Is there any AI performance of AM62 for reference?

AM62 has Quad-A53 cores and its SDK supports tflite runtime. See link below: https://software-dl.ti.com/processor-sdk-linux/esd/AM62X/08_03_00_19/exports/docs/linux/Foundational_Components_Machine_Learning.html

This allows AM62 to run tflite based Deep Learning models ( CNN’s, RNN’s) etc. Performance can be found in Section 4 of this appnote https://www.ti.com/lit/an/spracv1a/spracv1a.pdf  for AM62’s sister chip AM64 which has Dual A53 cores – note that those benchmarks are for A53’s at 1GHz. We will be posting a similar app note on AM62’s performance, and you can expect to see at least a 60% boost in performance boost although this may change based on the exact model.

Are there any particular applications for AI that you are interested in?

Hi Reese

The target application is Deep learning for falling detect (跌倒偵測) on bus like Vehicles.

Thanks

Daniel

Hi Reese

Does TI have any way to convert Tensor flow onto CPU?

Thanks

Daniel

Hi Daniel, good question.

Converting the model is not necessary -- when TensorFlow Lite runtime attempts to run inference using the model, it will run it on the CPU. This can run on multiple Arm cores by configuring the number of threads to use. You do not need to convert the .tflite model do this on AM62 to take advantage of this for the variants of AM62 with 2 or 4 A53 cores. The SDK image comes with TFLite.

Regards,

Reese