The Future Will Be Simulated

By James Reinders, Intel

The effect of expanding computer power is that we will do more simulations. If you are not using simulation in your work today, perhaps you should add simulations to your solution mix.

Increased simulations enable us to continue to advance our abilities to integrate real data into models, to run more sophisticated models on that data and finally, to do deeper analysis of the validity of those models. Simulations will change the world more than most human endeavors; computer simulation has replaced physical laboratories as the most significant source of new discovery.

High-performance computing (HPC) offers a changing range of compute power, because yesterday’s supercomputer is today’s workstation or desktop supercomputer, which will be tomorrow’s game console or cell phone.

In one recounting of the history of computing, we can summarize that “Problem A inspires the building of Computer A. Computer A can do much more and inspires the attempt to solve Problem B. Problem B strains Computer A and inspires the building of Computer B. However, Computer B can do much more ” In this explanation of computing history, it is evident that the way we solve problems evolves continually. In this world, the question “For what do we need all this computer power?” will always get an answer.

Another way to think of it might be “computers are like electricity.” Electricity was a discovery that saw rapid development and deployment with only a small set of uses. Electricity became available widely to solve known problems. Then more problems were invented, which required more electricity. It is hard to imagine what early electric pioneers would think of hydroelectric dams, arc furnaces, iPhones or remote control of exploration vehicles setting on the surface of other planets. I think it is safe to say they were not building their early electric generators with those problems in mind.

Personally, I’m drawn to considering the explosion of computing that is upon us now, and the profound effect it will have on many areas, especially simulation.

Parallelism’s Role
For decades, the general trend in computing is that machines increase in performance about 40% per year. It is more common, and equivalent, to simply say that performance doubles about every two years. There is a notable exception to this: the HPC field has a strong history of increasing at 80% per year.

This faster pace is fueled by use of parallelism: parallel hardware and parallel programming to utilize it. With the emergence of multicore processors over the past six years, parallelism is taking hold in every niche of computing. It is highly significant, because it continues our version of computing history where computers just keep getting more powerful—and inspiring computing problems to reciprocate.

Desktop workstation computers, with more than 100 general-purpose x86 cores, will be very common and affordable within the next year. And honestly, I doubt any of us can understand quite what changes this will make.

Key Growth Area: Simulations
Simulations wrestle with at least three challenges:

1. Integrating real data into simulations. Weather prediction is one such area where the timely integration of real-world data is paramount to forecasting the weather instead of verifying it retroactively.

2. Running complex-enough models to do justice to solving the problem. Finer grained models, running competing models, or having more sophisticated equations are all considerations in how to refine and improve models. Models may be fully empirical, or rely on Monte Carlo method, or a hybrid. Simulations may even run both, or more than two methods and select the best. The more simulations we can run, the more we will run.

3. Verifying the validity of the models and their results. In many ways, this is the least-advanced area of simulation. There are enormous opportunities to do deeper analysis of simulation results, but the biggest opportunities lie in being able to compare multiple simulations (after all, we will be running more of them), and to look at more feedback from analysis into the models to refine and guide them.

The effect of advances in computer power on simulations is clearly more simulations—lots more. And more simulations will help us advance our abilities to integrate real data into models, run more sophisticated models, and dig deeper in our analysis of results from models.

I predict that the decade ahead will add a great deal of fuel that is driving simulations, and should prove to be the greatest decade of simulation results in our history.

James Reinders is a senior engineer and director at Intel, based in Hillsboro, OR. Contact him via [email protected].