Unofficial Teardown NVENC Screen Recorder

After installing the Teardown beta on Steam it was apparent how slow FFMPEG CPU encoding is, so I set out to find the correct hardware acceleration commands that would meet the quality that is provided by the default script while maintaining an overall fast experience.

Downloads

You can find the default script provided by Tuxedo Labs along with the NVENC script and Linux variants in links below, they can also be found in the scripts folder. You'll need to have an FFmpeg build under or equal to 4.2.3 which supports NVENC, all needed files can be found under the Releases tab.

It's a drag and drop replacement for the original script and works the same.

[Windows - GPU] [Linux - GPU]

Hardware

I've tried to test this on a few PCs to see what the performance is like, below are the specs of those machines.

CPU	RAM	GPU	Storage (Where teardown saved the images to)	OS
AMD Ryzen 3700X 8-Core Processor	16GBs DDR4 @ 3000MHz	Gigabyte GeForce RTX™ 3070 VISIONS OC 8G	Seagate ST4000VX077 4TB Hard Drive	Windows 10 Enterprise 21H2
AMD Ryzen 1500X Quad-Core Processor	16GBs DDR4 @ 2133MHz	Gigabyte GeForce® 1660 Ti OC 6G	Samsung 850 EVO 250GB SSD	Windows 10 Enterprise 21H2

Game config

It's unlikely that this would effect anything but in case, below are a list of the video settings and mods I had enabled at the time.

• Video Settings:
  • Display:
    • Mode: Borderless window
    • Resolution: 4096 x 2160
  • Graphics:
    • Render Scale: 100%
    • Render Quality: High
    • Gamma correction: 1
    • Field of view: 120
    • Depth of field: Enabled
    • Darrel distortion: Enabled
    • Motion blur: Disabled
    • Vertical sync: Every Frame
• Mods:
  • Screen Recorder - [Built in]
  • Artzert's Utilities - [Steam Workshop]
  • REALISTIC NUCLEAR BOMB - [Steam Workshop]

Results

If you're just here for the results look no further the table below explains where the result came from, how fast the encode was and the final size of the video.

All tests used the same 1,171 images in a resolution of 4096 x 2160, they can be found here on Google Drive in the "capture.zip" archive, the accompanying videos can also be found in the same folder.

PC1 - AMD Ryzen 3700X | Gigabyte GeForce RTX™ 3070 8G

Encode Type	Encode Time	Encode Speed Avg.	Final File Size	YouTube - Google Drive	File Name
CPU - libx264	1 Minute 8 Seconds	x0.3	71.5 MB	[YouTube] - [Google Drive]	2022-03-03-16-29-51-cpu.mp4
GPU - cuda	12 Seconds	x1.7	81.5 MB	[YouTube] - [Google Drive]	2022-03-03-16-32-27-gpu.mp4

PC2 - AMD Ryzen 1500X | Gigabyte GeForce® 1660 Ti 6G

Encode Type	Encode Time	Encode Speed Avg.	Final File Size	YouTube - Google Drive	File Name
CPU - libx264	2 Minutes 3 Seconds	x0.1	73.2 MB	[YouTube] - [Google Drive]	2022-03-03-16-41-54-cpu.mp4
GPU - cuda	16 Seconds	x1.6	85.9 MB	[YouTube] - [Google Drive]	2022-03-03-16-45-55-gpu.mp4