After Effects on AWS EC2

Report · Feb 14, 2022

Hi all,

We're using After Effects on an EC2 instance - currently a g4dn.12xlarge, which we thought should work great however it doesn't. Works great with Resolve but struggles with After Effects.

Does anybody know why this might be? Or alternatively a recommendation for a different instance type to use? Thanks!

Report · Feb 15, 2022

A classic lack of understanding how AE works under the hood and what its limitations are, it seems, no offense. I suggest you educate yourself about these things by looking up the thousands of "How to optimize performance..." and similar threads here on this forum and on the web in general. The sticky points to keep in mind:

very limited GPU acceleration
limited multithreading and barely any actual parallel processing
even if the first two points work, it's easy enough to construct complex projects that make AE slow down to a crawl with nested comps, expressions, specific effects

Point in case and brief summary: AE is 90% CPU-based even today and in a way where it favors the single most powerful single- or dual-core CPU. Excessive GPU acceleration is useless and so are fancy multicore processors. Memory consumption and disk IO also matter, but there's a lot of myth building and misunderstandings around that, too. At the end of the day, the "lamest" EC2 instance that emulates a core 9i processor with a mid-range GPU is more useful than your twelve core setup with four tensor GPUs.

Mylenium

Report · Feb 18, 2022

Thanks Mylenium.

This has been really helpful! Indeed it seems I was labouring under a misconception of how AE and also the EC2 instances worked.

I am having much more success now with instances designed for compute processing.

Many Thanks!

@Mylenium wrote:
A classic lack of understanding how AE works under the hood and what its limitations are, it seems, no offense. I suggest you educate yourself about these things by looking up the thousands of "How to optimize performance..." and similar threads here on this forum and on the web in general. The sticky points to keep in mind:

very limited GPU acceleration
limited multithreading and barely any actual parallel processing
even if the first two points work, it's easy enough to construct complex projects that make AE slow down to a crawl with nested comps, expressions, specific effects

Point in case and brief summary: AE is 90% CPU-based even today and in a way where it favors the single most powerful single- or dual-core CPU. Excessive GPU acceleration is useless and so are fancy multicore processors. Memory consumption and disk IO also matter, but there's a lot of myth building and misunderstandings around that, too. At the end of the day, the "lamest" EC2 instance that emulates a core 9i processor with a mid-range GPU is more useful than your twelve core setup with four tensor GPUs.

Mylenium

Report · Jul 21, 2022

NVIDIA says otherwise: https://www.nvidia.com/en-us/studio/software/

Mylenium:

"I suggest you educate yourself about these things by looking up the thousands of 'How to optimize performance...'"... on EC2.

Here is part of an answer that isn't useless: https://helpx.adobe.com/after-effects/using/basics-gpu-after-effects.html

The only part of the first post that is partly true is that After Effects won't take full advantage of GPU capabilities, but that only means that you need to think about your workflow. The way that you could make this work to speed up your work considerably is to set up After Effects and Premiere on an EC2 virtual machine configured to the specs recommended by NVIDIS and AWS. Use AE for motion graphics and invididual scenes as necessary, but then move your assets to Premiere for rendering the bigger project. You should also set up EC2 to store/back-up your assets to S3. The big benefit here is that you won't slow your desktop down by filling up and fragmenting your local hard drive. https://smallbusiness.chron.com/full-should-let-hard-drive-get-79132.html

S3 provides security, versioning, geographical redundancy, and a means of organization that will speed things up when you need to go back and grab assets you've used previously and pull them into a new project. Production professionals will understand this challenge. A step further for professionals who have dozens or even hundreds of prior projects is to layer an asset management system onto your S3 bucket(s) so you can tag assets to make them easier to find later. (e.g. https://www.netx.net/using-amazon-s3-storage-with-netxposure)

If you mainly work on short videos or maybe car commercials, you may be better off following Mylenium's advice, although the processor is called an "i9." If you work on bigger productions with heavier assets and computationally intensive graphics, you will need to move to a server-based solution--either local metal or cloud-based. You're probably on the fastest track by figuring it out in AWS, but it will take some time to get your configurations right.