Nemotron 3 Puzzle 75B-A9B

Overview

NVIDIA's Nemotron 3 Puzzle: a 75B MoE model with 9B active parameters, tuned for reasoning, with 128K native context.

When to pick this model

• Reasoning-heavy tasks where you want more capacity than a 30B-class MoE without full dense-75B cost
• Multilingual reasoning and code generation
• You have access to 2x24GB GPUs or a 64GB+ unified-memory Mac
• Commercial deployment under NVIDIA's Open Model License

VRAM requirements by quantization

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 Puzzle 75B-A9B spills past single consumer GPUs even at Q4_K_M (44 GB) — think dual-GPU or workstation cards. Stepping up to Q8_0 nearly doubles the footprint to 80 GB, and unquantized FP16 weights take 150 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.

Without a GPU, Nemotron 3 Puzzle 75B-A9B needs roughly 98 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 1.5 tokens/sec on entry-level GPUs, on the order of 2.5 tokens/sec on a mid-range card, and up to 5 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 Puzzle 75B-A9B 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.

Which GPU should you buy to run Nemotron 3 Puzzle 75B-A9B?

To run Nemotron 3 Puzzle 75B-A9B locally at Q4, you need ~44 GB of VRAM. The best value for this is a Apple Mac Studio (64+ GB unified memory).

Strengths

• MoE (9B active of 75B) is lighter to run than an equivalent dense model
• Reasoning-focused tuning
• 128K native context
• NVIDIA Open Model License permits commercial use

Limitations

• 150GB in FP16 — needs substantial hardware even before quantization
• ~44GB VRAM even at Q4 (dual RTX 4090s or a 64GB+ Mac)
• No official Ollama tag — HuggingFace/vLLM install required

Typical workloads

In our catalog grid, Nemotron 3 Puzzle 75B-A9B is filed under Reasoning, Code Generation, Multilingual — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: multi-step reasoning and math-flavoured tasks; multilingual workloads.

The 125k-token context window covers long chats and mid-sized documents, though very large retrieval workloads will need chunking. 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: MoE, 75B total / 9B active · BF16 weights · Nemotron "Puzzle" architecture

Training: An NVIDIA Nemotron model focused on reasoning. Native 128K token context; training corpus not publicly detailed.

Quick start

# HuggingFace : nvidia/NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-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 Puzzle 75B-A9B need?

At the recommended Q4_K_M quantization, Nemotron 3 Puzzle 75B-A9B needs about 44 GB of VRAM. Q8_0 takes 80 GB, and unquantized FP16 weights take 150 GB.

Can Nemotron 3 Puzzle 75B-A9B run without a GPU?

Yes — with roughly 98 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 Puzzle 75B-A9B support?

Nemotron 3 Puzzle 75B-A9B supports a 125k-token context window (128,000 tokens).

Can I use Nemotron 3 Puzzle 75B-A9B commercially?

Nemotron 3 Puzzle 75B-A9B 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 Puzzle 75B-A9B on consumer hardware?

Our compatibility engine estimates on the order of 2.5 tokens/sec on a mid-range GPU and up to 5 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.

Which quantization of Nemotron 3 Puzzle 75B-A9B should I download first?

Start with Q4_K_M (44 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.