Back to search results
Qwen3-4B-Thinking-2507 · IQ2_XS
NVIDIA Jetson AGX Orin 64GB64GBllama.cpp v0.50L2 Cross-framework verified
This page aggregates 1 real-world runs of Qwen3-4B-Thinking-2507 (IQ2_XS) on NVIDIA Jetson AGX Orin 64GB with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.
92.94 tok/s
Decode
Decode speed
357.53 tok/s
Prefill
Prefill speed
2.35 s
TTFT
Time to first token
6.13 GB
VRAM
VRAM usage
L2 Cross-framework verified
1
Measured runs
1
Independent sources
Other
Source platforms
5 days ago
Last verified
Performance
- vLLM · Q3_K_S Decode 89.8 · Prefill 485.57 ·
- vLLM · Q5_K_S Decode 57.15 · Prefill 278.01 ·
- Ollama · Q2_K Decode 77.04 · Prefill 283.31 ·
- llama.cpp · IQ2_XS (current)Decode 92.94 · Prefill 357.53 ·
- Strata · Q5_K_S Decode 83.78 · Prefill 413.16 ·
Core figures
- Decode (avg)
- 92.94 tok/s
- Prefill (avg)
- 357.53 tok/s
- TTFT (avg)
- 2.35 s
- VRAM (avg)
- 6.13 GB
- MTP acceptance rate
- —
- TTFB
- 0.74 GB
- Power draw
- 18.3 W
Configuration
Member-level fields are taken from the most recent run
- Model
- Qwen3-4B-Thinking-2507
- Quantization
- IQ2_XS
- Framework
- llama.cpp
- Version
- v0.50
- Context length
- 4096 tokens
- Batch size
- 512
- GPU layers
- 33
- 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)
- 6.13 GB
- OS
- JetPack 6.0
- Driver
- NVIDIA L4T 36.3
- CUDA
- 12.6
- Power draw
- 18.3 W
Sources & evidence
1 measured records in total, each traceable to its original source
v0.50 · JetPack 6.0 · CUDA 12.6 · 4096 ctx
92.94 tok/s
Decode
357.53 tok/s
Prefill
2.35 s
TTFT
6.13 GB
VRAM
—
MTP
18.3 W
Power draw
L2 batch seed 2026-10-13: llama.cpp on hw_id=76 running model_id=57 at IQ2_XS