Arao AI
Back to search results

Llama-3-8B-Instruct · IQ3_XS

Start a discussion
NVIDIA RTX 509032GBvLLM v0.54L2 Cross-framework verified

This page aggregates 1 real-world runs of Llama-3-8B-Instruct (IQ3_XS) on NVIDIA RTX 5090 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

354.4 tok/s

Decode

Decode speed

1,815.43 tok/s

Prefill

Prefill speed

0.95 s

TTFT

Time to first token

12.52 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · IQ3_S Decode 315.36 · Prefill 1020.61 ·
  2. vLLM · IQ3_XS (current)Decode 354.4 · Prefill 1815.43 ·
  3. llama.cpp · Q8_0 Decode 162.5 · Prefill 1218.8 ·
  4. llama.cpp · Q5_K_M Decode 236.9 · Prefill 1776.8 ·
  5. llama.cpp · Q4_K_M Decode 285.3 · Prefill 2139.8 ·

Core figures

Decode (avg)
354.4 tok/s
Prefill (avg)
1,815.43 tok/s
TTFT (avg)
0.95 s
VRAM (avg)
12.52 GB
MTP acceptance rate
—
TTFB
2.84 GB
Power draw
476.9 W

Configuration

Member-level fields are taken from the most recent run

Model
Llama-3-8B-Instruct
Quantization
IQ3_XS
Framework
vLLM
Version
v0.54
Context length
16384 tokens
Batch size
16
GPU layers
999
Flash Attention
Off

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)
12.52 GB
OS
Ubuntu 22.04
Driver
570.86
CUDA
12.8
Power draw
476.9 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-18

    v0.54 · Ubuntu 22.04 · CUDA 12.8 · 16384 ctx

    354.4 tok/s

    Decode

    1,815.43 tok/s

    Prefill

    0.95 s

    TTFT

    12.52 GB

    VRAM

    —

    MTP

    476.9 W

    Power draw

    L2 batch seed 2026-10-18: vLLM on hw_id=13 running model_id=69 at IQ3_XS