Single RTX 3090 runs 4 coding agents at ~290 tok/s with custom CUDA engine

A developer released a new version of OpenJet, a custom CUDA megakernel engine for Qwen3.8-27B Q4KM, that serves 4 coding agents simultaneously on a single RTX 3090 with total throughput around 290 tok/s.

2026-10-27 ~ 2026-10-27 · 2 related posts

• 4 coding agents on one RTX 3090: custom CUDA engine hits ~290 tok/s with Qwen3.8-27B — Adorable_Weakness_39 · 2026-10-27
• OpenJet runs 4 coding agents on one RTX 3090 at ~290 tok/s via custom CUDA megakernel — Adorable_Weakness_39 · 2026-10-27