Only with CPU offload

Nemotron 3 Super 120B-A12B at Q4_K_M needs 70.4 GB but only 18.4 GB is addressable, so about 75% of the layers would stream from system RAM at roughly 60 GB/s. Expect around 2.7 tokens per second — usable for batch work, painful for chat.

25% on GPU 8K context Q4_K_M · 69.7 GB NVIDIA Open Model Released Mar 2026 Not in the Ollama library

The open-training-data 120B. Same hybrid layout as Lightning, so 128K context costs under a gigabyte.

What hardware do I need for Nemotron 3 Super 120B-A12B? →

The VRAM budget

weights 69.7 GB

Weights 69.7 GB KV cache @ 8K 0.06 GB Runtime overhead 0.6 GB Over budget 52.0 GB past 18.4 GB

Quantisation ladder

Quant	Weights	Total @ 8K	Max context	Tok/s	Quality	Fit
Q8_0	122.7 GB	123.4 GB	—	~1.4	−0.1% ppl	105.0 GB over
Q6_K	94.7 GB	95.4 GB	—	~1.8	−0.4% ppl	77.0 GB over
Q5_K_M	81.8 GB	82.5 GB	—	~2.2	−0.8% ppl	64.1 GB over
Q4_K_M	69.7 GB	70.4 GB	—	~2.7	−1.9% ppl	52.0 GB over
Q3_K_M	56.4 GB	57.1 GB	—	~3.6	−5.4% ppl	38.7 GB over
Q2_K	48.4 GB	49.0 GB	—	~4.5	−15% ppl	30.6 GB over

Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. Only 8 of its 88 blocks keep a per-token KV cache; the rest are linear-attention, Mamba or convolution blocks with a fixed-size state.

How to run it

llama.cpp LM Studio vLLM

terminal

$ llama-server \
    -hf nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16:Q4_K_M \
    -c 8192 -ngl 22

The engine underneath most of the others. Every knob is exposed. More on llama.cpp.

01Download is 69.7 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.

02Close anything else holding VRAM. A browser with hardware acceleration can sit on 1–2 GB.

03If it falls back to CPU silently, drop the context first, then step down a quantisation.

See all models for this rig Compare with another model