Copy link to clipboard
Copied
I have been rendering some fairly simple GFX (basic text animtors, shape layers, 2k images fading on and off etc) for a 1920x1080 60FPS presentation wall to ProRes4444 + Alpha files.
Even when rendering to a fast scratch SSD from RAM (almost always dark green bar during render) or from cache, AE seems to hit some sort of speed limit at around 30FPS.
Normally that's plenty fast, but some of these slideshows are 5min long comps, so a 5 min comp can take 10 min to output for 60FPS. My system is more or less idle while its writing from RAM and I am wondering what the limitation is that is keeping this from being a faster encode? ProRes 422HQ goes marignally faster but still tops out at around 30-40FPS
Seems like the bottleneck is just writing out the .mov file but the disk write max I can get is about 40MB/s which is a lot lower than what the SSD can handle.
@jenkmeister Any tips?
AE: 23.6.0 (build 62)
PC:
3960x Threadripper 24 core
2080ti
256GB ddr4 Ram
reading/writing to pci-e NVME drives.
Copy link to clipboard
Copied
There is no artificial limit on export FPS, but AE right now can only pass one frame at a time into the encoder (legacy, old code we hope to one day replace), so that might be a bottleneck. I'll try to find someone to repro but not sure if there is any "fix" I can share today.
Copy link to clipboard
Copied
Thanks for the prompt response. So it's not format specific just the encoder in general is single threaded. Hopefully thats in the works!
Copy link to clipboard
Copied
@jenkmeister curiously the write from RAM process goes about 2x as fast if you turn GPU acceleration off ((software only), which is strange behavior.
Copy link to clipboard
Copied
Also anecdotally the render seems to find a happy medium with Ram usage instead of maxing out all availiable ram with software only. I've been noticing my renders start to slow down when the ram maxes out.
Copy link to clipboard
Copied
I think the speed difference with CPU vs GPU encoding is still in line with the serial frames being sent. Because the frames have to get passed over to the GPU, but that still happens one at a time, the bandwidth of the PCI bus becomes a bottleneck. Whereas the CPU may encode slightly slower but that ends up being less time than the transfer to/from the GPU.
Utilizing up the RAM will definitely slow the render down as AE then needs to keep looking for memory to release/reuse. It's something I have one of the engineers looking at currently.
Copy link to clipboard
Copied
Yeah I thought it might be something like that. Hard to imagine the GPU would cause a 50% slowdown with PCI-E 4 and all that but I'm not super familiar with that side of things. I think for a lot of work I may switch to software only, GPU remains mostly unused when I hit render for a lot of my work.