Autor: Jakub Rusinowski · Ostatnia aktualizacja: 11 września 2026
Przyjazny konsumentom hybrydowy model SSM+attention od NVIDII, wydany w marcu 2026. Wariant 30B mieści się na pojedynczym RTX 4060 Ti (17 GB w Q4) i utrzymuje ~54 tokeny na sekundę — szybciej niż jakikolwiek czysty Transformer o porównywalnej jakości. Wykorzystuje selektywne warstwy state-space (Mamba2), przetwarzając długie konteksty 3× taniej niż mechanizm uwagi. Licencja Apache 2.0, dostępny w Ollama.
| Licence | What it permits | Applies to |
|---|---|---|
NVIDIA Open Model License | Commercial use permitted Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms. | Nemotron-Cascade 2 30B-A3B |
Apache-2.0 | Commercial use permitted Commercial use permitted. No usage restrictions beyond attribution. | Nemotron Cascade 2 70B (Unverified Listing) |
| Nemotron-Cascade 2 30B-A3B | Min 20 GB VRAM · Q4_K_M · 1,048,576 ctx · ollama run nemotron-cascade-2:30b |
| Nemotron Cascade 2 70B (Unverified Listing) | Min 43 GB VRAM · Q4_K_M · 128,000 ctx · |
The cheapest GPU that runs Nemotron Cascade 2 locally (min 20 GB VRAM) is the AMD Radeon RX 7900 XT (20 GB).
Install Ollama then run: ollama run nemotron-cascade-2:30b
Minimum VRAM: 20 GB. For best results use Q4_K_M quantization.
Nemotron Cascade 2 needs about 20 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: Nemotron-Cascade 2 30B-A3B (20 GB, Q4_K_M); Nemotron Cascade 2 70B (Unverified Listing) (43 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Yes — Nemotron Cascade 2 runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at Q4_K_M.
Q4_K_M is the best balance of quality and VRAM for Nemotron Cascade 2 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
Install Ollama, then run: ollama run nemotron-cascade-2:30b. This downloads Nemotron Cascade 2 and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.