Skip to content
Bitpute

How much VRAM does Llama 4 Scout 109B (17B active) Instruct need?

By Bitpute · Published 12 July 2026 · Updated 24 July 2026 · How we estimate · Sources · Editorial policy · Version history · Report an error · Figures re-checked against the calculation engine on every build

Llama 4 Scout 109B (17B active) Instruct needs about 65.4 GB of VRAM at Q4_K_M, or 114.0 GB at Q8_0 and 213.9 GB at FP16. No consumer card holds it at Q4_K_M. The smallest fit in our set is the A100 80GB (80 GB). Weights plus runtime overhead. Budget separately for KV cache, which at 1M is a material share.

Weights-plus-runtime footprint across the six most common quantizations. KV cache is context-dependent and comes on top. At 1M it is gigabytes on its own, before the weights; size it precisely in the GPU Memory Calculator.

This is a mixture-of-experts model: all 109B parameters must sit in memory, but only about 17B are active per token — so it loads like a 109B model and runs closer to a 17B one.

Parameters109B
Active / token17B
Context1000K
VendorMeta
LicenseLlama community
Released2025

Can your GPU run Llama 4 Scout 109B (17B active) Instruct?

Pick your card — the answer below is computed by the same engine that produces every figure on this page.

Yes — recommended
Estimated VRAM65.4 GB
GPU VRAM80 GB
Headroom10.6 GB

A100 80GB runs Llama 4 Scout 109B (17B active) Instruct at Q4_K_M with 10.6 GB to spare; comfortable at this size, where most cards cannot hold the weights at all; as a mixture-of-experts model all 109B parameters must be resident even though only 17B are active per token.

Weights plus runtime overhead. KV cache is not included — it depends on context length and this model’s published architecture.

Why
Weights (Q4_K_M)61.5 GB
Runtime overhead3.8 GB
Estimated total65.4 GB
Usable VRAM (95% of 80 GB)76.0 GB

Across every GPU we track

Estimates, not guarantees: real usage moves with runtime, driver, batch size and context. Size a specific context window in the GPU Memory Calculator.

How was this number calculated? Every figure here comes from Bitpute’s documented calculation methodology — parameters, precision, quantization, runtime overhead and usable VRAM, each formula written out in full.

Memory by quantization

QuantWeights+ runtime overhead
FP16203.0 GB213.9 GB
Q8_0107.9 GB114.0 GB
Q6_K83.2 GB88.2 GB
Q5_K_M72.2 GB76.6 GB
Q4_K_M61.5 GB65.4 GB
Q4_057.1 GB60.7 GB

Overhead = 0.75 GB + 5% of weights (CUDA context, buffers). Add KV cache on top. Weights dominate at this size, but a 1M window still costs gigabytes.

Which quantization should you actually run?

For Llama 4 Scout 109B (17B active) Instruct, the single-GPU entry point is Q4_K_M (65.4 GB) on A100 80GB, with plenty of headroom for long context. Higher quality (Q6_K 88.2 GB, FP16 213.9 GB) needs H200 141GB or multiple GPUs — for occasional use, renting is usually cheaper than buying. As a mixture-of-experts model it must hold all 109B in VRAM but activates only about 17B per token, so it runs much faster than its footprint suggests — memory is the limit here, not speed.

Single-GPU fit at Q4_K_M (65.4 GB + KV)

RTX 3060 12GBRTX 4060 Ti 16GBRTX 4070 Ti SUPERRTX 3090RTX 4090RTX 5090A100 40GBRTX A6000L40SA100 80GBH100 80GBH200 141GB

Size it exactly →Rent a GPU for it →

Where the memory goes

Memory budget for Llama 4 Scout 109B (17B active) Instruct on a A100 80GBStacked bar. Weights 61.5 GB, runtime overhead 3.8 GB, KV cache at 8K context 1.0 GB, against 76.0 GB usable on a A100 80GB.Memory budget for Llama 4 Scout 109B (17B active) Instruct on a A100 80GB76.0 GB usable of 80 GBWeights (Q4_K_M) — 61.5 GBRuntime overhead — 3.8 GBKV cache @ 8K — 1.0 GBHeadroom — 9.6 GB

Weights and overhead are exact for Q4_K_M. The KV bar is a generic 32-layer transformer at 8K; at this scale the weights dominate, but long context still adds gigabytes — size it in the calculator is for.

Deciding how much headroom to buy at this tier? A100 80GB vs H200 141GB compares them on bandwidth, power and which models each one holds.

Common questions

How much VRAM does Llama 4 Scout 109B (17B active) Instruct need?

Llama 4 Scout 109B (17B active) Instruct needs about 65.4 GB of VRAM at Q4_K_M, 114.0 GB at Q8_0, or 213.9 GB at FP16. That is weights plus about 0.75 GB of runtime overhead and 5% of weight size; KV cache is additional and depends on context length.

What GPU can run Llama 4 Scout 109B (17B active) Instruct?

At Q4_K_M the smallest card in our set that fits is the A100 80GB (80 GB usable at 95%). At FP16 no single GPU in our set is sufficient.

Can Llama 4 Scout 109B (17B active) Instruct run on 24 GB?

Not at Q4_K_M: it needs about 65.4 GB, more than the ~22.8 GB usable on a 24 GB card. Use a larger card, multiple GPUs, or offload.

More in the Llama 4 family

Llama 4 Maverick 400B (17B active) Instruct