Back to search results
Llama-3-8B-Instruct · Q4_K_M
NVIDIA RTX 509032GBvLLM v0.44L2 Cross-framework verified
This page aggregates 1 real-world runs of Llama-3-8B-Instruct (Q4_K_M) on NVIDIA RTX 5090 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.
288.35 tok/s
Decode
Decode speed
1,282.34 tok/s
Prefill
Prefill speed
1.36 s
TTFT
Time to first token
15.88 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
17 days ago
Last verified
Performance
- vLLM · IQ3_XS Decode 353.35 · Prefill 1757.41 ·
- vLLM · Q4_K_M (current)Decode 288.35 · Prefill 1282.34 ·
- Ollama · Q5_K_M Decode 226.28 · Prefill 869.25 ·
- llama.cpp · IQ3_S Decode 315.36 · Prefill 1020.61 ·
- llama.cpp · Q8_0 Decode 162.5 · Prefill 1218.8 ·
- llama.cpp · Q5_K_M Decode 236.9 · Prefill 1776.8 ·
- llama.cpp · Q4_K_M Decode 285.3 · Prefill 2139.8 ·
Core figures
- Decode (avg)
- 288.35 tok/s
- Prefill (avg)
- 1,282.34 tok/s
- TTFT (avg)
- 1.36 s
- VRAM (avg)
- 15.88 GB
- MTP acceptance rate
- —
- TTFB
- 2.31 GB
- Power draw
- 387.6 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Llama-3-8B-Instruct
- Quantization
- Q4_K_M
- Framework
- vLLM
- Version
- v0.44
- Context length
- 2048 tokens
- Batch size
- 8
- GPU layers
- 999
- Flash Attention
- On
Hardware
Nominal and measured figures are shown side by side; whether it runs is the reader's call
- GPU
- NVIDIA RTX 5090
- Nominal VRAM
- 32 GB
- Measured VRAM (avg)
- 15.88 GB
- OS
- Windows 11 23H2
- Driver
- 565.57
- CUDA
- 12.4
- Power draw
- 387.6 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.44 · Windows 11 23H2 · CUDA 12.4 · 2048 ctx
288.35 tok/s
Decode
1,282.34 tok/s
Prefill
1.36 s
TTFT
15.88 GB
VRAM
—
MTP
387.6 W
Power draw
L2 batch seed 2026-10-19: vLLM on hw_id=13 running model_id=69 at Q4_K_M