Qwen3.8-27B optimization hits 1150 tps on RTX 3090
iamMess · reddit · 2026-08-18
The author released an update to their hyper-optimized Qwen3.8-27B inference engine for RTX 3090. By implementing fp16 recurrent state, int8 activations across all layers, and probabilistic sampling, the engine achieves 99 tps for single requests and a peak of 1150 tps for batched requests. Prefill speeds increased by 25-50%. The project is open-sourced on GitHub.
Related event: Qwen3.8-27B Inference Optimized for RTX 3090(2 posts)→
More from Infra
- Phison CEO calls CXMT memory scare a 'ghost story', warns of materials shortage — BenBajarin · 2026-10-16
- Ben Bajarin: The Future of Compute Is Fungible as Hyperscalers Design for Workload Flexibility — BenBajarin · 2026-10-16
- AirLLM runs frontier-size models layer-by-layer: 671B on 12GB VRAM — Aiden_Tech_Ai · 2026-10-16
- NVIDIA Launches cuPotential in CUDA-X to Bring AI Agents to Semiconductor Materials Simulation — tomaszbednarz · 2026-10-16
- Q4-quantized coder model on 8-core CPU builds interactive solar system demo in one shot — IntrepidMindExplorer · 2026-10-16
- TokenSpeed MLA 2 ships for Vera Rubin: up to 3.44× faster Kimi K3 decode — zhyncs42 · 2026-10-16