Nemotron 3.5 Lightning 30B-A3B (BF16)

Overview

NVIDIA's BF16 checkpoint of Nemotron 3.5 Lightning: a 30B MoE model with just 3B active parameters, 256K context, and fast local inference for reasoning and code.

When to pick this model

• Local reasoning and coding workloads where inference speed matters more than raw model size
• You need 256K context without a dense-model VRAM footprint
• Commercial deployment under NVIDIA's Open Model License
• Multilingual reasoning tasks

VRAM requirements by quantization

Quantization	VRAM required
Q4_K_M (recommended)	25 GB
Q5_K_M	29 GB
Q8_0	35 GB
FP16 (no quantization)	66 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

In practice, Nemotron 3.5 Lightning 30B-A3B (BF16) spills past single consumer GPUs even at Q4_K_M (25 GB) — think dual-GPU or workstation cards. Stepping up to Q8_0 raises the footprint to 35 GB, and unquantized FP16 weights take 66 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.

Without a GPU, Nemotron 3.5 Lightning 30B-A3B (BF16) needs roughly 39 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 50 tokens/sec on entry-level GPUs, on the order of 85 tokens/sec on a mid-range card, and up to 130 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.

What hardware do you need

The table below matches Nemotron 3.5 Lightning 30B-A3B (BF16) to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.

GPU memory	Example cards	Best fit for Nemotron 3.5 Lightning 30B-A3B (BF16)
8 GB	RTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GB	Does not fit — needs 25 GB at Q4_K_M
12 GB	RTX 5070, RTX 5070 Ti Laptop, RTX 4080 Laptop	Does not fit — needs 25 GB at Q4_K_M
16 GB	RTX 5080, RTX 4080 Super, Radeon RX 9070 XT	Does not fit — needs 25 GB at Q4_K_M
24 GB	RTX 4090, Radeon RX 7900 XTX, RTX 5090 Laptop	Does not fit — needs 25 GB at Q4_K_M
32 GB	RTX 5090	Q5_K_M (29 GB used)

Which hardware should you buy to run Nemotron 3.5 Lightning 30B-A3B (BF16)?

To run Nemotron 3.5 Lightning 30B-A3B (BF16) locally at Q4, you need ~25 GB of VRAM. The best value for this today is a RTX 5090 32GB (GIGABYTE Windforce) (32 GB VRAM).

Strengths

• MoE (3B active of 30B) delivers high local throughput
• 256K native context
• Strong reasoning and code performance for its active-parameter count
• NVIDIA Open Model License allows commercial use

Limitations

• Still needs a 16-24GB GPU class for comfortable Q4 inference
• No Ollama tag — HuggingFace-only install path

Typical workloads

In our catalog grid, Nemotron 3.5 Lightning 30B-A3B (BF16) is filed under Local Reasoning, Agentic Code, Long Context — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: code generation and review (pair it with an editor integration like Continue.dev or Cline); multi-step reasoning and math-flavoured tasks; multilingual workloads.

The 256k-token context window is large enough to hold entire codebases' worth of files or long reports in a single prompt, which is what makes local RAG and document analysis practical. It ships under the NVIDIA Open Model License license — commercial use is generally possible but read the specific terms before embedding it in a product.

Architecture & training

Architecture: Mixture-of-Experts · 30B total parameters · ~3B active per token · 256K context · BF16 weights

Training: BF16 checkpoint of Nemotron 3.5 Lightning (NVIDIA). An MoE variant optimized for fast inference thanks to its low active-parameter count.

Quick start

# HuggingFace : nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.

Similar models worth comparing

Frequently asked questions

How much VRAM does Nemotron 3.5 Lightning 30B-A3B (BF16) need?

At the recommended Q4_K_M quantization, Nemotron 3.5 Lightning 30B-A3B (BF16) needs about 25 GB of VRAM. Q8_0 takes 35 GB, and unquantized FP16 weights take 66 GB.

Can Nemotron 3.5 Lightning 30B-A3B (BF16) run without a GPU?

Yes — with roughly 39 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.

What context window does Nemotron 3.5 Lightning 30B-A3B (BF16) support?

Nemotron 3.5 Lightning 30B-A3B (BF16) supports a 256k-token context window (262,144 tokens).

Can I use Nemotron 3.5 Lightning 30B-A3B (BF16) commercially?

Nemotron 3.5 Lightning 30B-A3B (BF16) ships under the NVIDIA Open Model License license. Commercial use is generally permitted subject to its terms — review the license text before shipping a product.

How fast is Nemotron 3.5 Lightning 30B-A3B (BF16) on consumer hardware?

Our compatibility engine estimates on the order of 85 tokens/sec on a mid-range GPU and up to 130 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.

Which quantization of Nemotron 3.5 Lightning 30B-A3B (BF16) should I download first?

Start with Q4_K_M (25 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It does not fit a single 24 GB consumer card — plan for multi-GPU or server hardware.