
02 TNR 4/4/5
04 TNR 2/2/2
06 TNR --- Didn't see a point in running.

09 Blur 6/6/7
18 Blur -/-/-
30 Blur -/-/-
66 Blur -/-/-

Out of curiosity I did a test of playback with an empty node tree and found the following

6K/UHD/HEVC
46/78/85

 

[rondocap]

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

 

[chfilm]

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

Thanks a lot! Seems like the for blur nodes it’s now on par with my dual Vega IIs.. a bit underwhelming I would say.. is TNR more CPU dependand? Cause there it shows almost double the numbers that I got.
I wonder if now is the time to upgrade for all the other advantages or to hold still till an m3 version comes out eventually… 🤔
And the other interesting question is if 192gigs of ram would improve those results further!

 

[AndreeOnline]

Updated the first post with a number of scores to make glancing comparisons a bit easier. Working on adding a graphic now to aid with legibility.

-------------

EDIT: graphic is up

 

[chfilm]

Updated the first post with a number of scores to make glancing comparisons a bit easier. Working on adding a graphic now to aid with legibility.

Thanks! very informative.. a shame that the studio just reaches the level of my Mac Pro with the dual Vegas. It seems the m3 ultra is the one to be holding out for.

 

[m3rob]

M2 Ultra 60 core GPU, 64 GB of Ram

Rocket Science Benchmark, UHD ProRes

09 Blur: 57 fps
18 Blur: 30 fps
30 Blur: 18 fps
66 Blur: 8 fps

1 TNR: 99 fps
2 TNR: 50 fps
4 TNR: 25 fps
6 TNR: 17 fps

Thanks for setting the up Andree! I love to see the comparisons. Anyone have a m2 max with 38 Core GPU they can test? I am considering that machine too

 

[pattielipp]

Just picked up a m2max studio with 12/38/64gb to replace my 5,1 with 6x2/radeon pro w5700 and 96gb. very happy so far.

6K/UHD/HEVC
01 TNR 59/59/59
02 TNR 31/32/32
04 TNR 15/16/16
06 TNR 11/10/10

09 Blur 37/38/38
18 Blur 19/19/19
30 Blur 11/12/12
66 Blur 6/6/5

 

[AlexVrg]

@m3rob & @pattielipp I'm about to buy a Mac Studio with either M2 Max or M2 Ultra, so your results are very enlightening, thank you very much!

But do you feel 64GB is enough RAM or would you upgrade to 96 or 128 GB?

 

[HDFan]

Interesting project, excellent instructions. Not sure how to get the node numbers since they are constantly changing, sometimes with red dot and sometimes green. Interestingly on a maxed out M2 Studio CPU is only ~115% (out of 2400%) and GPU is only at ~98% (out of 7600%?) using 7.6 GB of memory.