NVIDIANemotronGemma4ForConditionalGeneration

NVIDIA Nemotron 3.5 Bolt E4B BF16 — Hardware Requirements & GPU Compatibility

ChatFunctions

NVIDIA-Nemotron-3.5-Bolt-E4B-BF16 is a small language model from NVIDIA, post-trained from Google's Gemma-4-E4B-it and released in full BF16 precision. It targets local voice and device assistants, interactive personas, and embedded interfaces, with better multi-turn tool use, instruction following, and input grounding than the base model. It has about 8 billion parameters, 4 billion of them effective. A Q4 build needs roughly 5 GB of memory, and NVIDIA also publishes a separate 4 GB Q4_K_M release, so it runs on a modest GPU or a laptop. The context length is up to 64K tokens for input and output each. It is released under the OpenMDW-1.1 license, and the card says it is ready for commercial or non-commercial use. It was published on October 5, 2026. Its card reports 63.79 on IFBench against 42.84 for Gemma 4 E4B-it, and 68.51 on BFCL-v3. Thinking can be switched on or off in the chat template, and vLLM is the documented serving path.

8.0B params · dense66K contextOtherOct 2026249 downloads21 likes

Based on Gemma 4 E4B IT

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.

Runs great on a 24 GB GPU

Runs from 8 GB (Q4_K_M · 64K)Best on 24 GB (Q4_K_M · ~123 tok/s · 64K)Macs from 8 GB (Q2_K · ~18 tok/s)

Other

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run NVIDIA Nemotron 3.5 Bolt E4B BF16

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Other ways

each opens step-by-step

Run in cloud

Fits on RTX 3060 12GB (6 GB headroom) · Q4_K_M

Generation speed: ~44 tok/s; generation speed
Cost per 1M output tokens: $0.45; per 1M output tokens

Deployon Vast.aiBest optionfrom $0.071/hr

Compare GPUs →

Run NVIDIA Nemotron 3.5 Bolt E4B BF16

5.6 GB VRAM at 16K context · fits the RTX 4090 (18.4 GB free)

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/AtomicChat/NVIDIA-Nemotron-3.5-Bolt-E4B-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 paste https://huggingface.co/AtomicChat/NVIDIA-Nemotron-3.5-Bolt-E4B-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/AtomicChat/NVIDIA-Nemotron-3.5-Bolt-E4B-GGUF@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 AtomicChat/NVIDIA-Nemotron-3.5-Bolt-E4B-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/AtomicChat/NVIDIA-Nemotron-3.5-Bolt-E4B-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 AtomicChat/NVIDIA-Nemotron-3.5-Bolt-E4B-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-3.5-Bolt-E4B-BF16 --max-model-len 16384
Runs the original weights, not the Q4_K_M GGUF; needs ~18 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 DJLougen/NVIDIA-Nemotron-3.5-Bolt-E4B-MLX
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 DJLougen/NVIDIA-Nemotron-3.5-Bolt-E4B-MLX

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.
$AtomicChat/NVIDIA-Nemotron-3.5-Bolt-E4B-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 3060 12GBCheapestCheapest6 GB free	~44 tok/s	$0.45	Deploy on Vast.aifrom $0.071/hr (opens in new tab)	6 GB free · ~44 tok/s · $0.45 / 1M tokens
RTX 4060 (8 GB)2 GB free	~33 tok/s	$0.86	Deploy on Vast.aifrom $0.10/hr (opens in new tab)	8 GB · 2 GB free · ~33 tok/s · $0.86 / 1M tokens
RTX A4000 (16 GB)10 GB free	~55 tok/s	$0.56	Deploy on Vast.aifrom $0.11/hr (opens in new tab)	16 GB · 10 GB free · ~55 tok/s · $0.56 / 1M tokens
RTX 3070 (8 GB)2 GB free	~55 tok/s	$0.62	Deploy on Vast.aifrom $0.12/hr (opens in new tab)	8 GB · 2 GB free · ~55 tok/s · $0.62 / 1M tokens
RTX A5000 (24 GB)18 GB free	~94 tok/s	$0.47	Deploy on RunPodfrom $0.16/hr (opens in new tab)	24 GB · 18 GB free · ~94 tok/s · $0.47 / 1M tokens
RTX 3080 Ti (12 GB)6 GB free	~112 tok/s	$0.45	Deploy on RunPodfrom $0.18/hr (opens in new tab)	12 GB · 6 GB free · ~112 tok/s · $0.45 / 1M tokens
RTX 4000 Ada (20 GB)14 GB free	~44 tok/s	$1.26	Deploy on RunPodfrom $0.20/hr (opens in new tab)	20 GB · 14 GB free · ~44 tok/s · $1.26 / 1M tokens
RTX 3090 (24 GB)18 GB free	~114 tok/s	$0.53	Deploy on RunPodfrom $0.22/hr (opens in new tab)	24 GB · 18 GB free · ~114 tok/s · $0.53 / 1M tokens

Compare GPUs →

How Much VRAM Does NVIDIA Nemotron 3.5 Bolt E4B BF16 Need?

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

VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
3.2 GB	3.8 GB	—	3.5 GB	2-bit · i-quant medium — Importance-weighted 2-bit, medium	Fits: 64K context · ~204 tok/s
3.6 GB	4.2 GB	—	3.7 GB	3-bit · i-quant — Importance-weighted 3-bit	Fits: 64K context · ~181 tok/s
3.9 GB	4.5 GB	—	~3.4 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Fits: 64K context · ~167 tok/s
3.9 GB	4.5 GB	—	4 GB	3-bit · i-quant small — Importance-weighted 3-bit, small	Fits: 64K context · ~167 tok/s
4.4 GB	5 GB	—	~3.9 GB	3-bit medium — 3-bit medium quantization	Fits: 64K context · ~148 tok/s
4.8 GB	5.4 GB	—	4.5 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Fits: 64K context · ~136 tok/s
5.0 GB	5.6 GB	—	4.2 GB	4-bit · i-quant non-linear — Importance-weighted 4-bit, non-linear	Fits: 64K context · ~131 tok/s
5.3 GB	5.9 GB	—	5.1 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Fits: 64K context · ~123 tok/s
6.2 GB	6.8 GB	—	5.5 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Fits: 64K context · ~105 tok/s
7.1 GB	7.7 GB	—	7.4 GB	6-bit, very good — 6-bit quantization, very good quality	Fits: 64K context · ~92 tok/s
8.5 GB	9.1 GB	—	~8 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Fits: 64K context · ~77 tok/s
16.5 GB	17.1 GB	—	15.1 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Fits: 64K context · ~40 tok/s

est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

5.6GBat 16K context

Weights: 4.8 GB
KV cache: 0.5 GB
Overhead: 0.3 GB

Fits RTX 4090 (reference) · 18.4 GB free · ~123 tok/s

≥50 tok/s needs ≥409 GB/s · RTX 5090 ~219, RTX 3090 Ti ~123

Which Hardware Can Run NVIDIA Nemotron 3.5 Bolt E4B BF16?

Q4_K_M · 5.3 GB

NVIDIA Nemotron 3.5 Bolt E4B BF16 (Q4_K_M) requires 5.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 7+ GB is recommended. Using the full 66K context window can add up to 1.4 GB, bringing total usage to 6.7 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 151 devices with unified memory can run NVIDIA Nemotron 3.5 Bolt E4B BF16, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8 GB · RTX 4060Q4_K_M · 5.6 GB · ~33 tok/s · 64K5.6~3364K8 GB tier →

12 GB · RTX 3060 12GBQ4_K_M · 5.6 GB · ~44 tok/s · 64K5.6~4464K12 GB tier →

16 GB · RTX 5060 Ti 16GBQ4_K_M · 5.6 GB · ~55 tok/s · 64K5.6~5564K16 GB tier →

24 GB · RTX 4090referenceQ4_K_M · 5.6 GB · ~123 tok/s · 64K5.6~12364K24 GB tier →

Mac 32 GB · Mac Mini M4Q4_K_M · 5.6 GB · ~16 tok/s · 64K5.6~1664KMacs →

Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 5.6 GB · ~72 tok/s · 64K5.6~7264KMacs →

Alternatives to NVIDIA Nemotron 3.5 Bolt E4B BF16

Nemotron family · 44 models →

Similar models

6tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
ALoDLM 8BAmazon	—	7.6 GB	~119	same size
Llm Jp 4.1 8B Thinkingllm-jp	—	7.6 GB	~115	same size
Ling 3.0 TinyInclusion AI	—	5.5 GB	~364	same size
LFM2.5 8B A1BLiquid AI	—	5.6 GB	~335	same size
BitCPM CANN 8BOpenBMB	—	5.6 GB	~127	same size
Gemma 4 E4BGoogle	—	5.6 GB	~123	same size

Fine-tunes & variants(1)community builds of NVIDIA Nemotron 3.5 Bolt E4B BF16

NVIDIA Nemotron 3.5 Bolt E4B 4GB

Specifications

Publisher: NVIDIA
Family: Nemotron
Parameters: 8.0B
Architecture: Gemma4ForConditionalGeneration
Context Length: 65,536 tokens
Vocabulary Size: 262,144
Release Date: 2026-10-05
License: Other

Resources

Frequently Asked Questions

How much VRAM does NVIDIA Nemotron 3.5 Bolt E4B BF16 need?

NVIDIA Nemotron 3.5 Bolt E4B BF16 requires 5.3 GB of VRAM at Q4_K_M, or 16.5 GB at BF16. Full 66K context adds up to 1.4 GB (6.7 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 8.0B × 4.8 bits ÷ 8 = 4.8 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 ≈ 1.9 GB (at full 66K context)

VRAM usage by quantization

Q4_K_M

5.3 GB

Q4_K_M + full context

6.7 GB

Learn more about VRAM estimation →

What's the best quantization for NVIDIA Nemotron 3.5 Bolt E4B BF16?

For NVIDIA Nemotron 3.5 Bolt E4B BF16, Q4_K_M (5.3 GB) offers the best balance of quality and VRAM usage. Q5_K_M (6.2 GB) provides better quality if you have the VRAM. The smallest option is IQ2_M at 3.2 GB.

VRAM requirement by quantization

IQ2_M

3.2 GB~62%

IQ3_S

3.9 GB~75%

IQ4_NL

5.0 GB~88%

Q4_K_M ★

5.3 GB~89%

Q5_K_M

6.2 GB~92%

BF16

16.5 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run NVIDIA Nemotron 3.5 Bolt E4B BF16 on a Mac?

Yes — Mac Mini M2 (8 GB) and 145 other Macs can run NVIDIA Nemotron 3.5 Bolt E4B BF16, the smallest at Q2_K. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 5.3 GB of usable unified memory (RAM minus macOS overhead), from Mac Mini M2 Pro (16 GB).

Can I run NVIDIA Nemotron 3.5 Bolt E4B BF16 locally?

Yes — NVIDIA Nemotron 3.5 Bolt E4B BF16 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is NVIDIA Nemotron 3.5 Bolt E4B BF16?

At Q4_K_M, NVIDIA Nemotron 3.5 Bolt E4B BF16 can reach ~902 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~123 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 5.32 × 0.65 = ~977 tok/s

Estimated speed at Q4_K_M (5.3 GB)

NVIDIA B200

~977 tok/s

NVIDIA GeForce RTX 4090

~123 tok/s

NVIDIA B300

~977 tok/s

AMD Instinct MI350X

~902 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 3.5 Bolt E4B BF16?

At Q4_K_M, the download is about 4.80 GB. The full-precision BF16 version is 15.99 GB. The smallest option (IQ2_M) is 2.70 GB.

Which GPUs can run NVIDIA Nemotron 3.5 Bolt E4B BF16?

52 consumer GPUs can run NVIDIA Nemotron 3.5 Bolt E4B BF16 at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.

Which devices can run NVIDIA Nemotron 3.5 Bolt E4B BF16?

151 devices with unified memory can run NVIDIA Nemotron 3.5 Bolt E4B BF16 at Q4_K_M (5.3 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (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.