ENIAC Home - grumbiner/ENIAC GitHub Wiki
Welcome to the ENIAC wiki!
The ENIAC Numerical Weather Prediction model and challenge
Challenge1 is for building codes that implement the first successful Numerical Weather Prediction model -- (Charney, Fjørtoft and Von Neumann, 1950) https://onlinelibrary.wiley.com/doi/abs/10.1111/j.2153-3490.1950.tb00336.x. Fortran 77, Fortran 90, and C are already in place. There is also a reference result and a program to check differences between your run and the Fortran 77 standard results. It's also the case that the answers are unlikely to match exactly. The boundary conditions are unconditionally unstable, blowing up after approximately 2.5 days, and the model is very subject to the butterfly effect, so that even minor changes in numerics (compiler settings, languages, ...) produce observable differences in the results within the first day.
For historic views, see
- Platzman, 1979 The ENIAC Computations of 1950 -- Gateway to Numerical Weather Prediction
- Lynch, 2008 The ENIAC Forecasts: A Re-creation
Challenge2 is for the more substantial ****challenge -- writing the best model that takes only 10, or 1000, or 1 million, ... times as much compute as ENIAC. As a very rough ballpark estimate, you have easily 30 million times the computing power of the ENIAC, on one processor. The original calculation required almost 24 hours to carry out a 24 hour forecast (about 86,400 seconds). On one processor on my desk, it's about 86 milliseconds for 3 days, for something like 3 million times the speed.
On Model Evaluation and initialization:
- Teweles and Wobus 1954 S1 Skill Score
- Bolin 1956 Barotropic Model, Balance Equation
- Shuman 1957 i part 1 Balance Equation
- Shuman 1957 ii part 2 Smoothing and Filtering
- Cressman 1958 Divergence and Very Long Atmospheric Waves
- Cressman 1959 Objective Analysis