New in beta: ProRes export performance improvements

[Fergus H]

We are working on a series of ProRes export performance improvements in Premiere Pro and Media Encoder for macOS and Windows. 

The first of these improvements are now in beta and make ProRes exports up to three times as fast as in the current shipping products.

The largest improvements will be seen:

• Using simpler sequences.
• When exporting 16-bit ProRes files*
• When using a high-end computer, with the biggest improvements on Apple silicon computers

You will also see the same performance benefits when creating ProRes proxies, as the same engine powers both export and proxy creation. 

We’d love to get your feedback on the first part of our ProRes export performance improvements. 

Regards,

Fergus

* Currently, Premiere Pro’s ProRes export presets use 8-bit, which is silly. To get the correct quality export, customers need to manually modify the export setting. Coming soon to beta will be updated, 16-bit ProRes presets.

[scrozier]

I'm actually more excited for the default presets to be 16 bit than the marginal performance gains. One of those things that takes a lot of twirly time in the maligned export UI. Thanks! 

[Joost van der Hoeven]

Great gesture! So this is across the board? Not only on Macs with a Mx Max processor that has a ProRes feature?

[Fergus H]

@Joost van der Hoeven Yes. 

[Tranquil_Cosmos157F]

This is great – but why do we need the 8-bit option? For making bad video on purpose?

Please just make all ProRes render at 16-bit, and remove the 8-bit option. 

[Fergus H]

I agree. The quickest step is for us to update the preset, so that's what we'll do first. 

Fergus

[Alex367466572b5i]

Export to ProRes works in Pr 24.3 and does not work in Pr Beta 24.4/24.5. Attached screenshot. Is some library missing?

[Kyle Plumadore]

@Alex367466572b5i

Thanks for pointing this out. We have seen this pop-up on a few Windows machines internally as well and are currently investigating. 

Mac configs do not appear to be impacted and I would still encourage Mac users to give ProRes exports in the latest Premiere Pro Beta a try!

[David Simonton, DVA]

Are you still unable to get ProRes exports to work in 24.4 and 24.5 @Alex367466572b5i ? 

[Alex367466572b5i]

@David Simonton, DVA Switching to English UI (en_US) helped...

[Alex367466572b5i]

@David Simonton, DVA It turned out to be not about localization. And from which drive the video is being read. There is no increase in speed with simple 16-bit encoding, it has even become worse.
CPU load - 34%, SSD - 5%... Pr Beta 24.5.0.7

[David Simonton, DVA]

Thank you for getting back!  That's good information.

Regarding encode times, it's surprising that it's slower for you.  On Windows, I'd expect it to be about the same or slightly better than before, comparing 16-bit ProRes presets to 16-bit ProRes presets.  Have you quantified how much slower it is for you, meaning an export that took x seconds in prior version now takes y seconds in 24.5?

[Alex367466572b5i]

@David Simonton, DVA 

I wanted to test how much faster it became to work. Previously, I did this on Windows, it was the fastest for Premiere (Resolve Windows does not know how to export to ProRes).

v24.3 DJI UHD LongGOP read SSD NVMe --> ProRes 422 16-bit write SSD SATA --> 415 sec
v24.3 DJI UHD LongGOP read SSD SATA --> ProRes 422 16-bit write SSD NVMe --> 368 sec
v24.5 Beta DJI UHD LongGOP read HDD SATA --> ProRes 422 16-bit write SSD NVMe --> 393 sec

CPU utilization (i5-13500) is less than 40%, this is the bottleneck in the system.

[David Simonton, DVA]

Thanks for the additional information.  In addition to comparing 24.3 to 24.5, I also see variance in the drive/bus you're reading from and writing to.  That's also interesting information, but it somewhat complicates the isolation, because it changes the decode variables as well as the encode variables.  That ~7% degradation between 24.3 and 24.5 is concerning.  I'll have to see if we can reproduce that in house when reading from SSD vs HDD. 

[Alex367466572b5i]

@David Simonton, DVA Describe where you had an increase with what settings I will check with myself and post the result.

[David Simonton, DVA]

@Alex367466572b5i , I have not seen increased render times in internal testing, at least none that were related to the ProRes QT optimizations, which mostly affect Mac ARM systems, and are especially potent on higher end Apple Silicon like the Max and Ultra variants.

In your results above, I was most interested in the ~7% degradation you reported as:

v24.3 DJI UHD LongGOP read SSD SATA --> ProRes 422 16-bit write SSD NVMe --> 368 sec
v24.5 Beta DJI UHD LongGOP read HDD SATA --> ProRes 422 16-bit write SSD NVMe --> 393 sec

In the 24.3 test, you're reading from SSD, and in the 24.5 test, you're reading from HDD, which may or may not affect the total render time, but it might.  (similar with the longer render time when reading from the NVMe drive).  When I do A/B testing internally, I have to keep environmental variables tightly controlled while testing beween builds or beta flag states. 

If you want to do a test that would definitively answer whether or not what you're seeing is related to the change that went in for Windows environments, related to QT exports, drop me a PM and I can walk you through how to navigate that in a Beta build. 

[Alex367466572b5i]

@David Simonton, DVA 
This is the strangeness that 24.3 opens a video from a SATA SSD, and 24.5 Beta gives the error that I showed above.
Therefore, looking at the low CPU usage, I thought that the acceleration was somehow related to the disk system.
The SSDs are new and completely empty. Tomorrow I will try to put them in the same conditions and unsubscribe.

[Alex367466572b5i]

@David Simonton, DVA 
HDD read --> M.2 NVMe PCIe 4.0 x4 write
24.3
CUDA --> 396 sec.
Soft --> 197 sec.
24.5 Beta
CUDA --> 401 sec
Soft --> 308 sec
*Soft. I have been testing drives for many years in a real project in Premiere Pro, recording 10-bit uncompress 422. And turning off the video card gives an increase in recording speed, this also works with ProRes.

[Alex367466572b5i]

v24.3: Cuda 308 sec., Soft.only: 146 sec.

v24.5: Cuda 288 sec., Soft.only: 144 sec.

v24.6 (Beta): Cuda 161 sec., Soft.only: 149 sec.

[MyerPj]

Am I reading this right? So in all cases the software only is fastest?

This is a single h.264 long gop source clip -> prores 422

[Alex367466572b5i]

Yes, it has been like that for a long time, and not only with ProRes. You can check for yourself, we conducted a collective test.

[Tranquil_Cosmos157F]

Can you show us a screengrab of the timeline you're exporting? And the export settings? 

Sounds like there could be some clips or effects that gives parts of the timeline a red render bar (green after rendering) - which could lead to shuffling data between the CPU and the GPU, which again is slower than doing everything on the CPU.

[Alex367466572b5i]

1 clip! An sequence was created on its basis. Export.
Here are the screenshots:
vk.com/@adobe_premiere_pro_2024-apple-prores

[Fergus H]

Hi all, 

As an update to my original post, the ProRes export improvements described (including the 16-bit presets) are now shipping in Premiere Pro and Media Encoder v24.5. 

Now in beta are improvements to the export performance on Windows. Windows customers using the v24.6 beta (or later) should see a 2X improvement in ProRes export performance. We'd love your feedback on this!

Regards,

Fergus

[Alex367466572b5i]

@Fergus H 

v24.3: Cuda 308 sec., Soft.only: 146 sec.

v24.5: Cuda 288 sec., Soft.only: 144 sec.

v24.5, MXF: Cuda 239 sec., Soft.only: 107 sec.

v24.6x46 (Beta): Cuda 161 sec., Soft.only: 149 sec.

v24.6x75 (Beta): Cuda 165 sec., Soft.only: 149 sec.

v24.6x75 (Beta), MXF: Cuda 102 sec., Soft.only: 88 sec.