Add measured Snapdragon NPU / GPU section (Galaxy S26)
Browse filesNumbers from the 2026-08-23 Galaxy S26 sweep. Primary log:
litertlm-convert/portal_work/bench_all.txt + bench_redo.txt, summarised in
portal_work/NOTES.md (2026-08-23, the 50-model NPU/GPU comparison).
Every row carries device, SoC, runtime, N and thermal state. Measurements
are joined to this repo by exact .tflite file name.
README.md
CHANGED
|
@@ -76,3 +76,19 @@ Measured on a **Pixel 8a** (Tensor G3, Android 16) with the standard TFLite [`be
|
|
| 76 |
|
| 77 |
**Any on-device figure recorded when this model shipped came from a different runtime.** It was taken through LiteRT's own `CompiledModel` accelerator (logcat reports it as `LITERT_CL`), which is the path the Kotlin sample app and the LiteRT API use, and it appears elsewhere on this card. The rows above are the classic TFLite OpenCL delegate, measured with a tool anyone can download and re-run. The two are not comparable, so read the rows above as a reproducible floor rather than as this model's speed on LiteRT.
|
| 78 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 76 |
|
| 77 |
**Any on-device figure recorded when this model shipped came from a different runtime.** It was taken through LiteRT's own `CompiledModel` accelerator (logcat reports it as `LITERT_CL`), which is the path the Kotlin sample app and the LiteRT API use, and it appears elsewhere on this card. The rows above are the classic TFLite OpenCL delegate, measured with a tool anyone can download and re-run. The two are not comparable, so read the rows above as a reproducible floor rather than as this model's speed on LiteRT.
|
| 78 |
|
| 79 |
+
## Snapdragon NPU (Hexagon)
|
| 80 |
+
|
| 81 |
+
- `nima_aesthetic_fp16.tflite` — the NPU is **1.52x faster** than the GPU (0.565 ms against 0.858 ms) and loads 3.61x faster (101 ms against 366 ms).
|
| 82 |
+
|
| 83 |
+
- `nima_technical_fp16.tflite` — the NPU is **1.54x faster** than the GPU (0.554 ms against 0.854 ms) and loads 3.74x faster (99 ms against 371 ms).
|
| 84 |
+
|
| 85 |
+
| file | backend | inference (median / min) | load |
|
| 86 |
+
|---|---|---:|---:|
|
| 87 |
+
| `nima_aesthetic_fp16.tflite` | NPU (Hexagon v81) | 0.565 ms / 0.537 ms | 101 ms |
|
| 88 |
+
| `nima_aesthetic_fp16.tflite` | GPU (Adreno) | 0.858 ms / 0.724 ms | 366 ms |
|
| 89 |
+
| `nima_technical_fp16.tflite` | NPU (Hexagon v81) | 0.554 ms / 0.537 ms | 99 ms |
|
| 90 |
+
| `nima_technical_fp16.tflite` | GPU (Adreno) | 0.854 ms / 0.726 ms | 371 ms |
|
| 91 |
+
|
| 92 |
+
Measured on a **Samsung Galaxy S26** (Snapdragon 8 Elite Gen 5 / SM8850, Hexagon v81, Android 16), LiteRT `CompiledModel` 2.2.0, one accelerator per process, 5 warm-up runs then N=50 timed runs, median reported. Every run held thermal status `NONE` throughout. Headroom 0.67-0.67, where 1.0 is the throttling threshold.
|
| 93 |
+
|
| 94 |
+
The NPU rows here ran artifacts compiled ahead of time for SM8850 with QAIRT 2.47.0; the GPU rows ran the published files as they are. LiteRT can also compile for the NPU on the device at first load, which is what lets you ship the published file unchanged — that path and the ten runtime libraries it needs are in the [NPU recipe](https://github.com/john-rocky/hf-to-litertlm/blob/main/docs/android-npu.md), and we did not measure it here. GPU wiring is in the [GPU recipe](https://github.com/john-rocky/hf-to-litertlm/blob/main/docs/android-gpu.md).
|