SPRACZ2 August   2022 TDA4VM , TDA4VM-Q1

ADVANCE INFORMATION  

  1.   Abstract
  2. 1Introduction
    1. 1.1 Vision Analytics
    2. 1.2 End Equipments
    3. 1.3 Deep learning: State-of-the-art
  3. 2Embedded edge AI system: Design considerations
    1. 2.1 Processors for edge AI: Technology landscape
    2. 2.2 Edge AI with TI: Energy-efficient and Practical AI
      1. 2.2.1 TDA4VM processor architecture
        1. 2.2.1.1 Development platform
    3. 2.3 Software programming
  4. 3Industry standard performance and power benchmarking
    1. 3.1 MLPerf models
    2. 3.2 Performance and efficiency benchmarking
    3. 3.3 Comparison against other SoC Architectures
      1. 3.3.1 Benchmarking against GPU-based architectures
      2. 3.3.2 Benchmarking against FPGA based SoCs
      3. 3.3.3 Summary of competitive benchmarking
  5. 4Conclusion
  6.   Revision History
  7. 5References

3.2 Performance and efficiency benchmarking

Performance measurements

In order to run MLPerf Deep Learning Models on TDA4VM starter kit hardware, they need to be converted into a format that is understood by deep learning accelerators in the DAv4M device. TI already has done all the work of converting, optimizing and exporting several models from the original training frameworks in PyTorch, Tensorflow and MxNet into these MMA friendly formats. All these pre-compiled and optimized models are hosted in TI’s GitHub repository [22].

The TI Edge AI software development kit (SDK) comes packaged a few pre-imported models as part of the SD card image. For quick evaluation, these are good to run the demos out of box [19].

The starter kit is used for performance benchmarking as it is much smaller and lower cost.

All the benchmarking is done using the Mlperf models and the procedure described in the MLcommons specification. Below table shows FPS, FPS/TOPS, Watts and FPS/Watts measured on the TDA4x EVM.

Table 3-2 TDA4VM performance and efficiency measurements
Model FPS FPS/TOPS
ResNet-50 162 20.29
SSD MobileNet-V1 385 48.08

The performance numbers are published on ti.com/edgeai. We can use these numbers to compare with other leading GPU and FPGA based architectures.