TCL Smart TV Pro

4× Cortex-A73 @ 1.39 GHz, 32-bit userspace, 2.4 GB RAM (~700 MB free), Mali-G52 GPU, MDLA NPU via NNAPI (mtk-mdla_shim). Suspends ~3 min after power-off.

Manufacturer
TCL
SoC
MediaTek MT9689
OS
Google TV (Android 14)
RAM
2.4 GB
Accelerators
NPU, GPU, CPU

8 results

  • efficientnet-lite0-int8NPU · nnapi:mtk-mdla_shim

    6 msWorks

    invoke time

    version 1.0.0 · 2026-11-18

  • efficientnet-lite2-int8NPU · nnapi:mtk-mdla_shim

    9.4 msWorks

    invoke time

    version 1.0.0 · 2026-11-18

  • mobilenet-v2-int8NPU · nnapi:mtk-mdla_shim

    4.4 msWorks

    4.4–5.0 ms invoke

    version 1 · 2026-11-18

  • efficientnet-lite4-int8NPU · nnapi:mtk-mdla_shim

    22 msWorks with tweaks

    Tweak: Load the model from a memory-mapped buffer, not a path on shared storage: SELinux blocks the NN HAL (mtk_hal_neuralnetworks) from reading FUSE files, and NNAPI then silently falls back to its CPU reference path (~9 s per image).

    invoke p50 22 ms; full request (decode, render, pack, invoke) ~72 ms

    version 1.0.0 · 2026-11-18

  • mobilenet-v2-fp32NPU · nnapi:mtk-mdla_shim

    11.2 msWorks with tweaks

    Tweak: Enable allow_fp16 on the NNAPI delegate (setAllowFp16(true)).

    11.2 ms invoke with fp16 relaxation

    version 1 · 2026-11-18

  • clip-vit-b16Unknown · n/a

    –Fails

    fp32, 330 MB: out of memory while loading; the TV rebooted.

    version 1 · 2026-11-18

  • clip-vit-b16GPU · gpu (Mali-G52 delegate)

    –Fails

    The GPU delegate does not support the graph (unsupported CONCATENATION).

    version 1 · 2026-11-18

  • mobilenet-v2-fp32NPU · nnapi:mtk-mdla_shim

    –Fails

    The MDLA does not run fp32 graphs; without allow_fp16 NNAPI rejects the model for the NPU.

    version 1 · 2026-11-18

Open in the live table