Back to search results
Llama-3.2-3B-Instruct · FP16
NVIDIA H100 80GB HBM380GBllama.cpp v0.43L2 Cross-framework verified
This page aggregates 1 real-world runs of Llama-3.2-3B-Instruct (FP16) on NVIDIA H100 80GB HBM3 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
379.95 tok/s
Decode
Decode speed
1,874.2 tok/s
Prefill
Prefill speed
1.65 s
TTFT
Time to first token
13.76 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
16 days ago
Last verified
Performance
- llama.cpp · FP16 (current)Decode 379.95 · Prefill 1874.2 ·
- llama.cpp · Q5_K_M Decode 313.79 · Prefill 1508.71 ·
- Ollama · FP16 Decode 308.75 · Prefill 1127.72 ·
- Ollama · IQ2_XS Decode 295.44 · Prefill 1300.71 ·
Core figures
- Decode (avg)
- 379.95 tok/s
- Prefill (avg)
- 1,874.2 tok/s
- TTFT (avg)
- 1.65 s
- VRAM (avg)
- 13.76 GB
- MTP acceptance rate
- —
- TTFB
- 3.04 GB
- Power draw
- 590.5 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Llama-3.2-3B-Instruct
- Quantization
- FP16
- Framework
- llama.cpp
- Version
- v0.43
- Context length
- 32768 tokens
- Batch size
- 256
- GPU layers
- 33
- Flash Attention
- On
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- NVIDIA H100 80GB HBM3
- Nominal VRAM
- 80 GB
- Measured VRAM (avg)
- 13.76 GB
- OS
- Ubuntu 22.04
- Driver
- 570.86
- CUDA
- 13.0
- Power draw
- 590.5 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.43 · Ubuntu 22.04 · CUDA 13.0 · 32768 ctx
379.95 tok/s
Decode
1,874.2 tok/s
Prefill
1.65 s
TTFT
13.76 GB
VRAM
—
MTP
590.5 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=55 running model_id=70 at FP16