Hardware Accelerated Neural Block Texture Compression with Cooperative Vectors · HackerPapers discussionshttps://hackerpaper.kaipfister.de/p/2506.06040/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:16:22 GMTHardware Accelerated Neural Block Texture Compression with Cooperative Vectors — mentioned by cubefoxhttps://news.ycombinator.com/item?id=45158024urn:hackerpapers:mention:2506.06040:45158024Sun, 07 Sep 2025 13:34:12 GMT<p>The more efficient versions compress the entire material stack in one, to exploit inherent redundancies, including albedo, normals, roughness etc. The main problem, as far as I can tell, is the increased GPU performance cost compared to block compression. Though recent research shows some versions are already pretty manageable on modern GPUs with ML acceleration: https://arxiv.org/abs/2506.06040 They cite 0.55ms per frame at 1080p on a recent Intel GPU. The budget for one frame at 60 FPS is 16.7ms, so the performance cost would be 3.3% of that. Though higher resolutions would be more…</p><p><a href="https://hackerpaper.kaipfister.de/p/2506.06040/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.GR