Re:Math Breakthrough Speeds Supercomputer Simulations
by tehjoker on Saturday October 19, 2019 @06:25PM
(#21301115)
From a computer science perspective, a constant factor reduction isn't that interesting but it appears that what happened here is that the bottleneck part of the simulation can now be computed independent of time step (which is huge, algorithmicly speaking).
When working on the cutting edge of computing science problems, constant factor reductions can matter a lot. It can mean the difference between being able to do an experiment or not. If an experiment costs $10,000 of computation and your budget is $5,000, finding a 2x improvement is the difference between night and day.
That said, there are many situations where you need to do much better to make the impractical practical. However, problems are often chosen because they are just on the brink of tractibility, so the constant factor improvements matter more than you'd think especially in fields where the problems are chosen for e.g. biological interestingness vs computational interestingness.
When working on the cutting edge of computing science problems, constant factor reductions can matter a lot. It can mean the difference between being able to do an experiment or not. If an experiment costs $10,000 of computation and your budget is $5,000, finding a 2x improvement is the difference between night and day.
That said, there are many situations where you need to do much better to make the impractical practical. However, problems are often chosen because they are just on the brink of tractibility, so the constant factor improvements matter more than you'd think especially in fields where the problems are chosen for e.g. biological interestingness vs computational interestingness.