UCLA's optical generative model draws images with light, matching a 1.07B-param diffusion model in under 1ns
mtizard · x · 2026-09-18
UCLA researchers unveiled an "Optical Generative Model" that replaces GPU compute with the physics of light. A digital encoder imprints a noise pattern onto a laser, and light physically propagates through custom optical layers—the light itself performs the computation, with no silicon bottleneck. The system reportedly matches the output quality of a 1.07 billion-parameter diffusion model, generating faces, complex images, and Van Gogh-style art, with each image produced in under a nanosecond and near-zero energy cost during generation.
More from Infra
- Who Opens Up the Compute Middle Market? The $10-100M Deployment Standards Layer — AccBalanced · 2026-09-18
- Announced data centers hit 342GW vs 50GW installed, implying $26.3T in spend — BenBajarin · 2026-09-18
- Hyperbolic hires quant researchers to build GPU compute as a tradable asset class — YiMaTweets · 2026-09-18
- Crusoe raises $3.9B Series F at $30.9B valuation to fuel AI energy buildout — beffjezos · 2026-09-18
- Periodic Labs details its stack: 4.1x Megatron throughput, frontier-beating science models on 1,300 H200s — hsu_byron · 2026-09-18
- Bonsai quant hits 50 tok/s at 128k context on a 24GB card, letting users run two sessions at once — julianharris · 2026-09-18