gallery icon indicating copy to clipboard operation
gallery copied to clipboard

Published 20 hours ago • verified publisher icon google-ai-edge

NPU acceleration

Open nonetrix opened this issue 6 months ago • 4 comments
trafficstars

Would be cool if this ran on NPU as well for battery efficiency and possibly speed(?)

nonetrix avatar May 21 '25 06:05 nonetrix

Sure thing! Many devices are coming with these nowadays...

Raphango avatar May 21 '25 14:05 Raphango

GPU seems to have been added, but correct me if I'm wrong I think NPU and GPU are separate? GPU is faster but NPU is more power efficient, that is the case with Qualcomms laptops iirc at least

nonetrix avatar May 22 '25 08:05 nonetrix

GPU seems to have been added, but correct me if I'm wrong I think NPU and GPU are separate? GPU is faster but NPU is more power efficient, that is the case with Qualcomms laptops iirc at least

NPU and GPU are separate devices.

NPU may go faster than GPU if you use Qualcomm QNN or MTK NeuroPilot framework to convert and deploy your model. But it's hard to write an efficient universal framework without reconvert the model for QNN/NeuroPilot because of the runtime/hardware limitation.

null-define avatar May 22 '25 12:05 null-define

@nonetrix Thank you for raising this feature request! We will keep tracking this feature request and share with the team. Thanks for the input!

dpknag avatar Jun 03 '25 21:06 dpknag