Arao AI
Back to search results

Qwen3.5-4B · Q4_K_M

Start a discussion
NVIDIA RTX Spark128GBllama.cpp v0.38L2 Cross-framework verified

This page aggregates 2 real-world runs of Qwen3.5-4B (Q4_K_M) on NVIDIA RTX Spark with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

90.26 tok/s

Decode

Decode speed

398.49 tok/s

Prefill

Prefill speed

0.8 s

TTFT

Time to first token

8.99 GB

VRAM

VRAM usage

L2 Cross-framework verified

2

Measured runs

1

Independent sources

Other

Source platforms

16 days ago

Last verified

Performance

  1. llama.cpp · Q4_K_M (current)Decode 90.26 · Prefill 398.49 ·
  2. llama.cpp · Q5_K_M Decode 71.83 · Prefill 346.83 ·
  3. llama.cpp · Q8_0 Decode 64.66 · Prefill 241.03 ·
  4. Ollama · FP16 Decode 40.67 · Prefill 196.29 ·
  5. Ollama · Q5_K_M Decode 69.06 · Prefill 289.88 ·
  6. Ollama · Q8_0 Decode 56.21 · Prefill 286.64 ·
  7. llama.cpp · Q3_K_S Decode 110.57 · Prefill 596.61 ·
  8. Ollama · Q4_K_M Decode 76.41 · Prefill 245.09 ·

Core figures

Decode (avg)
90.26 tok/s
Prefill (avg)
398.49 tok/s
TTFT (avg)
0.8 s
VRAM (avg)
8.99 GB
MTP acceptance rate
—
TTFB
0.73 GB
Power draw
72.85 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
Q4_K_M
Framework
llama.cpp
Version
v0.38
Context length
8192 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 Spark
Nominal VRAM
128 GB
Measured VRAM (avg)
8.99 GB
OS
Ubuntu 24.04
Driver
615.78
CUDA
13.0
Power draw
78.6 W
Reference price
$2,399 · updated 2026-10-19

Prices are not live; check the store page

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-20

    v0.38 · Ubuntu 24.04 · CUDA 13.0 · 8192 ctx

    96.11 tok/s

    Decode

    444.14 tok/s

    Prefill

    1.33 s

    TTFT

    8.99 GB

    VRAM

    —

    MTP

    78.6 W

    Power draw

    L2 batch seed 2026-10-20 (Linux 116GiB unlock): llama.cpp on Spark running model_id=48 at Q4_K_M
  2. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-19

    v0.53 · Windows 11 24H2 · CUDA 13.0 · 16384 ctx

    84.41 tok/s

    Decode

    352.83 tok/s

    Prefill

    0.28 s

    TTFT

    8.99 GB

    VRAM

    —

    MTP

    67.1 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=110 running model_id=48 at Q4_K_M