How much VRAM does Llama 4 Scout (17B active / 109B total) need at Q4_K_M? Native 10M Context
Full context is a multi-terabyte estimate even with expert offload. Reduce context before planning a small deployment, and verify engine support, offload behavior, and actual memory use.
At Q4_K_M and 10,485,760 tokens, this model estimates 2335 GB with all 109B parameters resident. The configured architecture has 48 layers, 8 KV heads, and head dimension 128. Each token only uses 17B parameters, but that doesn't shrink what has to stay loaded, because different tokens pick different experts. These are planning estimates, not guarantees for a particular inference engine.
By TechCompare · Updated
Calculator
Estimated VRAM required
2335 GB
109B params at Q4_K_M, 10,485,760 token context, batch 1, inference.
Estimate accuracy: Weights within ~2%. KV cache within ~5% for standard GQA models, ~10% for MLA or hybrid state estimates. Real VRAM may vary with framework (vLLM vs llama.cpp vs Transformers), Flash Attention, and driver overhead.
KV cache exceeds model weights: Consider lowering the context length to save on VRAM. Contexts between 8K and 64K are generally more typical for local setups.
Hardware that fits
Apple entries show installed unified memory separately from a model budget based on an 8 GB reserve. That budget is a planning heuristic, not a fixed hardware limit.
No single GPU in our catalog has enough memory. Multi-GPU or CPU offload required.
How this is calculated
Weights are 109B x 0.56 bytes = 61.04 GB. FP16 KV cache is 8 heads x 128 head dimension x 48 layers x 10,485,760 tokens x 2 (K and V) x 2 bytes = 2061.58 GB. The calculator adds 10% of weights plus cache as overhead: 212.26 GB, for 2334.89 GB total.
Verdict
Each token needs 196,608 bytes of FP16 KV storage under this model, or 196.6 GB per million tokens. The configured 10,485,760 tokens require 2061.58 GB of cache. Adding weights and 10% overhead gives 2334.89 GB resident. Dividing resident memory by nominal capacity gives at least 30 80-GB GPUs or 17 141-GB GPUs, before distribution constraints and additional runtime needs. At 32,768 tokens, the resident estimate drops to about 74 GB.
More Llama scenarios
Related guides
Frequently asked questions
What's iRoPE and why does it support 10M context?
Can Scout run on a single 24 GB GPU?
What about Maverick (17B/400B)?
Related tools
RAM Latency Calculator
Convert DDR3/DDR4/DDR5 timings (CL, tRCD, tRP, tRAS) into true latency in nanoseconds.
Use tool ➜Power Cost Estimator
Estimate annual electricity costs for your PC, Server, or TV.
Use tool ➜Data Transfer Calculator
Estimate transfer times for files over USB, WiFi, Ethernet, and more.
Use tool ➜Memory and Storage Throughput Visualizer
Visualize the massive speed difference between CPU cache, RAM, and storage.
Use tool ➜