← ALL RECIPES
NVIDIA NEMOTRONOPTIMIZING

NEMOTRON CASCADE-2 30B-A3B

Hybrid Mamba-2 + MoE on GPU — including a larger-than-VRAM mode that stays deterministic.

PARAMETERS

30B (3B active)

ARCHITECTURE

Hybrid SSM

QUANT

Q4_K_M / TurboHybrid 4.06 bpe

VRAM

18–24 GB (any budget via unified memory)

01

OVERVIEW

Cascade-2 mixes Mamba-2 state-space layers with sparse MoE — the architecture class powering the fastest agentic decoders. IronWorks runs the full hybrid natively on GPU: warp-level SSM scan kernels, fused multi-expert GEMVs, and graph-captured decode.

It is also the showcase for unified memory: the 24.7 GB Q4_K_M file does not fit a 24 GB card. Where OS paging gives erratic 2–3 tok/s, IronWorks' expert LRU cache with learned pinning delivers deterministic throughput at 98–99% expert hit rates — same weights, exact outputs, explicit budget.

02

PREREQUISITES

  • 24 GB GPU (Q4_K_M via expert cache) or 18 GB (TurboHybrid mode)
  • Measured: RTX 3090 Ti
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 (expert cache + learned pinning)

terminal
IRONWORKS_MOE_CACHE_GB=12 \
iwx complete -m Nemotron-Cascade2-30B-A3B-Q4_K_M.gguf --gpu \
  -p "Hi" -n 400
04

TUNING

Explicit VRAM budget

terminal
iwx complete -m Nemotron-Cascade2-30B-A3B-Q4_K_M.gguf --gpu \
  --vram-budget 20G -p "Hi" -n 400
05

MEASURED PERFORMANCE

METRICMEASUREDNOTES
Decode (Q4_K_M, n=400)~107 tok/sllama.cpp same-session tg96: 190.7 — kernel-efficiency gap being closed
TurboHybrid mode (4.06 bpe, 17 GB)45.2 tok/s end-to-endIronWorks-only quantization mode; not comparable to Q4_K_M
Oversubscribed (24.7 GB on 24 GB)6.56 tok/s deterministicvs 2.2–3.1 erratic under OS paging; expert cache 98–99% hits, prewarmed

RTX 3090 Ti, single GPU, greedy, same-session A/Bs 2026-06-04 (head-to-head) and 2026-07-20 (unified memory).

Expert usage statistics persist to <model>.iwx_usage; second and later runs start pre-warmed (+40% measured over cold cache).