Arao AI
Back to search results

Qwen3.5-4B · IQ2_M

Start a discussion
NVIDIA Jetson AGX Orin 64GB64GBllama.cpp v0.20L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (IQ2_M) on NVIDIA Jetson AGX Orin 64GB with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

96.05 tok/s

Decode

Decode speed

357.59 tok/s

Prefill

Prefill speed

0.11 s

TTFT

Time to first token

6.47 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. llama.cpp · IQ2_M (current)Decode 96.05 · Prefill 357.59 ·
  2. Ollama · Q3_K_M Decode 81.64 · Prefill 331.85 ·
  3. llama.cpp · IQ3_S Decode 86 · Prefill 348.05 ·
  4. llama.cpp · Q3_K_M Decode 81.53 · Prefill 347.31 ·
  5. Strata · Q3_K_S Decode 126.8 · Prefill 658.95 ·
  6. Strata · Q5_K_S Decode 96.92 · Prefill 440.11 ·

Core figures

Decode (avg)
96.05 tok/s
Prefill (avg)
357.59 tok/s
TTFT (avg)
0.11 s
VRAM (avg)
6.47 GB
MTP acceptance rate
—
TTFB
0.77 GB
Power draw
24.2 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
IQ2_M
Framework
llama.cpp
Version
v0.20
Context length
4096 tokens
Batch size
512
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 Jetson AGX Orin 64GB
Nominal VRAM
64 GB
Measured VRAM (avg)
6.47 GB
OS
JetPack 6.0
Driver
NVIDIA L4T 36.3
CUDA
12.6
Power draw
24.2 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-16

    v0.20 · JetPack 6.0 · CUDA 12.6 · 4096 ctx

    96.05 tok/s

    Decode

    357.59 tok/s

    Prefill

    0.11 s

    TTFT

    6.47 GB

    VRAM

    —

    MTP

    24.2 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=76 running model_id=48 at IQ2_M