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

[AndreeOnline]

Hi guys,

For some time, there has been a manual benchmark in DaVinci Resolve referred to as "Standard Candle". It's based on a short h.264 HD clip with a few standardised node setups: Blur and Noise Reduction. Maybe you are familiar with it?

At any rate, as many have moved to UHD projects I've wanted to move the benchmark to this resolution. Below, I'm providing a new project and a new clip in three flavors. The short description might assume some familiarity with Resolve and the concept. If you want to participate but don't understand what you're supposed to do, raise your hand and I'll go into more detail. I'm pasting this description from a post I made in another color grading forum:

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

Similar to the previous project, it's a very short clip. But this time it comes in 3 flavors:

• 6K ProRes_LT (6144x3456)
• UHD ProRes_LT (3840x2160)
• UHD HEVC 10bit (3840x2160)

The project and the 3 clips can be found in the .zip file.

The timeline has the clips organised in the order above and is set to loop over a single clip, so you can focus on one clip at the time. The node structure is the familiar one and just copied over from the well known 'Standard Candle': 09-18-30-66 Blur and 01-02-04-06 TNR.

For new users: you can move between the various node groups with CMD+N (next grade) or CMD+B (previous grade). On Windows I assume it's CTRL+N and B.
Also, after you import the project (right click in the project window, chose "import project" and navigate to the .drp file) the media will be 'missing'. On the media page, select the 3 clips, right click and choose "relink selected media" and navigate to where you put the downloaded clips.

The download is a manageable 265MB or so.

https://drive.google.com/file/d/1Objogjc28KbSOPTT3tJaivg6R18Tf7jU/view?usp=sharing

It would be cool if some of you with interesting Mac Pro setups, new and old, would like to participate. I'll find a way to collect the results—together with contributions from PC users—a bit down the road.

 

[AndreeOnline]

A few initial results. While the test includes 3 clips, the fps for each grade turn out to be similar. This means that it is the grade itself, much more than the underlying clip, that determines playback performance. To make the data easier on the eyes, I'm just posting the middle clip (UHD ProRes) fps for a few Macs:

UPDATE nov 2023: Added a number of results, including a MacBook Pro M3 Max 40-core. Tweaked graphic by removing the 01 TNR node that was "overperforming" across all Macs, making it less relevant. The noise section is now 2 TNR nodes and 6 TNR nodes.

NOTE: People with a keen eye might find 'discrepancies' in the graphic below where you think one system should be faster than another. I'm providing the data as I get it. These are not tests done by myself on a single day, but user-collected data provided over time in different versions of Resolve and macOS. But the benchmark setup has remained the same.

Mac Pro 7.1 2019 | 28 core | 2x W6800x Duo (quad W6800x 32GB):

09 Blur: 120 fps (Top Score)
18 Blur: 73 fps (Top Score)
30 Blur: 46 fps (Top Score)
66 Blur: 21 fps

1 TNR: 109 fps (Top Score)
2 TNR: 53 fps (Top Score)
4 TNR: 28 fps (Top Score)
6 TNR: 19 fps (Top Mac Score)

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

MacPro 7.1 2019 | 16-core | 2x Vega Pro II Duo (quad Vega Pro II 32GB):

09 Blur: 50 fps
18 Blur: 45 fps
30 Blur: 42 fps
66 Blur: 26 fps (Top Score)

1 TNR: 47 fps
2 TNR: 32 fps
4 TNR: 23 fps
6 TNR: 18 fps

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

Mac Pro 7.1 2019 | 24 core | two 6900 XT

09 Blur: 59 fps (capped)
18 Blur: 59 fps (capped)
30 Blur: 41 fps
66 Blur: 19 fps

1 TNR: 59 fps (capped)
2 TNR: 46 fps
4 TNR: 23 fps
6 TNR: 16 fps

-------------------------PC reference with Nvidia 4090 --------

AMD 5950x | Inno3D 4090 24GB | Rocky Linux 9 - Nvidia 520.56.06

66 Blur: 17 fps

6 TNR: 21 fps (Top Score)

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

Mac Pro M2 Ultra, 76 core GPU, 128GB RAM:

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

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

Mac Pro 7.1 2019 | 24 core | W6800x Duo:

09 Blur: 59 fps (capped)
18 Blur: 51 fps
30 Blur: 31 fps
66 Blur: 15 fps

1 TNR: 59 fps (capped)
2 TNR: 39 fps
4 TNR: 20 fps
6 TNR: 14 fps

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

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

09 Blur: 78 fps
18 Blur: 43 fps
30 Blur: 27 fps
66 Blur: 13 fps

1 TNR: 64 fps
2 TNR: 35 fps
4 TNR: 18 fps
6 TNR: 13 fps

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

Mac Pro 7.1 2019 | 12 core | Vega Pro II 32GB:

09 Blur: 46 fps
18 Blur: 24 fps
30 Blur: 15 fps
66 Blur: 07 fps

1 TNR: 53 fps
2 TNR: 30 fps
4 TNR: 16 fps
6 TNR: 11 fps

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

Mac Pro 7.1 2019 | 24 core | dual W5700x:

09 Blur: 59 fps (capped)
18 Blur: 30 fps
30 Blur: 19 fps
66 Blur: 9 fps

1 TNR: 51 fps
2 TNR: 26 fps
4 TNR: 14 fps
6 TNR: 10 fps

Mac Pro 5.1: 12 Core, Vega Frontier Edition 16GB

09 blur: 31 fps
18 blur: 16 fps
30 blur: 10 fps
66 blur: 05 fps

01 TNR: 31 fps
02 TNR: 17 fps
04 TNR: 09 fps
06 TNR: 06 fps

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

iMac 2020 | 10 core | 5700XT 16GB:

09 Blur: 18 fps
18 Blur: 10 fps
30 Blur: 06 fps
66 Blur: 03 fps

1 TNR: 30 fps
2 TNR: 16 fps
4 TNR: 9 fps
6 TNR: 6 fps

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

MacBook Pro 16" 2019 | 8 core i9 | 5500M 8GB:

09 Blur: 13 fps
18 Blur: 07 fps
30 Blur: 04 fps
66 Blur: 02 fps

1 TNR: 17 fps
2 TNR: 08 fps
4 TNR: 04 fps
6 TNR: 02 fps

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

MacBook Air M1 2020:

09 Blur: 07-10 fps

Looking for more complete data set here.

As the data set gets larger, I'll present it in a chart. I'm looking actively to see if I can get someone with a Mac Mini M1 to test it.

 

[chfilm]

Wow- so my second Vega doesn’t help AT ALL with noise reduction, but only with blur nodes! That’s a major bummer. Is that expected?

 

[AndreeOnline]

No, I wouldn't say that it's expected. You can see on Barefeats' site that when he doubles up on GPUs he gets increased TNR performance too. But he isn't running this exact test. Still.

Let's make a note of our observations. I'll verify my numbers on my side. Once we have bit more data it's easier to spot anomalies.

But this is one of the reasons for doing a test like this. Sometimes we can spot settings that need to be tweaked and sometimes we learn something about a specific software that is unexpected.

But it's a bit too early for that.

 

[chfilm]

No, I wouldn't say that it's expected. You can see on Barefeats' site that when he doubles up on GPUs he gets increased TNR performance too. But he isn't running this exact test. Still.

Let's make a note of our observations. I'll verify my numbers on my side. Once we have bit more data it's easier to spot anomalies.

But this is one of the reasons for doing a test like this. Sometimes we can spot settings that need to be tweaked and sometimes we learn something about a specific software that is unexpected.

But it's a bit too early for that.

hmmm, I was very frustrated when I popped the 2nd card in and made the same experience in a project that I was working on. Before I was reaching around 19fps with NR turned on, and afterwards, pretty much the same - the crazy thing is, it's not like one GPU was sitting idle and doing nothing, no they're both working full blast, but still no improvement fps wise over your results with just one card. How is that even possible?

 

[ArchStudent]

Hmmm...

MacPro 5,1 with dual X5675 CPUs, RX 580 8GB, 64GB of ram, using an NVMe drive, Mojave and DaVinci Resolve 17 results with MIDDLE CLIP UHD Pro:

09 blur : 18 fps
18 blur : 09 fps
30 blur : 05 fps
66 blur : 01 fps

TNR 01 : 20 fps
TNR 02 : 11 fps
TNR 04 : 2 fps
TNR 06 : 1 fps

I will follow up with a Vega 64 and Frontier Edition.

I will also report back in DaVinci Resolve 16 as well.

 

[AndreeOnline]

Hmmm...

MacPro 5,1 with dual X5675 CPUs, RX 580 8GB, 64GB of ram, using an NVMe drive, Mojave and DaVinci Resolve 17 results with MIDDLE CLIP UHD Pro:

09 blur : 18 fps
18 blur : 09 fps
30 blur : 05 fps
66 blur : 01 fps

TNR 01 : 20 fps
TNR 02 : 11 fps
TNR 04 : 2 fps
TNR 06 : 1 fps

I will follow up with a Vega 64 and Frontier Edition.

I will also report back in DaVinci Resolve 16 as well.

Thanks for posting!

Your results are basically completely determined by the RX580.

 

[ArchStudent]

Thanks for posting!

Your results are basically completely determined by the RX580.

I would be very curious to see if there is any difference between resolve 16 & 17 however minimal it may be.

I will also test with an R9 280X

 

[deppest]

Would it be possible to provide the project file in resolve 15 compatible format? V15 being the last version running on HS 10.13 I believe.

 

[meisterjaan]

Mac Mini M1, Big Sur and Resolve 17.1B (free version). Had Chrome and Textedit open, don't know if it matters.

My results:

6K prores
09 blur - 7
18 blur - 3
30 blur - 2
66 blur - 1

1 tnr - 8
2 tnr - 4
4 tnr - 2

UHD prores
09 blur - 7
18 blur - 3.5
30 blur - 2
66 blur - 1

1 tnr - 9
2 tnr - 4
4 tnr -2

hevc
09 blur - 7
18 blur - 3.5
30 blur - 2
66 blur - 1

1 tnr - 9
2 tnr - 4
4 tnr - 2

 

[myassin]

Hi guys!

I added the project files to DaVinci Resolve 17 on Windows. What to do next? Apologies I'm not a professional.

 

[AndreeOnline]

Hi myassin,

This is a pretty old skool test, where you just play the different clips with various pre-defined node configurations. So if you've imported the project and relinked the clips as I describe in the first post, you just go to the color page and play the timeline, or perhaps better, one clip at the time in loop mode while you evaluate the fps that is displayed.

The project has already been setup so you don't need to change anything expect make sure the media is linked on your machine.

Did you import the project or just add the media to a new empty project?

 

[cobra521]

Andree,

Thanks for doing the heavy lifting on this!

It all worked for me after I updated to the v17 beta. It even looks like it loops correctly on the color page. Except, maybe my eyesight is too bad; I can't find the fps. Can you point me to it?

Tom

 

[AndreeOnline]

I can't find the fps. Can you point me to it?

Top left, above your viewer. There's a percentage that indicates the Viewer's size, and to the right of that you have a coloured dot with the fps number as well. It shows up when playing back.

If you have a green dot, it means 'realtime' and the number will normally be 24, 25.. or whatever. If you're not achieving realtime playback, the dot will be red.

So the numbers reported for various clips and settings is this number, while playing back the footage on the color page.

 

[cobra521]

Andree,

Thank you - I have the dot and the number. So far, so good.

How do I find out which of the seven cases I am observing, i.e., blur (there's 4 different ones) or tnr (3 different)?

Tom

 

[AndreeOnline]

Cobra521, I take it you read the text in bold in my first post that says "for new users"? =)

The keyboard shortcut CMD+N goes to the next grade/node setup and CMD+P goes to the previous.

Either look at the text overlay that shows up in the viewer for like 2 seconds after switching grade, or just look at the node setup itself. You can just count the nodes, since you're either looking at Blur nodes or TNR nodes.

This little "benchmark" has been around a while, but was based on a HD clip. It sort of assumes basic familiarity with Resolve already—like how to import a project and move between grades. Sorry if the first post is confusing.

I trust you got what you need now! Please post results and specs if you've got some numbers! All results are interesting (both high and low fps).

 

[Scrotorr]

Mac Pro 5.1: 12 Core (x5675) 48GB DDR3, Radeon Fury Nano 4GB

09 blur = 15 fps
18 blur = 08 fps
30 blur = 05 fps
66 blur = 02 fps

01 TNR = 25 fps
02 TNR = 13 fps
04 TNR = 07 fps
06 TNR = 04 fps

Also of note - These were results with Metal. OpenCL cut GPU utilization and results by about half.

I'll give it a shot when I get home with a R9 Fury X, R9 270X 2GB, and a Radeon VII

 

[Scrotorr]

Well, I had another machine on the other side of the office, so I checked on there too...

Mac Pro 4.1->5,1: 4 Core (2.67Ghz W5675) 16GB DDR3, R9 270X 2GB

09 blur = 11 fps
18 blur = 05 fps
30 blur = 03 fps
66 blur = 01 fps

01 TNR = 16 fps
02 TNR = 07 fps
04 TNR = 02 fps
06 TNR = 00 fps

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

Mac Pro 4.1->5,1: 4 Core (2.67Ghz W5675) 16GB DDR3, RX 550 4GB (Baffin Core Version)

09 blur = 05 fps
18 blur = 02 fps
30 blur = 01 fps
66 blur = 00 fps

01 TNR = 07 fps
02 TNR = 03 fps
04 TNR = 01 fps
06 TNR = 00 fps

 

[Scrotorr]

Strangely my numbers weren't much better on the GPU compute box here in the office.

Windows 10 Pro, i7 4790, 32GB DDR3, Nvidia Geforce GTX Titan X 12GB

Numbers were almost identical to my R9 Fury Nano in the Mac Pro above.

 

[cobra521]

The keyboard shortcut CMD+N goes to the next grade/node setup and CMD+P goes to the previous.

Either look at the text overlay that shows up in the viewer for like 2 seconds after switching grade, or just look at the node setup itself. You can just count the nodes, since you're either looking at Blur nodes or TNR nodes.

Andree,

Thank you! As my grandfather would have said "I see, said the blind man as he tripped over the hammer and saw!"

The pointers about where to look and when were spot on. When I get a few minutes I'll gather the data from my setup and post it.

Tom

 

[rondocap]

Here are some of my results so far on the new 6800 and 6900 XTs.

Only limitation could be the 8 core and 32gb of ram, I'm upgrading this base model and waiting on parts - I have a 2nd 24 Core Xeon with 96gb of ram and Dual W5700x that I will test later as well.

Middle Clip, UHD Pro Res

Mac Pro 7.1 2019 | 8 core | 6900 XT

09 Blur: 59 fps
18 Blur: 35 fps
30 Blur: 21 fps
66 Blur: 10 fps

1 TNR: 59 fps
2 TNR: 34 fps
4 TNR: 18 fps
6 TNR: 13 fps

Mac Pro 7.1 2019 | 8 core | 6900 XT And 6800 XT

09 Blur: 59 fps
18 Blur: 59 fps
30 Blur: 40 fps
66 Blur: 19 fps

1 TNR: 59 fps
2 TNR: 40 fps
4 TNR: 22 fps
6 TNR: 15 fps

 

[rondocap]

Some more scores with the W5700x

Mac Pro 7.1 2019 | 24 core | 2x W5700X

09 Blur: 59 fps
18 Blur: 30 fps
30 Blur: 19 fps
66 Blur: 9 fps

1 TNR: 51 fps
2 TNR: 26 fps
4 TNR: 14 fps
6 TNR: 10 fps


Mac Pro 7.1 2019 | 24 core | Single W5700X

09 Blur: 31 fps
18 Blur: 16 fps
30 Blur: 10 fps
66 Blur: 5 fps

1 TNR: 41 fps
2 TNR: 21 fps
4 TNR: 11 fps
6 TNR: 7 fps

 

[rondocap]

From my tests above, it seems like a single 6900XT is almost as fast as 2x Pro Vega ii in blur, and actually faster in TNR vs 2x Vega ii.

When combining the 6900xt and the 6800xt, they are considerably faster than the 2x Vega ii pro.

Very impressive so far! I would love to see 2x 6900xt vs 2x Pro Vega ii duo, I bet the 2 6900xt would give those 4 gpu a run for their money for considerably less cost

 

[rondocap]

Aside from this newer test above, Here is my Candle Benchmark score, I think It may be the highest we have ever seen. This is with 2 GPUs, but the single 6900 XT is impressive too.

Almost perfect, all get at least 60 FPS except the 64 blur nodes which got 49 FPS. I think if I put in a 2nd 6900 xt in place of the 6800 xt, it will get close to 60 FPS too.

I think this makes the Pro Vega ii duo look outdated, especially for the price, if video/metal is your forte. Of course in compute those are still great.

2019 Mac Pro, 28 Core, 96GB ram, 6900 XT And 6800 XT

6900 XT + 6800 XT
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

Single 6900 XT
1 NR 2 60 FPS
3 NR 2 48 FPS
6 NR 2 25 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

 

[AndreeOnline]

Thanks rondocap for taking the time!

Some nice results in there... ?