Python - Parallel Processing of complex Objects

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello dudes,

is it possible to process complex python objects in parallel with a combination of the
Jetson TK1 and something like nVidia PyCUDA? I have multiple objects that (should) simultaneously wait for input-parameters to arrive and to be processed and passed to another object.

For me the main advantage of this is not only the faster processing time, but also (very important) to keep the right order of processing the arriving inputs etc.

I want to use this for my (yet) python neural network with a number x of neurons that are all (simulated is sufficient, as long as the order is correct) simultaneously active. They wait for input (constantly active, object-related method) and if something arrives they wait a certain period of time for other inputs that arrive at the same time slot and consider that in the processing for the spike (access object related variables to send output to connected axons from other neurons)…

The plan is to run one python-neuron-object per available CUDA-Core.

Is this somehow possible with the Jetson TK1?

Thank you for support :)

2

I think it shoudl be possible to run pyCUDA on Jetson TK1

With GPUs, you don’t really get to identify what will run on a particular CUDA core, but you can write GPU kernels that essentially run “independent” threads, which could effectively mean running “one python-neuron-object per available CUDA-Core”. This is a fairly naive approach (at least what I have in mind) and may not yield the full power of the GPU, however.

3

Hello

@DJW24 I want to access Jetson TK1 GPU threads and grids using PyCuda.Is it possible to access GPU using PyCuda on Jetson TK1 board.