Arao AI
Back to search results

Qwen3-1.7B · Q5_K_S

Start a discussion
NVIDIA RTX 309024GBllama.cpp v0.23L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-1.7B (Q5_K_S) on NVIDIA RTX 3090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

175.42 tok/s

Decode

Decode speed

578.57 tok/s

Prefill

Prefill speed

0.86 s

TTFT

Time to first token

5.31 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · Q3_K_L Decode 227.45 · Prefill 862.55 ·
  2. llama.cpp · Q5_K_S (current)Decode 175.42 · Prefill 578.57 ·
  3. Strata · Q6_K Decode 215.56 · Prefill 902.66 ·

Core figures

Decode (avg)
175.42 tok/s
Prefill (avg)
578.57 tok/s
TTFT (avg)
0.86 s
VRAM (avg)
5.31 GB
MTP acceptance rate
—
TTFB
1.4 GB
Power draw
418 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-1.7B
Quantization
Q5_K_S
Framework
llama.cpp
Version
v0.23
Context length
16384 tokens
Batch size
512
GPU layers
99
Flash Attention
Off

Hardware

Nominal and measured figures are shown side by side; whether it runs is the reader's call

GPU
NVIDIA RTX 3090
Nominal VRAM
24 GB
Measured VRAM (avg)
5.31 GB
OS
macOS 15.0
Driver
565.57
CUDA
12.6
Power draw
418 W

Sources & evidence

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

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

    v0.23 · macOS 15.0 · CUDA 12.6 · 16384 ctx

    175.42 tok/s

    Decode

    578.57 tok/s

    Prefill

    0.86 s

    TTFT

    5.31 GB

    VRAM

    —

    MTP

    418 W

    Power draw

    L2 batch seed 2026-10-10: llama.cpp on hw_id=2 running model_id=55 at Q5_K_S