pgagarinov

""" I was writing a dataloader from a video stream. I ran some numbers.
# in a nutshell. 
-> np.transpose() or torch.permute() is faster as uint8, no difference between torch and numpy
-> np.uint8/number results in np.float64, never do it, if anything cast as np.float32
-> convert to pytorch before converting uint8 to float32
-> contiguous() is is faster in torch than numpy
-> contiguous() is faster for torch.float32 than for torch.uint8
-> convert to CUDA in the numpy to pytorch conversion, if you can.
-> in CPU tensor/my_float is > 130% more costly than tensor.div_(myfloat), however tensor.div_()
does not keep track of gradients, so be careful using it.

pgagarinov

###
# Proxmox or other server kernel params cheap tune and secure.
# Try it if you have heavy load on server - network or memory / disk.
# No harm assumed but keep your eyes open.
#
# @updated: 2020-02-06 - more params used, adjust some params values, more comments on params
#

### NETWORK ###