← ALL RECIPES
ALIBABA QWEN3AT PARITY

QWEN3 30B-A3B (MOE)

185 tok/s MoE decode at parity with llama.cpp — 30B-class quality at 3B-class speed.

PARAMETERS

30B (3B active)

ARCHITECTURE

MoE

QUANT

Q4_K_M

VRAM

18 GB full-resident

01

OVERVIEW

Qwen3 30B-A3B activates ~3B parameters per token across 128 experts, delivering 30B-class quality at small-model decode speeds. It is the most popular locally-run MoE family.

IronWorks runs it fully GPU-resident with device-side routing, fused expert GEMVs (gate+up+SwiGLU fused; expert kernels measured at the ~670 GB/s bandwidth ceiling), and whole-decode-loop graph capture — measured at decode parity with llama.cpp (0.99×) in same-session A/B.

02

PREREQUISITES

  • 18 GB+ VRAM for full residency (measured: RTX 3090 Ti)
  • ~17.5 GB model file (Q4_K_M)
  • Lower-VRAM expert-cache mode in development
03

DEPLOY

Install IronWorks

terminal
# CLI (CUDA build)
cargo install ironworks-cli --features cuda

# or build from source
git clone https://github.com/nervosys/IronWorks && cd IronWorks
cargo build --release -p ironworks-cli --features cuda

Run

terminal
iwx complete -m Qwen3-30B-A3B-Q4_K_M.gguf --gpu \
  -p "Compare Rust and C++ for systems programming." -n 512

iwx serve -m Qwen3-30B-A3B-Q4_K_M.gguf --port 8080
04

TUNING

Persistent KV cache

terminal
iwx serve -m Qwen3-30B-A3B-Q4_K_M.gguf --port 8080 \
  --kv-cache-dir ~/.cache/ironworks/kv
05

MEASURED PERFORMANCE

METRICMEASUREDNOTES
Decode (tg128)185.4 ± 0.6 tok/sllama.cpp same-session: 187.0 ± 11 (0.99× = parity, tighter variance)
Prefill (pp512)~475–506 tok/sllama.cpp leads prefill; MoE-prefill megakernel is the known remaining lever

RTX 3090 Ti, single GPU, greedy tg128, same-session A/B vs llama.cpp CUDA build, clean-GPU protocol, 2026-06-08.