NVIDIANemotronNemotronHForCausalLM

NVIDIA Nemotron 3 Super 120B A12B BF16 — Hardware Requirements & GPU Compatibility

ChatFunctions

NVIDIA's Nemotron 3 Super 120B A12B is a mixture-of-experts model with roughly 120 billion total parameters, of which about 12 billion are active for any given token, the figure that primarily drives inference speed. Because the full expert set still needs to be resident in memory, running it locally needs multi-GPU or server-class hardware rather than a single consumer card, even though the active-parameter count keeps generation relatively fast for a model this large. It is tuned for general chat and instruction-following. The model is distributed in bfloat16 precision and supports a 256K token context window. It is released under NVIDIA's own license terms rather than a standard open-source license, and was published on March 10, 2026, as part of the Nemotron 3 series.

123.6B params · MoE13B active262K contextOtherMar 20261.1M downloads438 likes23.5K quant downloads

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

NVIDIA Nemotron 3 Super 120B A12B BF16 runs on a 96 GB+ Mac or a workstation GPU

Best local pick → Mac Studio (M3 Ultra, 96GB) · Q4_K_M · 74.7 GB · ~32 tok/s

Macs from 48 GB (IQ2_M · ~36 tok/s)

Score 95.8#71 of 106 open#33 of 62 that fit ≤96 GBOther

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

llmrun Composite Score →

95.8

90% range 86.0–105.6
#71 of 106 open models
#33 of 62 that fit ≤96 GB

Benchmarks · tier chart ↓

Among 106 open modelsHow the Composite Score is built

Coding95.690% range 86–106
AgentsNot enough data yet
MathNot enough data yet
Science97.690% range 75–120
ReasoningNot enough data yet

Run NVIDIA Nemotron 3 Super 120B A12B BF16

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Ways to runOther ways

each opens step-by-step

Run in cloud

Your own private copy · Q4_K_M · 96 GB GPU

The easiest way to try this model
Running in a few minutes, easy setup
Chat right here or connect any app

$5+ bonus for new users

Run NVIDIA Nemotron 3 Super 120B A12B BF16

74.7 GB VRAM at 16K context

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF:Q4_K_M
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and search nvidia/nemotron-3-super, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get nvidia/nemotron-3-super@q4_k_m
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF:Q4_K_M -c 16384 -ngl 99 --port 8080

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF:Q4_K_M
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF:Q4_K_M -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1.
$vllm serve nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 --max-model-len 16384
Runs the original weights, not the Q4_K_M GGUF; needs ~272 GB of GPU memory.

macOS only.

MLX supports only macOS, so follow these steps for that setup.

Install mlx-lm
$pip install mlx-lm
Chat in the terminal
$mlx_lm.chat --model mlx-community/NVIDIA-Nemotron-3-Super-120B-A12B-4bit
Downloads the model on first run. The MLX repo has its own quantisation; it may not match the quant selected above.
Or start a local server
Listens on http://localhost:8080.
$mlx_lm.server --model mlx-community/NVIDIA-Nemotron-3-Super-120B-A12B-4bit

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
RTX PRO 6000 Max-Q (96 GB)CheapestCheapest21 GB free	~86 tok/s	$4.29	Deploy on Vast.aifrom $1.33/hr (opens in new tab)	96 GB · 21 GB free · ~86 tok/s · $4.29 / 1M tokens
RTX PRO 6000 (96 GB)21 GB free	~86 tok/s	$4.94	Deploy on Vast.aifrom $1.54/hr (opens in new tab)	96 GB · 21 GB free · ~86 tok/s · $4.94 / 1M tokens
RTX PRO 6000 Server (96 GB)21 GB free	~81 tok/s	$5.82	Deploy on RunPodfrom $1.69/hr (opens in new tab)	96 GB · 21 GB free · ~81 tok/s · $5.82 / 1M tokens
RTX PRO 6000 Server (96 GB)21 GB free	~81 tok/s	$5.99	Deploy on Vast.aifrom $1.74/hr (opens in new tab)	96 GB · 21 GB free · ~81 tok/s · $5.99 / 1M tokens
H200 SXM (141 GB)66 GB free	~136 tok/s	$7.34	Deploy on RunPodfrom $3.59/hr (opens in new tab)	141 GB · 66 GB free · ~136 tok/s · $7.34 / 1M tokens
H200 NVL (141 GB)66 GB free	~136 tok/s	$7.71	Deploy on Vast.aifrom $3.77/hr (opens in new tab)	141 GB · 66 GB free · ~136 tok/s · $7.71 / 1M tokens
B200 (192 GB)117 GB free	~157 tok/s	$10.55	Deploy on RunPodfrom $5.98/hr (opens in new tab)	192 GB · 117 GB free · ~157 tok/s · $10.55 / 1M tokens
B300 (288 GB)213 GB free	~157 tok/s	$12.25	Deploy on RunPodfrom $6.94/hr (opens in new tab)	288 GB · 213 GB free · ~157 tok/s · $12.25 / 1M tokens

Compare GPUs →

Your own private copy of NVIDIA Nemotron 3 Super 120B A12B BF16 · Q4_K_M · OpenAI-compatible API

The pod’s first start downloads ~74 GB, billed while it loads.

Open the pod on RunPod
Deploy on RunPodRTX PRO 6000 · $2.49/hr (opens in new tab)
New to RunPod? Sign up and add credit first. If that GPU is gone, pick another with at least 96 GB.
2
Set the model and your key
Press Set overrides → Environment variables. The first variable is already there; set its value:
Key
LLAMA_ARG_HF_REPO
Value
unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF:UD-Q4_K_M
Press Add Environment variable for your key:
Key
LLAMA_API_KEY
Value
Make up your own key
A long password works. You’ll paste it below as the API key. Without it, anyone who finds the pod’s address can use it.Also set Container disk to 100 GB.Then Set overrides.
Deploy
Press Deploy Pod. Usually ready in 15–60 min.

Use it

RunPod → your pod → Connect → HTTP 8080, copy the link. The key is the one you made up in step 2.

Endpoint URL

API key

Chats and keys stay only in this browser.

https://<POD_ID>-8080.proxy.runpod.net/v1

RunPod → your pod → Connect → HTTP 8080 gives this address. The API key is the one from step 2.

from openai import OpenAI
client = OpenAI(base_url="https://<POD_ID>-8080.proxy.runpod.net/v1", api_key="YOUR_KEY")
r = client.chat.completions.create(
    model="any",
    messages=[{"role": "user", "content": "Hello!"}])
print(r.choices[0].message.content)

Works with any OpenAI client.

https://<POD_ID>-8080.proxy.runpod.net/v1

RunPod → your pod → Connect → HTTP 8080 gives this address. The API key is the one from step 2.

Free on Windows, macOS, Linux, iOS, Android and the web.

Open the web app ↗ (opens in a new tab)Download ↗ (opens in a new tab)

1. Settings → Model Provider → Add → Add Custom Provider
2. API Mode OpenAI API Compatible; paste the API key and the base URL as API Host
3. Model → Fetch, pick the model
4. New chat → pick the provider → start chatting

Free and open source for Windows, macOS and Linux.

Download Jan ↗ (opens in a new tab)

1.Settings → next to Model Providers press +
2.API format OpenAI-compatible; Base URL is the base URL above (with /v1); API key is your key; Create
3.If no model appears, click + in Models and type any
4.New chat → pick the model → start chatting

How Much VRAM Does NVIDIA Nemotron 3 Super 120B A12B BF16 Need?

Select a quantization level; the listed hardware adheres to it.

VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
34.5 GB	34.7 GB	35.8 GB	~34 GB	2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss	Does not fit: needs 34.5 GB
37.6 GB	37.8 GB	38.9 GB	52 GB	2-bit · i-quant extra small — Importance-weighted 2-bit, extra small	Does not fit: needs 37.6 GB
39.1 GB	39.3 GB	40.4 GB	52.1 GB	2-bit · i-quant small — Importance-weighted 2-bit, small	Does not fit: needs 39.1 GB
42.2 GB	42.4 GB	43.5 GB	52.7 GB	2-bit · i-quant medium — Importance-weighted 2-bit, medium	Does not fit: needs 42.2 GB
48.4 GB	48.6 GB	49.7 GB	56.6 GB	3-bit · i-quant — Importance-weighted 3-bit	Does not fit: needs 48.4 GB
51.5 GB	51.7 GB	52.8 GB	64.5 GB	3-bit · i-quant extra small — Importance-weighted 3-bit, extra small	Does not fit: needs 51.5 GB
53.0 GB	53.2 GB	54.3 GB	56.6 GB	3-bit · i-quant small — Importance-weighted 3-bit, small	Does not fit: needs 53 GB
53.0 GB	53.2 GB	54.3 GB	54.8 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Does not fit: needs 53 GB
54.6 GB	54.7 GB	55.9 GB	61.7 GB	3-bit small — 3-bit small quantization	Does not fit: needs 54.6 GB
56.1 GB	56.3 GB	57.4 GB	66.3 GB	3-bit · i-quant medium — Importance-weighted 3-bit, medium	Does not fit: needs 56.1 GB
60.7 GB	60.9 GB	62 GB	61.7 GB	3-bit medium — 3-bit medium quantization	Does not fit: needs 60.7 GB
62.3 GB	62.5 GB	63.6 GB	71 GB	4-bit · legacy — 4-bit legacy quantization	Does not fit: needs 62.3 GB
63.8 GB	64 GB	65.1 GB	66.4 GB	3-bit large — 3-bit large quantization	Does not fit: needs 63.8 GB
66.9 GB	67.1 GB	68.2 GB	64.5 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Does not fit: needs 66.9 GB
70.0 GB	70.2 GB	71.3 GB	64.5 GB	4-bit · i-quant non-linear — Importance-weighted 4-bit, non-linear	Does not fit: needs 70 GB
70.0 GB	70.2 GB	71.3 GB	79 GB	4-bit small — 4-bit small quantization	Does not fit: needs 70 GB
70.0 GB	70.2 GB	71.3 GB	76.6 GB	4-bit · legacy+ — 4-bit legacy quantization with offset	Does not fit: needs 70 GB
74.7 GB	74.8 GB	75.9 GB	82.5 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Does not fit: needs 74.7 GB
76.2 GB	76.4 GB	77.5 GB	87.4 GB	4-bit large — 4-bit large quantization	Does not fit: needs 76.2 GB
85.5 GB	85.7 GB	86.8 GB	89.7 GB	5-bit small — 5-bit small quantization	Does not fit: needs 85.5 GB
88.6 GB	88.7 GB	89.9 GB	107.3 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Does not fit: needs 88.6 GB
90.1 GB	90.3 GB	91.4 GB	96.9 GB	5-bit large — 5-bit large quantization	Does not fit: needs 90.1 GB
102.5 GB	102.6 GB	103.8 GB	114.8 GB	6-bit, very good — 6-bit quantization, very good quality	Does not fit: needs 102.5 GB
124.1 GB	124.3 GB	125.4 GB	128.5 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Does not fit: needs 124.1 GB
247.7 GB	247.9 GB	249 GB	241.5 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Does not fit: needs 247.7 GB

Selected Q4_K_M

4-bit medium quantization

74.7GBat 16K context

Weights: 74.2 GB
KV cache: 0.2 GB
Overhead: 0.3 GB

Doesn't fit RTX 4090 (reference) · needs 50.7 GB more

≥30 tok/s needs ≥424 GB/s · M3 Ultra ~32, M2 Ultra ~32

Or .

Which Hardware Can Run NVIDIA Nemotron 3 Super 120B A12B BF16?

Q4_K_M · 74.7 GB

NVIDIA Nemotron 3 Super 120B A12B BF16 (Q4_K_M) requires 74.7 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 98+ GB is recommended. Using the full 262K context window can add up to 2.8 GB, bringing total usage to 77.4 GB. No consumer GPU has enough memory. 30 devices with unified memory can run NVIDIA Nemotron 3 Super 120B A12B BF16, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio (M3 Ultra, 96GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8–16 GB · RTX 4060, RTX 3060 12GB, RTX 5060 Ti 16GBDoesn't fit — smallest quant needs 34.5 GB8 GB tier →12 GB tier →16 GB tier →

24 GB · RTX 4090referenceDoesn't fit — smallest quant needs 34.5 GB24 GB tier →

Mac 32 GB · Mac Mini M4Doesn't fit — smallest quant needs 34.5 GBMacs →

Mac 64 GB · Mac Studio M4 MaxIQ3_M · 56.1 GB · ~30 tok/s · 256K56.1~30256KMacs →

How good is NVIDIA Nemotron 3 Super 120B A12B BF16?

The llmrun Score merges 4 public benchmark boards into a single figure, where 100 represents the average for open models.

Score vs VRAM in the ≤96 GB tier

NVIDIA Nemotron 3 Super 120B A12B BF16
Other models
Best Score for its VRAM
x: VRAM (GB, log scale) · y: llmrun Score

#33 of 62 that fit ≤96 GB

Each dot is a model at the best quant that fits ≤96 GB. The line steps up to the best Score available at each amount of VRAM; a dot below it is beaten by a model that needs no more memory.

Full ≤96 GB ranking →

Benchmarks

LMArena Text
LMArena · Oct 2026Arena
1359.9
#70/188 open
Arena
SciCode
Epoch AI · Mar 2026Lab-reported
36.0%
#50/72 open
Lab-reported
WeirdML
Epoch AI · Mar 2026Independent
38.0%
#29/47 open
Independent
ALE-Bench
Epoch AI · Mar 2026Lab-reported
213.9
#47/48 open
Lab-reported
CritPt
Epoch AI · Mar 2026Lab-reported
3.1%
#28/84 open
Lab-reported

All 5 NVIDIA Nemotron 3 Super 120B A12B BF16 benchmarks

In the ≤96 GB class

62 models fit ≤96 GB

Same Score, only models that fit ≤96 GB at their best quant: the ranking a ≤96 GB owner actually chooses from.

#1
MiMo V2.6 Flash RLXiaomi
131.9
~153 tok/s · 93.8 GB
#2
DeepSeek V4 Flash 0731DeepSeek
127.6
~103 tok/s · 95.5 GB
#3
Qwen3.8 27BAlibaba
122.0
~67 tok/s · 17.9 GB
#4
Inkling SmallThinking Machines
119.6
~356 tok/s · 92.9 GB
#5
DeepSeek V4 FlashDeepSeek
115.3
~162 tok/s · 91.4 GB
#6
MiniMax M2.5MiniMax
111.8
~139 tok/s · 93.9 GB

Full ≤96 GB ranking →RTX PRO 6000 Blackwell Workstation Edition page →

Alternatives to NVIDIA Nemotron 3 Super 120B A12B BF16

Nemotron family · 45 models →

Official models with an llmrun Score, each answering one question; numbers are at the best quant that fits a 24 GB GPU.

Best alternative at this size
Mistral Medium 3.5 128B
Mistral AI
- 6.3 higher Score
- 8.2 GB more VRAM
Score 102.1 · 82.8 GB
Compare with NVIDIA Nemotron 3 Super 120B A12B BF16 Open Mistral Medium 3.5 128B
Other vendor
Qwen3 235B A22B Thinking 2507
Alibaba
- 12.6 higher Score
- 68.3 GB more VRAM
Score 108.4 · 142.9 GB
Compare with NVIDIA Nemotron 3 Super 120B A12B BF16 Open Qwen3 235B A22B Thinking 2507

Similar models

10tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
MiniMax M2.5MiniMax	111.8	139.6 GB	—needs > 24 GB	similar score
Qwen3.5 122B A10BAlibaba	109.4	75.5 GB	—needs > 24 GB	similar score
MiniMax M2.7MiniMax	107.4	139.6 GB	—needs > 24 GB	similar score
MiniMax M2.1MiniMax	106.2	139.6 GB	—needs > 24 GB	similar score
MiniMax M2MiniMax	104.0	139.6 GB	—needs > 24 GB	similar score
Qwen3 235B A22BAlibaba	100.0	142.9 GB	—needs > 24 GB	similar score
Qwen3 235B A22B Instruct 2507Alibaba	98.5	142.9 GB	—needs > 24 GB	similar score
Qwen3.5 4BAlibaba	98.5	3.4 GB	~201	similar score
Qwen3 30B A3B Instruct 2507Alibaba	96.7	19.4 GB	~173	similar score
GPT OSS 20BOpenAI	96.4	13.1 GB	~190	similar score

Fine-tunes & variants(2)community builds of NVIDIA Nemotron 3 Super 120B A12B BF16

Specifications

Publisher: NVIDIA
Family: Nemotron
Parameters: 123.6B
Active parameters: 13B per token
Architecture: NemotronHForCausalLM
Context Length: 262,144 tokens
Vocabulary Size: 131,072
Release Date: 2026-03-10
License: Other

Resources

Frequently Asked Questions

How much VRAM does NVIDIA Nemotron 3 Super 120B A12B BF16 need?

NVIDIA Nemotron 3 Super 120B A12B BF16 requires 74.7 GB of VRAM at Q4_K_M, or 247.7 GB at BF16. Full 262K context adds up to 2.8 GB (77.4 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 123.6B × 4.8 bits ÷ 8 = 74.2 GB

KV Cache + Overhead ≈ 0.5 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 3.2 GB (at full 262K context)

VRAM usage by quantization

Q4_K_M

74.7 GB

Q4_K_M + full context

77.4 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run NVIDIA Nemotron 3 Super 120B A12B BF16?

No — NVIDIA Nemotron 3 Super 120B A12B BF16 requires at least 34.5 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for NVIDIA Nemotron 3 Super 120B A12B BF16?

For NVIDIA Nemotron 3 Super 120B A12B BF16, Q4_K_M (74.7 GB) offers the best balance of quality and VRAM usage. Q4_K_L (76.2 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 34.5 GB.

VRAM requirement by quantization

IQ2_XXS

34.5 GB~53%

IQ3_S

53.0 GB~75%

Q3_K_L

63.8 GB~86%

Q4_K_M ★

74.7 GB~89%

Q4_K_L

76.2 GB~90%

BF16

247.7 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run NVIDIA Nemotron 3 Super 120B A12B BF16 on a Mac?

Yes — Mac Studio M5 Max (48 GB) and 55 other Macs can run NVIDIA Nemotron 3 Super 120B A12B BF16, the smallest at IQ2_M. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 74.7 GB of usable unified memory (RAM minus macOS overhead), from Mac Studio (M3 Ultra, 96GB).

Can I run NVIDIA Nemotron 3 Super 120B A12B BF16 locally?

Yes — NVIDIA Nemotron 3 Super 120B A12B BF16 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is NVIDIA Nemotron 3 Super 120B A12B BF16?

At Q4_K_M, NVIDIA Nemotron 3 Super 120B A12B BF16 can reach ~52 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)

Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.

Example: NVIDIA B200 → 7.9 GB active ÷ (8000 × 0.65) = 1.51 ms, plus 88 layers × 0.055 ms = 4.84 ms, so 1000 ÷ 6.35 ms = ~157 tok/s

Estimated speed at Q4_K_M (74.7 GB)

NVIDIA B200

~157 tok/s

NVIDIA B300

~157 tok/s

NVIDIA GH200 Grace Hopper Superchip

~137 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of NVIDIA Nemotron 3 Super 120B A12B BF16?

At Q4_K_M, the download is about 74.17 GB. The full-precision BF16 version is 247.22 GB. The smallest option (IQ2_XXS) is 33.99 GB.

Which GPUs can run NVIDIA Nemotron 3 Super 120B A12B BF16?

No single consumer GPU has enough VRAM to run NVIDIA Nemotron 3 Super 120B A12B BF16 at Q4_K_M (74.7 GB). Multi-GPU or professional hardware is required.

Which devices can run NVIDIA Nemotron 3 Super 120B A12B BF16?

30 devices with unified memory can run NVIDIA Nemotron 3 Super 120B A12B BF16 at Q4_K_M (74.7 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.

NVIDIA Nemotron 3 Super 120B A12B BF16 — Hardware Requirements & GPU Compatibility

NVIDIA Nemotron 3 Super 120B A12B BF16 runs on a 96 GB+ Mac or a workstation GPU

llmrun Composite Score →

Run NVIDIA Nemotron 3 Super 120B A12B BF16

Recommended

Ways to runOther ways

Run in cloud

How Much VRAM Does NVIDIA Nemotron 3 Super 120B A12B BF16 Need?

Selected Q4_K_M

Which Hardware Can Run NVIDIA Nemotron 3 Super 120B A12B BF16?

How good is NVIDIA Nemotron 3 Super 120B A12B BF16?

Score vs VRAM in the ≤96 GB tier

Benchmarks

In the ≤96 GB class

Alternatives to NVIDIA Nemotron 3 Super 120B A12B BF16

Mistral Medium 3.5 128B

Qwen3 235B A22B Thinking 2507

Similar models

Specifications

Resources

Frequently Asked Questions