[Runtime Feature]: Implement RTX Neural Texture Compression

What is the context of the feature/improvement?

There are many ways that current GPUs struggle with Remix. Currently, VRAM is a major one. Lower end Nvidia GPUs often struggle with Remix due to this. Back when I was using a 3070 ti, I would very often run out of VRAM while playing Remix mods for extended durations.

Implementing RTXNTC would help resolve that issue.

Please describe the feature/improvement?

If RTXNTC is implemented, the vast majority of GPUs capable of running Remix would be able to utilize NTC-on-Sample. This would significantly reduce VRAM consumption from textures, while also allowing packaged Remix mods to be smaller.

From the ReadMe, the savings on a 2048x2048 texture vs the BCn compression we currently use:

What would be your solution? (optional)

Implement RTXNTC, and possibly introduce a system to utilize it automatically to reduce VRAM load when necessary. Loading only a certain quantity of textures (and maybe at certain resolutions?) as NTC on Load would avoid unnecessary performance hits vs loading all textures with NTC on Sample.

Version

Latest

Media

No response

Start by reviewing the linked RTXNTC project and how RTX Remix currently handles BCn texture compression. Define the integration scope for NTC-on-Sample or NTC-on-Load, then verify that the completed work reduces texture VRAM usage and packaged mod size without unacceptable performance impact.