Skip to content
Run LLM Locally

Llama 4 Scout 17B-16E Instruct on the H100 PCIe

With 80 GB of VRAM, the H100 PCIe runs Llama 4 Scout 17B-16E Instruct at Q4_K_M entirely on the GPU with an 8K context (about 68.6 GB needed). At Q8_0 it does not fit in VRAM.

ShareFacebookXWhatsAppLinkedInEmail

Llama 4 Scout 17B-16E Instruct

Q4_K_M · 8K context

Compatible

The model needs about 68.6 GB and fits in the 77.6 GB available (9 GB left).

Your hardware
H100 PCIe 80 GB · 64 GB RAM
Estimated memory
~68.6 GB estimate
VRAM (usable memory)
77.6 GB
GPU only
Yes
CPU/GPU offload
Not required
Generation speed
76–115 tok/s estimate
Largest context in fast memory
191K tokens
Confidence: low

Based on: memory bandwidth (2,000 GB/s) and model size

Memory breakdown

~68.6 GB (66.9 GB–70.3 GB) needed · estimate

Fits entirely in VRAM · 9 GB left

  • Model weights62.9 GB
  • KV cache1.5 GB
  • Runtime overhead0.9 GB
  • Safety margin3.3 GB
Assumptions: context 8K tokens · KV cache f16 · llama.cpp (CUDA) · flash attention on · weights = real file size

EstimateData updated 24 days agoSource: Run LLM Locally engine, Hugging Face config.json + model card

What makes this estimate less reliable

  • no benchmark for this hardware and model yet
  • mixture-of-experts: speed depends on expert routing

Other quantizations on the H100 PCIe

Llama 4 Scout 17B-16E Instruct on H100 PCIe + 64 GB RAM · 8K context
Llama 4 Scout 17B-16E Instruct Q8_08K~113 GB · 3–6 tok/s estimateCPU/GPU offload
Llama 4 Scout 17B-16E Instruct Q6_K8K~89 GB · 9–17 tok/s estimateCPU/GPU offload
Llama 4 Scout 17B-16E Instruct Q5_K_M8K~77.4 GB · 70–105 tok/s estimateCompatible with limits
Llama 4 Scout 17B-16E Instruct Q4_K_M8K~68.6 GB · 76–115 tok/s estimateCompatible
Llama 4 Scout 17B-16E Instruct IQ4_XS8K~61.1 GB · 82–125 tok/s estimateExcellent
Llama 4 Scout 17B-16E Instruct Q3_K_M8K~55.7 GB · 87–134 tok/s estimateExcellent

Alternatives

GPUs that run Llama 4 Scout 17B-16E Instruct Q4_K_M entirely in VRAM

Other models that fit the H100 PCIe

Hardware that runs Llama 4 Scout 17B-16E Instruct Q4_K_M entirely in memory

Prices and links follow the market you choose.

Radeon PRO W7900

AMD

ExcellentRuns Llama 4 Scout 17B-16E Instruct Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

NVIDIA DGX Spark (GB10)

Unified memory PC

ExcellentRuns Llama 4 Scout 17B-16E Instruct Q4_K_M entirely in memory

Price not available

Availability unknown

View on amazon.com

Affiliate link

No current offer imported for this market: the link opens a marketplace search.

Frequently asked questions

Can the H100 PCIe run Llama 4 Scout 17B-16E Instruct?
The model needs about 68.6 GB and fits in the 77.6 GB available (9 GB left).
Which quantization should I use on the H100 PCIe?
Q5_K_M is the highest-precision quantization that fits entirely in VRAM at 8K context. ~77.4 GB / 77.6 GB.
What is the largest context on the H100 PCIe?
At Q4_K_M, about 191K tokens fit entirely in VRAM with an f16 KV cache. A q8_0 KV cache roughly doubles that.
How fast is Llama 4 Scout 17B-16E Instruct (H100 PCIe)?
We estimate 76–115 tok/s (estimate, confidence low). Based on: memory bandwidth (2,000 GB/s) and model size.