# Thread-Efficient Decoding for Neural Texture Compression

arXiv: https://arxiv.org/abs/2608.27888 · HTML: https://mlsanity.com/paper/2608.27888/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 7 of its 14 references that are in the map.
- First posted on arXiv 28 Aug 2026.

## Builds on (most cited first, 7 of 7)

- [Recipes for Adapting Pre-trained Monolingual and Multilingual Models to Machine Translation](https://mlsanity.com/paper/2004.14911/) (2020)
- [Random-Access Neural Compression of Material Textures](https://mlsanity.com/paper/2305.17105/) (2023)
- [Freezing Subnetworks to Analyze Domain Adaptation in Neural Machine Translation](https://mlsanity.com/paper/1809.05218/) (2018)
- [Not All Adapters Matter: Selective Adapter Freezing for Memory-Efficient Fine-Tuning of Language Models](https://mlsanity.com/paper/2412.03587/) (2024)
- [Hardware Accelerated Neural Block Texture Compression with Cooperative Vectors](https://mlsanity.com/paper/2506.06040/) (2025)
- [Real-Time Neural Materials using Block-Compressed Features](https://mlsanity.com/paper/2311.16121/) (2023)
- [Rethinking the adaptive relationship between Encoder Layers and Decoder Layers](https://mlsanity.com/paper/2405.08570/) (2024)

## Abstract (arXiv)

Neural texture compression (NTC) achieves higher compression ratios than BCn formats but suffers from GPU thread divergence, which significantly reduces runtime performance. In this work, we propose a shared decoder MLP architecture -- trained with a gradual decoder freezing schedule -- combined with texture clustering to reduce thread divergence by 25%-52% while preserving rendering quality. We evaluate our method on over 500 textures and multiple real rendering scenes, demonstrating up to 8.48x speedup on the Radeon RX 9070 XT GPU compared to non-shared baselines. Our key contributions include: (1) a unified shared decoder architecture that reduces divergence by grouping textures; (2) a training recipe with gradual decoder freezing that improves stability and reconstruction accuracy; (3) a semantic clustering strategy using CLIP embeddings that groups similar textures for effective decoder sharing; and (4) comprehensive performance and ablation studies validating our approach.