Accumulated TOPS ?
" Between the CPU, NPU, and integrated GPU, the House of Zen claims its top-specced chips are able to churn out 39 TOPS of inferencing performance for on-device AI/ML functionality. "
Ok...so 39 TOPS of AI/ML of combined CPU/GPU/NPU is nothing to sneeze at. On paper at least. Which of course is what the marketing department gets paid for. Until all those TOPS and the models get sophisticated enough to "downsize" the marketing department altogether.
That said, what is the TOPS figure of the Xilinx based NPU alone? With Apple with their M3 it's 18 TOPS alone. In the A17 Pro chip its NPU does 35 TOPS or so.
And what is out in the wild right now that a consumer could use where all three blocks (CPU/GPU/NPU) are used simultaneously in this new spanking AMD chippery ?