From: janburse@fastmail.fm   
      
   Hi,   
      
   It seems I am having problems pacing with   
   all the new fancy toys. Wasn't able to really   
   benchmark my NPU from a Desktop AI machine,   
      
   picked the wrong driver. Need to try again.   
   What worked was benchmarking Mobile AI machines.   
   I just grabbed Geekbench AI and some devices:   
      
   USA Fab, M4:   
      
        sANN    hANN    qANN   
   iPad CPU    4848    7947    6353   
   iPad GPU    9752    11383    10051   
   iPad NPU    4873    36544    *51634*   
      
   China Fab, Snapdragon:   
      
        sANN    hANN    qANN   
   Redmi CPU    1044    950    1723   
   Redmi GPU    480    905    737   
   Redmi NNAPI    205    205    469   
   Redmi QNN    226    226    *10221*   
      
   Speed-Up via NPU is factor 10x. See the column   
   qANN which means quantizised artificial neural   
   networks, when NPU or QNN is picked.   
      
   The mobile AI NPUs are optimized using   
   mimimal amounts of energy, and minimal amounts   
   of space squeezing (distilling) everything   
      
   into INT8 and INT4.   
      
   Bye   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)