Recommend a workflow for using the GPU on a AWS instance

[JonnyBurger]

Remotion 4.0 does not support Glibc 2.33 anymore, but there are no templates on AWS that have GPU drivers installed but use a new enough version of Glibc

[bilalmughal]

While working on GPU acceleration for headless Chrome/Chromium, I couldn't help but wonder: why not harness the full power of GPUs for video transcoding as well? I have just finished with my cost-centric analysis comparing CPU vs GPU for video transcoding and the results were astonishing, benchmark data reveal GPUs proved to be both faster and more cost-effective. You can check out my detailed findings here. So this compelled me even more to pitch this idea to the Remotion community.
This will not only offload more rendering workload to GPU but it will also free up the CPU, to perform so many general purpose tasks we generally need to perform.

And the best part is it's within our reach. All it takes is recompiling the FFmpeg binary with hardware acceleration support. I'm more than eager to lend a hand in this and see the results for myself.
Here is a graph from the post reflecting the stark difference in performance.

https://mirzabilal.com/cpu-vs-gpu-for-video-transcoding-challenging-the-cost-speed-myth

[UmungoBungo]

nice work! We have found that arm has better price performance than x86-based CPUs, which your post shows as well. I think it would be good to mention the architecture of each instance as I had to jump back and forth between your post and the AWS docs to understand 😄

[UmungoBungo]

so based on the method in your post, you are suggesting that the Remotion renderer would first check if a GPU is available, and if it is, perform transcoding using the GPU?

[UmungoBungo]

Related: #3013

[bilalmughal]

@UmungoBungo Thank you, and the reason behind not mentioning the architecture as AWS has now finally standardized the naming convention, for example, any instance name with g is AWS Graviton, like G5g.xlarge and C7g.4xlarge, similarly i is for Intel. As in C7i.2xlarge and a is for AMD like C7a.xlarge

nice work! We have found that arm has better price performance than x86-based CPUs, which your post shows as well. I think it would be good to mention the architecture of each instance as I had to jump back and forth between your post and the AWS docs to understand 😄

[UmungoBungo]

ahh I see, thanks for pointing that out! I had guessed g for Graviton and i for Intel but wasn't sure about the a

[bilalmughal]

Related to #3038

[JonnyBurger]

True, this is done now! https://www.remotion.dev/docs/miscellaneous/cloud-gpu