Llama 4 Scout 17B — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
The most accessible Llama 4 model. 17B active parameters in a 109B MoE architecture with a massive 10M token context window. All 109B parameters must be resident in memory even though only 17B activate per token, so it needs roughly 67 GB at Q4_K_M — an 80 GB datacentre card, a 96 GB+ unified-memory machine, or multi-GPU. Meta states it fits a single H100 at int4.
Llama 4 Scout 17B needs about 67 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
VRAM and speed by quantization
Quoted against NVIDIA RTX 4090 (24 GB). Includes the KV cache at 8K context, so it reads higher than the headline figure.
| Quant | Memory | VRAM | Speed (est.) | Fit |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 38.2 GB | ~14 tok/s | Offload | |
| Q3_K_M 3.41 bpw | 48.9 GB | ~12 tok/s | Offload | |
| Q4_K_M 4.83 bpw | 68.2 GB | — | Too big | |
| Q5_K_M 5.67 bpw | 79.7 GB | — | Too big | |
| Q6_K 6.56 bpw | 91.8 GB | — | Too big | |
| Q8_0 8.50 bpw | 118.2 GB | — | Too big | |
| F16 16.00 bpw | 220.4 GB | — | Too big |
Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. Llama 4 Scout 17B VRAM calculator →
Get Llama 4 Scout 17B running
The cheapest catalogued GPU that runs Llama 4 Scout 17B is the NVIDIA DGX Spark (128 GB).
or compare on Vast.ai from $0.77/hr (typical low · varies)
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
How to run Llama 4 Scout 17B
Install Ollama, then run:
ollama run llama4:scoutSpecifications
Curated — A hand-written entry from before this catalogue recorded its sources. The figures are long-standing but their provenance is not on file.
- Parameters
- 109 Billion (17B active)
- Context window
- 10,000,000
- Architecture
- MoE (16 experts)
- Provider
- Meta
- Licence
- Llama 4 Community
- Practical deployment
- Workstation (large unified memory or pro GPU)
- Specified at
- Q4_K_M
- System RAM
- 16 GB
- Record updated
- 2025-04-05
Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms.
Quality and use cases
Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.
Can I run Llama 4 Scout 17B on my GPU?
- Llama 4 Scout 17B on AMD Ryzen AI Max+ 395
- Llama 4 Scout 17B on Apple M1 Max
- Llama 4 Scout 17B on Apple M1 Pro
- Llama 4 Scout 17B on Apple M1 Ultra
- Llama 4 Scout 17B on Apple M2 Max
- Llama 4 Scout 17B on Apple M2 Pro
- Llama 4 Scout 17B on Apple M3 Max
- Llama 4 Scout 17B on Apple M3 Pro
- Llama 4 Scout 17B on Apple M3 Ultra
- Llama 4 Scout 17B on Apple M4
- Llama 4 Scout 17B on Apple M4 Max
- Llama 4 Scout 17B on Apple M5
- Llama 4 Scout 17B on Apple M5 Max
- Llama 4 Scout 17B on Apple M5 Pro
- Llama 4 Scout 17B on NVIDIA A100 80GB (PCIe)
- Llama 4 Scout 17B on NVIDIA DGX Spark
- Llama 4 Scout 17B on NVIDIA GeForce RTX 4090
- Llama 4 Scout 17B on NVIDIA GeForce RTX 5090
- Llama 4 Scout 17B on NVIDIA H100 80GB (PCIe)
- Llama 4 Scout 17B on NVIDIA L40S
- Llama 4 Scout 17B on NVIDIA RTX 6000 Ada Generation
- Llama 4 Scout 17B on NVIDIA RTX PRO 6000 Blackwell
Other Llama 4 sizes
Llama 4 Scout 17B — frequently asked questions
How much VRAM does Llama 4 Scout 17B need?
About 67 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does Llama 4 Scout 17B run on an RTX 4090 (24 GB)?
No. Llama 4 Scout 17B needs about 67 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.
How do I run Llama 4 Scout 17B locally?
Install Ollama and run `ollama run llama4:scout`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.
What other sizes does Llama 4 come in?
Llama 4 Scout 17B (67 GB), Llama 4 Maverick 17B (242 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.