Llama-4-Scout-109B-A17B (MoE) on NVIDIA DGX Spark GB10 (128GB Unified): VRAM Requirement & Speed Estimate
Yes. At Q4_K_M 4-bit quantization with a 16K context window, Llama-4-Scout-109B-A17B (MoE) requires 71.92 GB of VRAM (66.08 GB weights + 3.22 GB KV cache + 2.62 GB overhead), fitting within DGX Spark 128G's 108.8 GB usable memory at ~15.1 tokens/s.
Physical 128 GB · Load 66.1%
MoE: reads only 10.31 GB active weights/step
Enable Q8/Q4 KV quantization to scale batch concurrency
OLLAMA_CONTEXT_LENGTH=16384 ollama run llama4:scoutDGX Spark 128G Fits Llama-4-Scout-109B-A17B (MoE) · Cost & TCO
Approx. $3,999 (~$31/GB VRAM) · 170W TDP
📊 All Quantization Levels for Llama-4-Scout-109B-A17B (MoE) on DGX Spark 128G
| Quantization | Total VRAM (16K) | Verdict | Est. Speed | Max Context |
|---|---|---|---|---|
| FP16 / BF16 | 226.88 GB | Out of VRAM (OOM) | — | — |
| FP8 / Q8_0 | 122.65 GB | Out of VRAM (OOM) | — | — |
| GGUF Q6_K | 95.69 GB | Tight Fit (Watch Context) | ~10.7 tok/s | 81K |
| GGUF Q5_K_M | 83.6 GB | Smooth Fit | ~13.2 tok/s | 141K |
| Q4_K_M / AWQ | 71.92 GB | Smooth Fit | ~15.1 tok/s | 199K |
| GGUF IQ4_XS | 63.59 GB | Smooth Fit | ~16.9 tok/s | 240K |
| GGUF Q3_K_M | 58.86 GB | Smooth Fit | ~18.2 tok/s | 264K |
| GGUF Q2_K | 48.3 GB | Smooth Fit | ~21.7 tok/s | 316K |
💡 Recommended Alternative GPUs & Macs for Llama-4-Scout-109B-A17B (MoE)
Frequently Asked Questions
Can NVIDIA DGX Spark GB10 (128GB Unified) run Llama-4-Scout-109B-A17B (MoE) locally?
Yes. At Q4_K_M 4-bit quantization with a 16K context window, Llama-4-Scout-109B-A17B (MoE) requires 71.92 GB of VRAM (66.08 GB weights + 3.22 GB KV cache + 2.62 GB overhead), fitting within DGX Spark 128G's 108.8 GB usable memory at ~15.1 tokens/s.
What is the best quantization for Llama-4-Scout-109B-A17B (MoE) on DGX Spark 128G?
The highest-quality quantization that fits cleanly on DGX Spark 128G is GGUF Q6_K (Ultra High Quality), using 95.69 GB VRAM at ~10.7 tokens/s with up to 81K context.
How much VRAM does Llama-4-Scout-109B-A17B (MoE) need in full FP16 precision?
In unquantized FP16/BF16 precision (16K context), Llama-4-Scout-109B-A17B (MoE) requires 226.88 GB VRAM (218 GB for weights alone). Because it uses a Mixture-of-Experts (MoE) architecture, all 109B parameters must sit in memory, while only 17B active parameters are read per token.