I just got my hands on a brand new computer, 4Ghz 8 cores i9, and a Rtx 2080 ti.
I was expecting the i9 to blow my mind during renders, so I fed him a Vr conversion project.
Input is a 360 equirectangular, 5.7K video. Converted using the VR Converter effect, to a 5670 x 1000 h.265 video via media encoder.
Thought it would heat up the CPU for a while.
However, when I open up the performance tab through windows' task manager, the cpu is running on average at 20%. The Gpu sometimes climbs up to 10% (which does not surprise me since I'm already aware that media encoder doesn't know how to encode using the gpu).
What I'm concerned about is the low cpu usage. I thought it would go to 100% and render 4 times faster...
Drives don't seem to be a bottleneck. It's a m.2 SSD and i'm not asking to write 1Tb/s, the file ends up around 500Mb.
I have tried different input files, different resolutions (in and out) not much change. I also tried without effects, simple "3 videos stitched together", cpu usage goes to around 40%...
Is this normal, anything I should be aware of ? Workeraounds ?
Media encoder uses the GPU on select operations.
Did you look at the single cores. May be the operation that was processed did not allow for parallel work. The same for the GPU. May be you are hitting Amdahls law.
I understand your concern. Hardware utilization while export is dependent on the complexity of the composition.
Copy link to clipboard
Hi, thanks for your answers.
- I did look at the cores (or threads) individually. They all look very identical, no thread (out of the 16 I see)is left unused.
It could be related to Amdahls law, but regarding what Jeff posted, the "sweet spot" seems to be around 8 to 10 cores, which is my case. (Unless he is referring to threads, this is not clear).
- I haven't tried exporting from after effects internally, but it doesn't seem convinient since I need hevc. However I will try to export to pngs then export the sequence to hevc using AME. To see if the precessor is best used. But I doubt that the double encoding will be a better option.
(...) but regarding what Jeff posted, the "sweet spot" seems to be around 8 to 10 cores, which is my case. (Unless he is referring to threads, this is not clear).
He's for sure referring to cores.
A thread is an offspring of a process doing some independent work. That's software. A thread can run on the same core than the parent or on a different core.
Hyper-threading is a technology promoted by Intel fooling the system to have more cores than are available. However, the hyperthreaded thread is sharing the core with a different thread, making it only effective for specific situations.
Copy link to clipboard
So I tried different solutions.
When I exported directly from after effects, in prores, the cpu usage was around 60 to 70%. Which is much better than 10%.
Then I converted this prores file to h265 via AME only, and cpu is almost 100%.
Right now I'm just coverting 4 png sequences side by side, from after effects via AME, CPU is around 100% as well. This leads me to believe that with the effects that I wanted to render (VR Converter), AME was being bottlenecked by after effects' ability to provide the images. I haven't really had the "free time" to time the process to see which is effectively faster, but I think this is a good lead.