LLM Inference on Qwen3 Samsung Galaxy S25 Ultra 0.6B (test)
1,709.9Prefill Throughput (min)llama.cpp
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| llama.cppHardware=GPU, GS=322026.05 | 1,709.9 | 1,718 | 77.8 | 78.8 | 442 | |
| ET VulkanHardware=GPU, GS=1282026.05 | 1,538.01 | 1,556.21 | 62.35 | 63.52 | 337 | |
| ET QNNHardware=NPU, GS=322026.05 | 1,462.86 | 1,542.17 | 61.02 | 62.38 | 681 | |
| ET VulkanHardware=GPU, GS=322026.05 | 1,206.42 | 1,246.45 | 57.98 | 58.23 | 457 | |
| ET XNNPACKHardware=CPU, GS=1282026.05 | 837.58 | 848.39 | 74.92 | 75.25 | 376 | |
| llama.cppHardware=CPU, GS=322026.05 | 747.7 | 750.4 | 97.9 | 100.8 | 442 | |
| ET XNNPACKHardware=CPU, GS=322026.05 | 716.62 | 732.59 | 72.34 | 72.73 | 417 | |
| ONNXHardware=CPU, GS=1282026.05 | 483.27 | 554.74 | 63.52 | 65.39 | 323 | |
| ONNXHardware=CPU, GS=322026.05 | 387.68 | 443.93 | 60.32 | 65.09 | 376 | |
| llama.cppHardware=NPU, GS=322026.05 | 343.1 | 409.1 | 22.5 | 23.4 | 442 | |
| LiteRTHardware=CPU, GS=1282026.05 | 172.57 | 173.22 | 12.56 | 13.23 | 299 | |
| LiteRTHardware=CPU, GS=322026.05 | 150.83 | 153.38 | 11.88 | 12.48 | 326 |