PIVlab memory usage

[Erich's Lab]

How much memory does PIVlab use when processing in serial or parallel? For example, what would be the memory usage for a 1280x800 pixels image with windowing of 64x64> 32x32> 16x16>12x12 with overlap of 50%>50%50%>75%? Would it be any higher than python with PyFFTW (python wrapper of FFTW) and 3D stacks of interrogation windows for vectorization?

On a side note, does PIVlab store everything on RAM or on disk with .mat files when batch processing?

PS, sorry if the question is unanswerable. I just don't have access to MATLAB to try it my self yet :D

[William Thielicke]

Hi Erich,

I can only give a rough guess using the task manager.:

Before starting matlab: 5.3 gb in use
After starting Matlab: 6.2 GB in use

pivlab started: 6.4 gb

Analyzing 10 1 MP images: Oscillating up to 6.9 GB

Analyzing 10 10 MP images: Oscillating up to 8.8 GB

But this is not with explicit parallel processing. In parallel processing, Matlab seems to simply double the RAM that is used

I am not storing anything during or after processing on the hard disk.

[Erich's Lab]

Thanks, my current laptop won't be able to handle anything like that (hence I created a Python GUI using HDF5 as a database and leaves RAM for processing only). I only have at most 2.8 GB of DDR3 RAM available when everything is closed except the GUI. :(

On your 1 MP images, oscillations of > 0.5 GB seems quite high. On my software, I get oscillations of up to 0.4 GB with windowing of 64x64>32x32>24x24 and overlap of 50%>50%>74%, totaling 27,300 vectors at 0.499 vec/ms on one core. Maybe I'm misunderstanding 3D FFTs with FFTW?