Arao AI
Back to search results

Qwen3-1.7B · Q6_K

Start a discussion
NVIDIA RTX 509032GBllama.cpp v0.27L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-1.7B (Q6_K) on NVIDIA RTX 5090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

345.13 tok/s

Decode

Decode speed

1,779.25 tok/s

Prefill

Prefill speed

0.42 s

TTFT

Time to first token

5.67 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. llama.cpp · Q5_K_M Decode 353.77 · Prefill 1906.09 ·
  2. llama.cpp · Q6_K (current)Decode 345.13 · Prefill 1779.25 ·
  3. Ollama · FP16 Decode 199.27 · Prefill 1057.11 ·
  4. Strata · Q2_K Decode 300.65 · Prefill 1243.6 ·
  5. Strata · Q4_K_M Decode 280.21 · Prefill 1416.33 ·

Core figures

Decode (avg)
345.13 tok/s
Prefill (avg)
1,779.25 tok/s
TTFT (avg)
0.42 s
VRAM (avg)
5.67 GB
MTP acceptance rate
—
TTFB
2.76 GB
Power draw
480 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-1.7B
Quantization
Q6_K
Framework
llama.cpp
Version
v0.27
Context length
2048 tokens
Batch size
256
GPU layers
28
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)
5.67 GB
OS
Ubuntu 24.04
Driver
560.94
CUDA
12.8
Power draw
480 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

    v0.27 · Ubuntu 24.04 · CUDA 12.8 · 2048 ctx

    345.13 tok/s

    Decode

    1,779.25 tok/s

    Prefill

    0.42 s

    TTFT

    5.67 GB

    VRAM

    —

    MTP

    480 W

    Power draw

    L2 batch seed 2026-10-10: llama.cpp on hw_id=13 running model_id=55 at Q6_K