Arao AI
Back to search results

Llama-3-8B-Instruct · IQ4_XS

Start a discussion
NVIDIA H100 80GB HBM380GBllama.cpp v0.52L2 Cross-framework verified

This page aggregates 1 real-world runs of Llama-3-8B-Instruct (IQ4_XS) on NVIDIA H100 80GB HBM3 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

283.18 tok/s

Decode

Decode speed

981.02 tok/s

Prefill

Prefill speed

1.67 s

TTFT

Time to first token

14.53 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

16 days ago

Last verified

Performance

  1. llama.cpp · IQ3_S Decode 289.19 · Prefill 992.46 ·
  2. llama.cpp · IQ4_XS (current)Decode 283.18 · Prefill 981.02 ·
  3. vLLM · Q2_K Decode 349.23 · Prefill 1726.15 ·

Core figures

Decode (avg)
283.18 tok/s
Prefill (avg)
981.02 tok/s
TTFT (avg)
1.67 s
VRAM (avg)
14.53 GB
MTP acceptance rate
—
TTFB
2.27 GB
Power draw
566.9 W

Configuration

Member-level fields are taken from the most recent run

Model
Llama-3-8B-Instruct
Quantization
IQ4_XS
Framework
llama.cpp
Version
v0.52
Context length
8192 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 H100 80GB HBM3
Nominal VRAM
80 GB
Measured VRAM (avg)
14.53 GB
OS
Ubuntu 24.04
Driver
565.57
CUDA
12.8
Power draw
566.9 W

Sources & evidence

1 measured records in total, each traceable to its original source

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-19

    v0.52 · Ubuntu 24.04 · CUDA 12.8 · 8192 ctx

    283.18 tok/s

    Decode

    981.02 tok/s

    Prefill

    1.67 s

    TTFT

    14.53 GB

    VRAM

    —

    MTP

    566.9 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=55 running model_id=69 at IQ4_XS