Proteus Generates Custom GPU Kernels for Qwen3 122B, Up to 5.2x Faster Than vLLM
matei_zaharia · x · 2026-09-09
Databricks published a blog on Proteus, a system that generates specialized GPU kernels for the exact tensor shapes a model encounters at runtime, questioning why models from 1B to 1T parameters share the same generic kernels.
Key result: kernels generated for Qwen3 122B are 1.8–5.2x faster than the best implementations available in vLLM.
Verification is the hard part: in their initial harness, knowledge and context management—not candidate generation—consumed most of the token budget. Verification requires running candidates on real GPUs in isolation, often multiple times, and it must be airtight: a kernel must be proven functionally correct, not just fast.
More from Infra
- AI computing demand may never be sated, says Nvidia partner Iren's CEO — pstAsiatech · 2026-09-09
- China targets 9,800 EFLOPS of AI computing capacity by 2030, a fourfold boost — pstAsiatech · 2026-09-09
- Nvidia reportedly adopting copper-diamond composite materials for cooling — zephyr_z9 · 2026-09-09
- FT: Huawei invests across lithography supply chain as China targets 12 domestic DUV machines by year-end — EleanorOlcott · 2026-09-09
- Google Cloud August AI Infra Roundup: gVisor Sandboxes on Ray, Filestore on Colossus — dl_weekly · 2026-09-09
- Provably private inference services promise prompts unreadable to providers — corbtt · 2026-09-09