How to Deploy Llama 4 Scout 17B 16E Locally
Llama 4 Scout 17B 16E is a MoE model with 108.6B total parameters and 17B active per token. At Q4 it needs 64GB of VRAM to sit entirely on the GPU; with experts offloaded to system RAM it needs only 10GB of VRAM plus 55GB of RAM. The cheapest card that runs it today is the Ryzen AI Max+ 395 128GB at about $3,913, at roughly 11.7 tokens/s. In offload mode the cheapest card that runs it is the Tesla V100 16GB, at roughly 12 tokens/s.
Llama 4 Scout 17B 16E details
How much VRAM does Llama 4 Scout 17B 16E need for local deployment?
MoE: all 108.6B parameters must stay resident in memory, while speed is set by the 17B active per token.
| Quantization | WeightsGB | Total at 8KGB | Total at 32KGB | Total at 128KGB |
|---|---|---|---|---|
| 1-bit | 35GGUF | 28 | 32.5 | 50.5 |
| 2-bit | 42.4GGUF | 42.9 | 47.4 | 65.4 |
| 3-bit | 49GGUF | 54.6 | 59.1 | 77.1 |
| 4-bit | 65.4GGUF | 63.1 | 67.6 | 85.6 |
| 5-bit | 76.5GGUF | 75.8 | 80.3 | 98.3 |
| 8-bit | 114.5GGUF | 115.1 | 119.6 | 137.6 |
| FP16 | 215.6GGUF | 215 | 219.5 | 237.5 |
Total = weights + KV cache (FP16) + 1GB runtime overhead. Weights marked GGUF are measured file sizes from the repository; the rest are estimated from bytes per parameter. Contexts beyond this model's 10M limit show a dash. Each tier takes whichever file of that bit width the repository has, preferring unsloth dynamic quants, so the exact file name varies by model.
Which GPUs can run Llama 4 Scout 17B 16E?
AmazonJP
| Device | VRAMGB | Price | Est. t/s | Suggested context |
|---|---|---|---|---|
 Ryzen AI Max+ 395 128GB | 128 | $3,9136 listings | 11.7 | 128K |
 DGX Spark 128GB | 128 | $8,2233 listings | 17.1 | 128K |
| CMP 170HX 64GB | 64 | —No listings | 71.5 | 8K |
| Instinct MI210 | 64 | —No listings | 59.7 | 8K |
| RTX PRO 5000 Blackwell 72GB | 72 | —No listings | 66.2 | 32K |
| RTX PRO 6000D | 84 | —No listings | 74 | 32K |
| RTX PRO 6000 Blackwell | 96 | —No listings | 81.2 | 128K |
| Mac Studio M4 Max 128GB | 128 | —No listings | 25.8 | 128K |
| Mac Studio M5 Max 128GB | 128 | —No listings | 28.6 | 128K |
| MacBook Pro M4 Max 128GB | 128 | —No listings | 25.8 | 128K |
| MacBook Pro M5 Max 128GB | 128 | —No listings | 28.6 | 128K |
| Mac Studio M3 Ultra 512GB | 512 | —No listings | 36.8 | 128K |
Fits when weights + KV cache + runtime overhead is at or below usable memory, single card. "Offload" on MoE models means it runs with expert weights in system RAM (only attention layers and KV cache stay in VRAM); that speed assumes 70 GB/s RAM bandwidth. The context column is the largest tier that fits.
Recommended GPUs for Llama 4 Scout 17B 16E with CPU/GPU offloading
AmazonJP
Expert weights live in system RAM (needs at least 55 GB); VRAM holds only attention layers, shared experts and the KV cache, at least 10 GB at Q4. Speed is bound by RAM bandwidth (estimated at 70 GB/s) and is far slower than a full-VRAM setup.
| Device | VRAMGB | Price | RAM neededGB | Est. t/s |
|---|---|---|---|---|
| Tesla V100 16GB | 16 | $659 | 55 | 11.6 |
| Radeon RX 9070 XT | 16 | $884 | 55 | 10.3 |
| GeForce RTX 5060 Ti 16GB | 16 | $924 | 55 | 10.3 |
| GeForce RTX 5070 | 12 | $956 | 55 | 11.1 |
| Radeon RX 7800 XT | 16 | $966 | 55 | 10.3 |
| Tesla V100 32GB | 32 | $1,141 | 55 | 11.6 |
 Arc Pro B60 | 24 | $1,164 | 55 | 9.2 |
| Radeon RX 7900 XT | 20 | $1,244 | 55 | 10.8 |
| Radeon RX 6950 XT | 16 | $1,404 | 55 | 10.1 |
| GeForce RTX 5070 Ti | 16 | $1,576 | 55 | 11.6 |
| Radeon RX 7900 XTX | 24 | $1,738 | 55 | 11.2 |
| GeForce RTX 5080 | 16 | $1,868 | 55 | 11.7 |
| Arc Pro B70 | 32 | $2,074 | 55 | 10 |
| GeForce RTX 4080 | 16 | $2,187 | 55 | 11.2 |
| Radeon AI PRO R9700 | 32 | $2,244 | 55 | 10.3 |
| GeForce RTX 3090 | 24 | $2,520 | 55 | 11.7 |
| RTX PRO 4000 Blackwell | 24 | $3,956 | 55 | 11.1 |
| GeForce RTX 4090 | 24 | $4,905 | 55 | 11.8 |
| GeForce RTX 5090 | 32 | $7,489 | 55 | 12.4 |
| GeForce RTX 2080 Ti 22GB | 22 | — | 55 | 11 |
| GeForce RTX 5090 D V2 | 24 | — | 55 | 12.1 |
| Arc Pro B65 | 32 | — | 55 | 10 |
| GeForce RTX 4080 SUPER 32GB | 32 | — | 55 | 11.3 |
| GeForce RTX 5090 D | 32 | — | 55 | 12.4 |
| Instinct MI100 | 32 | — | 55 | 11.6 |
| Instinct MI50 32GB | 32 | — | 55 | 11.3 |
| Radeon PRO W7800 | 32 | — | 55 | 10.1 |
| RTX PRO 4500 Blackwell | 32 | — | 55 | 11.6 |
| A100 40GB PCIe | 40 | — | 55 | 12.3 |
| CMP 170HX 40GB | 40 | — | 55 | 12.3 |
| GeForce RTX 4090 48GB | 48 | — | 55 | 11.8 |
| Radeon PRO W7900 | 48 | — | 55 | 11 |
| RTX PRO 5000 Blackwell 48GB | 48 | — | 55 | 12.1 |
Deploy Llama 4 Scout 17B 16E locally with Ollama, llama.cpp or vLLM
Ollama pulls and runs it in one command; the quantization comes from the official tag.
ollama run llama4:scout
llama.cpp pulls the GGUF straight from Hugging Face; -c sets the context length.
llama-server -hf unsloth/Llama-4-Scout-17B-16E-Instruct-GGUF:Q4_K_M -c 8192
vLLM serves the original-precision weights, which needs far more memory than GGUF and usually more than one GPU.
vllm serve meta-llama/Llama-4-Scout-17B-16E-Instruct
File names follow that week's model-library snapshot; the definitions are in the methodology. Methodology
FAQ
How much VRAM does Llama 4 Scout 17B 16E need?
At Q4_K_M with an 8K context Llama 4 Scout 17B 16E needs about 64GB of VRAM; 76GB leaves comfortable headroom.
What is the cheapest GPU that runs Llama 4 Scout 17B 16E?
The Ryzen AI Max+ 395 128GB: 128GB of VRAM, currently about $3,913, at roughly 11.7 tokens/s.
Can you run Llama 4 Scout 17B 16E with Ollama?
Yes: ollama run llama4:scout.