DepthAI documentation reports that 8 is the optimal number of SHAVEs to choose, when a model is converted to blob to execute on the RVC2 MyriadX Luxonis device. My question is, is there any case in which one should choose a lower number of SHAVEs?
Should one compile the model with a lower number of SHAVEs if his computer vision app runs multiple neural network concurrently?
Should one compile the model with a lower number of SHAVEs if he wants the device to reduce energy consumption?
Hi @brick
Optimal number of shaves depends on the pipeline you are running. If there are other processes (manips, videoEncoder, ..), they will take up the resources resulting in less available shaves.
brick Should one compile the model with a lower number of SHAVEs if his computer vision app runs multiple neural network concurrently?
The shaves are automatically distributed, so no.
brick Should one compile the model with a lower number of SHAVEs if he wants the device to reduce energy consumption?
That would work, since you are essentially stunting the performance of VPU, leading to less frames processed in the the same amount of time, so less energy expenditure.
Thanks
Jaka
