That's right. People have been using clusters for high-performance computing for some time and now for training deep-learning neural networks.Please excuse my ignorance. I've seen people setting up Pi as clusters for handling multiple tasks, but how does that actually work ? is that the same thing as parallel processing ? is there any generic setup of multiple Pis for handling regular task faster ? as far as I understand there is a 'control unit' to determine which Pi to handle what task or something like that. I've very little knowledge of Mini and Mainframe that has multiple CPU cards, that can be added to improve performance
I've trained a neural operator model using GPUs to reproduce a certain lifting related to PDEs, but that was small scale and didn't involve a cluster. I have more experience with traditional HPC and describe setting up and using a cluster of Pi Zero's at
viewtopic.php?t=199994
In my experience cluster-based parallel processing doesn't increase the speed of computation but the size of what problems can be handled. Obtaining results can be frustrating and takes a combination of mathematical algorithms and engineering.
Unfortunately, it's difficult to know what a regular task is. The use of computers has changed and the greatest change will happen tomorrow. Said another way, if a regular task is using a mixture of AI experts to access other software today, what might tomorrow bring?
Statistics: Posted by ejolson — Sun Nov 09, 2025 7:32 am