AI accelerators: SoC packages and development

I conducted a 3 hour session on 'AI accelerators: SoC packages and development' to PhD students and Faculty . 

This was an opportunity for me to show hashtag#Khadas-Vim3-Pro #NPU, Cloud #TPU-VM, GPU and present #ANE(Apple Neural Engine) and MS Copilot+ PC with NPU. In terms of performance, these AI accelerators range from 5 to 1800 TOPS. Deployments can go from Edge to Cloud. 

After NVIDIA Jetson Nano, Khadas is the second SoC, I am able to experiment and share. I encourage students to try-out useful AI applications on these amazing piece of low-cost hardware. 

Though I have used #XLA in the past, I found its role in optimizing(padding, fusion, broadcast) the data layout to feed TPU Pods to be very unique. Single TPU #Coral(USB pluggable) SoC can be bought from Amazon is tempting. Thanks google for permitting me to use TPU PODs for free.