My colleague Narayan Rangarajan, Altair Engineering team leader, has contributed the following guest post:
I recently read an interesting article that stated “two Google searches on a desktop computer produce 14g of CO2, which is roughly the equivalent of boiling an electric kettle.” While Google claims a much smaller CO2 cost, the math is inescapable. With over 2 million searches a day, it is “the server farms, not the animal farms, that will be increasingly responsible for global warming!”
Once thought of as a green alternate to physical prototypes, today, the carbon footprint due to CAE clusters is increasing exponentially. It is especially galling when one considers that roughly half of all CAE runs terminate with errors. In simple terms, more erroneous runs = a larger carbon footprint.
Does this mean engineers should stop computing? Stop using clusters? How can computing be made more energy efficient, and what makes green computing sustainable?
Well, to start off we need to adopt “The Toyota Way,” i.e. eliminate waste. This means to submit certifiably correct models to the cluster and obtain successful runs more often. The golden rule in this context: any model must be checked for correctness before submitting the job to the cluster. This principle has led to a new breed of tools in the CAE world – “Model Checkers.”
Model Checkers check the sanity of models in the desktop environment and inform users if there are modeling errors such as negative density, unrealistic modulus of elasticity or non-associated components. Today, industry is focusing on making Model Checkers more intelligent and ensuring that we get the runs right the first time.
To me, Model Checkers are the unsung heroes of the CAE world. Let us give them the credit they deserve.
Can more be done? Yes, of course. There are numerous opportunities to reuse the heat generated by compute farms in innovative ways. This is one of the ways in which “green computing” is achieved.