Arao AI
Back to search results

Llama-3-8B-Instruct · IQ3_S

Start a discussion
NVIDIA RTX 509032GBllama.cpp v0.31L2 Cross-framework verified

This page aggregates 1 real-world runs of Llama-3-8B-Instruct (IQ3_S) on NVIDIA RTX 5090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

315.36 tok/s

Decode

Decode speed

1,020.61 tok/s

Prefill

Prefill speed

1.99 s

TTFT

Time to first token

13.19 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 (current)Decode 315.36 · Prefill 1020.61 ·
  2. vLLM · IQ3_XS 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)
315.36 tok/s
Prefill (avg)
1,020.61 tok/s
TTFT (avg)
1.99 s
VRAM (avg)
13.19 GB
MTP acceptance rate
—
TTFB
2.52 GB
Power draw
375.6 W

Configuration

Member-level fields are taken from the most recent run

Model
Llama-3-8B-Instruct
Quantization
IQ3_S
Framework
llama.cpp
Version
v0.31
Context length
32768 tokens
Batch size
256
GPU layers
99
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)
13.19 GB
OS
Windows 11 23H2
Driver
572.83
CUDA
12.6
Power draw
375.6 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

    v0.31 · Windows 11 23H2 · CUDA 12.6 · 32768 ctx

    315.36 tok/s

    Decode

    1,020.61 tok/s

    Prefill

    1.99 s

    TTFT

    13.19 GB

    VRAM

    —

    MTP

    375.6 W

    Power draw

    L2 batch seed 2026-10-18: llama.cpp on hw_id=13 running model_id=69 at IQ3_S