Hacker News Reader

I would guess the main goal is to perform parallel scientific computations on a shared chunk of data, not to run multiple web servers & services.

Seems pretty good if you compare it to a GPU.

The impressive thing is the memory bandwidth though. That's the one thing I've always loved GPUs for.

Perhaps it could be a 'high CPU' micro instance. 16 512MB instance with 3 cores (6 threads) per instance...