High Performance Computing

How It Works

Artificial Machines designs custom HPC Computers based on RPi CM4 boards and Google Coral boards. There are many applications where you need a lot of isolated cores at a low cost, but don't need too much computing power per core. In these applications the need for dedicated Tensorflow processors is more important.

In under $1200/- you can get 52 isolated ARM cores running at 1.5GHZ, along with 104 Trillion OPS of Google Coral Tensorflow computing. While the ARM cores are themselves low power and have basic performance capability, they still provide opportunities of running frameworks like Kubernetes for isolated computing.

The Modular RPi4 and Google Coral Dual TPU design allows you to scale from a few Trillion OPS to over 100 Trillion OPS. Even the RPi cores can be scaled easily. This also allows for swapping boards to keep the system running all the time. RAM can go from 1GB to 8GB per RPi board (13GB to 104GB per HPC), and eMMC from 8GB to 32GB per RPi board (104GB to 416GB per HPC). SSD drives can be from 120GB to 2TB per HPC. Of course, you can fit 10 of these boards in a PC cabinet to give you 1 Quadrillion Tensorflow Operations per second with an additional 500 ARM Cores and 1 Terabyte of RAM, in about the power it takes to run a room Air Conditioner.