Arao AI
Back to search results

Qwen3-4B · IQ2_M

Start a discussion
NVIDIA Jetson AGX Orin 64GB64GBllama.cpp v0.52L2 Cross-framework verified

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

92.76 tok/s

Decode

Decode speed

306.43 tok/s

Prefill

Prefill speed

2.34 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 92.76 · Prefill 306.43 ·
  2. vLLM · IQ2_M Decode 96.3 · Prefill 512.46 ·
  3. vLLM · Q2_K Decode 95.78 · Prefill 454.64 ·
  4. Ollama · IQ3_S Decode 73.25 · Prefill 282.68 ·
  5. Ollama · Q4_K_M Decode 58.47 · Prefill 300.76 ·

Core figures

Decode (avg)
92.76 tok/s
Prefill (avg)
306.43 tok/s
TTFT (avg)
2.34 s
VRAM (avg)
6.47 GB
MTP acceptance rate
—
TTFB
0.74 GB
Power draw
43.6 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-4B
Quantization
IQ2_M
Framework
llama.cpp
Version
v0.52
Context length
2048 tokens
Batch size
512
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 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
43.6 W

Sources & evidence

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

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

    v0.52 · JetPack 6.0 · CUDA 12.6 · 2048 ctx

    92.76 tok/s

    Decode

    306.43 tok/s

    Prefill

    2.34 s

    TTFT

    6.47 GB

    VRAM

    —

    MTP

    43.6 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=76 running model_id=54 at IQ2_M