RTX 4090 hits 90 t/s on Qwen3 27B with llama.cpp — full config shared
SirApprehensive7573 · reddit · 2026-08-21
A user shares an exact llama.cpp setup running Qwen3-27B (UD-IQ4XS) on a fully-free RTX 4090: all GPU layers, FlashAttention, q80 KV cache, batch 256/ubatch 64, plus MTP-2 speculative decoding, reaching 90 t/s at 242k context. Without MTP the full 262k context fits but speed drops to 45-50 t/s; he pairs it with opencode on large codebases and prefers speed. Other Q4 quants and FP16 KV cache showed no noticeable difference for this model.
More from Infra
- Muse Spark Open runs locally on a Mac M3 Ultra via llama.cpp: frontier model on desktop — pcuenq · 2026-10-16
- Muse-Spark open-weights drop: 816B-A42B MoE model gets 359GB MLX 3-bit build — ns123abc · 2026-10-16
- vLLM ships Day-0 LoRA RL for Muse Spark Open, lifting AIME from 0.56 to 0.85 — vllm_project · 2026-10-16
- DSpark draft model for Muse Spark Open proposes 7 tokens per pass with 23-42% acceptance — vllm_project · 2026-10-16
- Cloudflare Stream's WebRTC ultra-low-latency live streaming is now Generally Available — ritakozlov · 2026-10-16
- YC-backed Overshoot launches inference optimization, claims $5M saved for AI teams — ycombinator · 2026-10-16