Back to search results
Llama-3.1-8B-Instruct · Q3_K_M
NVIDIA H2097GBllama.cpp v0.45L2 Cross-framework verified
This page aggregates 1 real-world runs of Llama-3.1-8B-Instruct (Q3_K_M) on NVIDIA H20 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
65.35 tok/s
Decode
Decode speed
226.54 tok/s
Prefill
Prefill speed
1.18 s
TTFT
Time to first token
13.52 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 · Q3_K_M (current)Decode 65.35 · Prefill 226.54 ·
- Ollama · IQ3_S Decode 56.57 · Prefill 208.48 ·
- Ollama · Q5_K_M Decode 43.67 · Prefill 144.84 ·
Core figures
- Decode (avg)
- 65.35 tok/s
- Prefill (avg)
- 226.54 tok/s
- TTFT (avg)
- 1.18 s
- VRAM (avg)
- 13.52 GB
- MTP acceptance rate
- —
- TTFB
- 0.52 GB
- Power draw
- 351.5 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Llama-3.1-8B-Instruct
- Quantization
- Q3_K_M
- Framework
- llama.cpp
- Version
- v0.45
- Context length
- 2048 tokens
- Batch size
- 512
- 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 H20
- Nominal VRAM
- 97 GB
- Measured VRAM (avg)
- 13.52 GB
- OS
- Windows 11 24H2
- Driver
- 560.94
- CUDA
- 12.6
- Power draw
- 351.5 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.45 · Windows 11 24H2 · CUDA 12.6 · 2048 ctx
65.35 tok/s
Decode
226.54 tok/s
Prefill
1.18 s
TTFT
13.52 GB
VRAM
—
MTP
351.5 W
Power draw
L2 batch seed 2026-10-19: llama.cpp on hw_id=56 running model_id=68 at Q3_K_M