Supercomputer trend data

Karl Rupp has a nice blog series chronicling the developments in microprocessors over the past 42 years. He took the data originally compiled by M. Horowitz, F. Labonte, O. Shacham, K. Olukotun, L. Hammond, and C. Batten. I wanted to compare this data with parallel developments in supercomputing. So I noted down the number of cores in the #1 supercomputer of the Top500 going back to June 1993. The top machine at that time was Thinking Machines Corporation’s CM-5. In only 21 years, we’ve gone from the top machine having 140 cores (the Numerical Wind Tunnel) to having 10.6 million... Read more

Embedding FORTRAN in Python

I recently started developing a simple Python framework for playing around with neural nets and data assimilation. I wanted to see whether a neural net could be used as a surrogate for a traditional numerical model to perform the forecast step in an ensemble Kalman filter. I decided on Python because I wanted to use the ridiculously simple Keras framework to train the model. To train an accurate neural net, you need a lot of data. Now, all I was trying to do was to emulate an existing model as well as possible. I could use this existing model to... Read more