I was stunned by seeing Nvidia’s Tesla V100’s spec numbers. 14 tera flops single precision and 112 tera flops for deep learning.
At $10,000 price tag it isn’t cheap. But one server will suffice a hundred computers.
Need to get up to speed with ternsorflow. A lot to do.
Here is a picture of Cray II, running at 2Gflops, and cooled by liquid Nitrogen, which once we treated as supercomputer.