Reuse what is already there
The design divides images into tiles and separates stored content from its rendering map. Reusable regions may avoid full-picture reconstruction and repeated memory writes.
VR research / RESEARCH PROTOTYPE
NX Warp.
Less reconstruction. More useful pixels.
An experimental video codec for rendered VR, combining Vulkan encoding, tile reuse, and GPU-native reconstruction in the custom WiVRn NX stack.
01 — CAPABILITIESNX WARP
Bring reconstruction to the GPU
Vulkan encoder and decoder paths connect to an atlas renderer. Tile-aware meshes explore moving warp calculations out of repeated fragment work.
Measure the complete path
Encoding, transport, decoding, rendering, and compositor scheduling have distinct costs. The research tracks quality and timing together rather than treating one fast stage as an end-to-end result.
02 — GETTING STARTED
Follow the research.
Requirements and setup vary by project. The repository documentation is the source for complete instructions.
Read the documentation ↗- 01
Read the design
Start with the project README and design paper for architecture, assumptions, and the current experimental scope.
- 02
Build the reference paths
Use the documented build instructions and fixtures. Keep the exact build identity and configuration alongside any results.
- 03
Compare evidence
Read the measured-results and measurement-rules sections. Use controlled comparisons before interpreting visual captures or individual timings.
PLATFORM & SETUP
Developer-oriented C++20 and Vulkan research, with experimental WiVRn NX integration. Primary measured hardware includes a Radeon RX 7900 XTX host and Pico 4 headset.
CURRENT SCOPE
No end-user download is available. Visible artifacts and incomplete quality gates remain; consistent 240 Hz delivery and physical motion-to-photon latency have not been demonstrated.
Keep going.
Documentation, development, and help.