First TPU Inference Megakernel Open-Sourced, Beats GB200

inferact open-sourced the first TPU inference megakernel, running the entire Kimi K3 model in a single Pallas kernel and hitting 709 tokens/s in low-concurrency decoding—56% faster than NVIDIA's GB200 NVL72 baseline, per vLLM maintainers' tests.

2026-09-24 ~ 2026-09-24 · 3 related posts

• Open-sourced TPU megakernel runs Kimi K3 at 709 tokens/s, beating GB200 — vllm_project · 2026-09-24
• vLLM Tests Show TPUv7 Hits 700 tok/s/user, 56% Faster Than NVIDIA GB200 NVL72 — woosuk_k · 2026-09-24

1 near-duplicate retellings: woosuk_k