Copy link to clipboard
Copied
Multi-frame Rendering is here for After Effects beta users. Right now, you will have access to Multi-Frame Rendering for export only. Keep your eye on this forum, as we will be rolling out new features until we launch.
Internally, we have been testing a representative sample of projects with a suite of hardware configurations, and we are excited to finally put this feature in your hands and get your feedback. Test your unique projects on your own hardware so we can ensure that our performance updates benefit all of our customers and meet speed and quality metrics before we launch.
Beta testing of Multi-Frame Rendering will last a little longer than some of our other features because we currently do not support Multi-Frame Rendering in Preview, Motion Graphics templates, Dynamic Link, Adobe Media Encoder and AERender Command Line Interface.
Check out the blog and FAQ post to learn the following:
Start Testing!
We want to hear from you! Share your thoughts and ideas in this forum.
Test Your Comps
Test your comps in Multi-Frame Render mode vs. Single-Frame Render
Leave a comment and the following information in this forum:
If you detect any problems with render speed or quality, submit your projects
Share the following information in your comment:
Test our benchmark project and see how your hardware compares
We would love to know how your hardware compares to the results we have gathered from our test suite:
To test the benchmark project on your machine:
Share the following information in this forum:
Copy link to clipboard
Copied
Windows PC
R9 3900X 12 Cores - 24 Threads | RTX 3070 8GB | 32 GB RAM | 3500MB Read / 3000MB Write NVMe SSD
AE Benchmark Single Frame: 17:18
AE Benchmark Multi Frame: 07:05 ~ 2.5x Faster
Personal Project Single Frame: 12:39
Personal Project Multi Frame: 04:21 ~ 3x Faster
2015 MacBook Pro 15"
i7-4980HQ 4 Cores / 8 Threads | AMD R9 M370 | 16GB RAM | 1600MB Read / 500MB Write SSD
AE Benchmark Single Frame: 32:10
AE Benchmark Multi Frame: 27:01 ~ 16% Faster (I might be experiencing some thermal throttling tho, I will redo the benchmark after repasting the CPU)
You have no idea how satisfying it felt seeing AE utilizing 98% of my 3900X Aaahhhh that felt awesome!
Copy link to clipboard
Copied
Very happy to add my results to this long list:
So MF is 32% faster on my system in the Beta (on this project, on a single test)
Copy link to clipboard
Copied
Here are my results:
CPU: AMD Ryzen 9 5950X 16-Core Processor 3.40 GHz
RAM: 64GB
GPU: GeForce RTX3080 10GB
OS: Win 10(20H2) 64-bit
Multi-Frame: 4 min 05 sec
Single Frame: 9 min 13 sec
Copy link to clipboard
Copied
I'm really impressed about the speed gain of the Ryzen CPU's between the two architectures of ZEN2 and ZEN3 !
Very impressive. Should have done my build with a ZEN3 Chip.
Copy link to clipboard
Copied
Thanks for taking out time to test Multi-Frame rendering and sharing your results.
Copy link to clipboard
Copied
Can't wait for the official version of this update!
Here are my results:
AMD Ryzen 9 3950X 16 core
128 GB RAM
Nvidia RTX 2080 TI 11GB Vram
Single core: 14:02
Multy core: 6:08
Copy link to clipboard
Copied
Thanks for taking out time to test Multi-Frame rendering and sharing your results.
Copy link to clipboard
Copied
AMD Threadripper 3990X 64-Core / 256GB RAM / RTX 3090 24GB (x2) / Windows 10 64 (20H2):
Very exciting results using MFR. Would love to see the other half of those cores being utilized someday :D.
Copy link to clipboard
Copied
Are people finding that their GPU's VRAM is the limiting factor in many multi-frame render projects (not necessarily the Adobe benchmark one)? I've found that if I'm using large source material (4K, 8K), my VRAM usage shoots up well over 8GB (I have 11GB) and only one or two frames are rendered simultaneously, presumably because there's just no more available VRAM to hold more than a couple of frames or something. I have 64GB RAM, and AE is (on my current render) using 26GB of that, so there's plenty to space if AE needed it.
On my next system, I'm hoping to get a 24GB RTX3090, when they actually come into existence. I've seen a few users here have that card - have you taken a look at the VRAM usage while AE is rendering? Does it shoot up when using large source material, but you can still render quite a few frames simultaneously?
Copy link to clipboard
Copied
Interestingly, the same thing happens when using the latest Beta (18.1.0 Build 34) even when I'm using low res. source material. When MFR was first launched, my systemn would render 6 frames at a time (with its 6 cores). Now I rarely even get 2 frames rendering simultaneously. But the AE Benchmark project still renders happily with all my 6 cores.
Copy link to clipboard
Copied
The benchmark project is in Mercury Software mode by default, so in that case, MFR isn't checking the GPU VRAM to determine concurrent frame limits. When using Mercury GPU mode, MFR is performing some rough calculations on composition resolution versus GPU VRAM in an attempt to avoid GPU OOM errors but it's still pretty rough. We are working on analyzing individual frames and looking at ways we can increase the number of concurrent frames when using Mercury GPU modes.
Copy link to clipboard
Copied
Copy link to clipboard
Copied
Rendered as Quicktime using GoPro codec on one of our newest AE workstations:
Hardware specs: AMD Ryzen 9 5950X, 16 cores/32 threads, 128GB RAM, RTX 3080 with 10GB VRAM
Single-Frame render speed: 10:53
Multi-Frame render speed: 4:35 🙂 🙂
Also tried Multi-Frame with Mercury GPU Acceleration enabled, and then the render speed was 4:22 (all cache purged before each render).
I noticed that it used a high percentage of the CPU, which is very good. Also didn't use much of the GPU until I tested with GPU acceleration, then it used around 6-7GB of the VRAM.
I will also test on an almost similar workstation, but with Intel 10900K and RTX3070
Copy link to clipboard
Copied
Did another render on another workstation:
Intel i9-10900K CPU, 10 cores/20 threads, 128GB RAM, RTX3070 with 8GB VRAM
Single-Frame render speed: 12:43
Multi-Frame render speed: 7:23
Multi-Frame with GPU acceleration enabled: 6:36
So this shows that a higher core count is now the key, instead of single core frequency.
I have to buy some stocks in AMD 😉
Copy link to clipboard
Copied
Intel(R) Core(TM) i9-10900K CPU @ 3.70GHz 3.70 GHz
128 GB RAM
RTX 3080 10GB VRAM
this is a VR360 project, using a lot of native VR effects that are part of the VR Comp Editor workflow. also 3rd party plugins mostly deep glow and stardust. guess it's not optimized for MFR, that's why I am getting only 1 concurrent frame rendering. especially at the begnining where it is calculating.
on other types of projects I have seen good results, but I want this to work in practice.
I am missing the information about what operations are being rendered like in the old render queue. I would like to have more information in the render queue.
Copy link to clipboard
Copied
> I am missing the information about what operations are being rendered like in the old render queue. I would like to have more information in the render queue.
Can you let us know which information you want back and what you were using it for?
Copy link to clipboard
Copied
the information in previous render queue gives you in detail the operations that are being rendered, this would help me see what causes a bottleneck. I wouldn't want to lose that. now you just get the frame that is being rendered with no idea what's being calculated. there should be specific information on what the render is doing, this would help optimize projects, also since we are dealing with MFR, further information is required - how many CPU's are being used, how much the GPU is being used, how much RAM, VRAM per frame etc...
here's an example of the previous render queue:
Copy link to clipboard
Copied
Thanks, that's helpful. Yes, we took out the rendering section in the new RQ design as trying to communicate that information when there could be a lot of frames rendering concurrently is rather difficult to do well.
Also, being forced to use the RQ to figure out what's slow is not something that makes a lot of sense (to me) to make users do. It seems like it might be better to give you that information while you're putting together your compositions, perhaps alert you when something you did is going to hurt rendering performance? In the timeline perhaps? (no promises or announcements, just thinking out loud).
That aside, we do now have more information in the MFR system to identify what hardware might be the bottleneck for a comp and I'd like to find a way to surface that information someday. Rather than you having to look at a task manager/activity monitor-like solution inside AE, it'd be good to find a way to tell you what the bottleneck is comp by comp, rather than you having to figure out on your own. (again, no promises, just a bit of transparency on some of my own thoughts).
Anyway, that's some late night musings for you. Thanks for the feedback, it's very helpful.
Copy link to clipboard
Copied
Hi,
8700K i7 6/12 @ 5GHz / 64 GB DDR4 3200MHz / Radeon Vega 64 8GB HBM2 / macOS 10.14.6 / Ae 18.2.0 (Build 10)
First i tryed to render into ProRes 422 HQ with clean render cache:
Single-Frame: 14min, 15sec
Multi-Frame: 09min, 43sec
Then i switched to pure Animationwith clean render cache:
Single-Frame: 0min, 40sec
Multi-Frame: 0min, 17sec
I thought that couldn`t be true, but i checked both rendered clips and everything is there.
Strange…
Copy link to clipboard
Copied
I suspect you didn't do Edit > Purge > All Memory & Disk Cache... between renders.
Copy link to clipboard
Copied
Your hardware specs: # cores in your CPU, GB of memory, GPU model and amount of VRAM
Mac Pro (2019) macOS Big Sur 11.2.3
3.5GHz 8-Core Intel Xeon W
112GB 2666MHz RAM
AMD Radeon Pro W5700X 16GB
Single-Frame render speed: 19 Min, 46 Sec
Multi-Frame render speed: 12 Min, 57 Sec
Wow! Really impressed, look forward to this making it's way into AE! 🙂
Copy link to clipboard
Copied
What surprises me more is that my i9 MacBook Pro 16” has the same scores as your Mac Pro with Xeons.
Copy link to clipboard
Copied
I'm not surprised, I believe the i9 in your MacBook Pro can go up to 5.0GHz! I'd not recommend the 8-core Xeon to anyone 😞
Copy link to clipboard
Copied
I've just upgraded my CPU to 16-cire 3.2GHz Xeon, and I got a multi-frame time of 5min 40sec 🙂
Copy link to clipboard
Copied
Re-run the test to see if any more inporvements.
Mac Pro 2019
AE BETA - 18.2.0 - Build10
16 core
1x AMD Radeon Pro Vega II (32gbRAM)
192gb Ram
Mac OS 11.2.3
gpu acceleration (metal)
Multiframe: 4min 40sec (10 seconds faster)
Singleframe (with Metal): 12min 30sec (slightly slower)