Home High Performance 2013: It’s time to stop chasing flops

2013: It’s time to stop chasing flops

by admin

2013: It's time to stop chasing flops
From the translator : Exascale computing – is such an ambitious project to achieve performance on the order of ExaFLOPS by 2018.There is an opinion that science-based computing is already cramped in petaflops.Is this really the case? Reflections on this topic by William Gropp, director of the Parallel Computing Institute, were published in The Exascale Report.
William D. Gropp

Everyone reading this believes in the power of computing technology. It seems self-evident to us that the performance of the most powerful computing systems must continue to grow at the same rate in order to meet the needs of society. Nevertheless, this is not so indisputable.
What’s worse, the emphasis on any measure of performance (not necessarily ExaFLOPS) instead of the emphasis on abilities solve pressing problems can (and does!) force us to focus on the technology rather than what can be achieved with that technology. In turn, projects like Exascale are pulling funding from, say, projects in Big Data
So, if we stop talking about productivity, what should we talk about then? For me it is obvious : we should discuss and describe the challenges and opportunities we face, from basic science to commercial solutions, and only then the important role that high-performance systems play in solving these problems. The problem itself must be clearly defined before its computational complexity, and only then can the need for faster computers be shown.
Too much effort to date has been put into chasing flops, when first we should be asking : what can we do with them? The danger is that approaches that don’t really need exascale – simplistic, suboptimal algorithms or distributed research – are being proposed in this way. This will do more harm than good, and even a layman can easily debunk the claim that such computing power is needed.
Don’t get me wrong, I believe that we do need much more powerful systems to solve the problems we face, whether it be understanding the functioning of life and the universe or developing sustainable infrastructures. But the need for them must come clearly from the problems.
We can start to move away from a focus on FLOPS (especially if it’s the result of benchmarks, which can be misleading, although the largest CS communities take them seriously) and focus on solving the hardest computational problems. Among other things, it provides a better rationale for developing the new technologies needed to create the increasingly fast machines we all believe are so necessary. Yes, without a simple metric like ExaFLOPS, it will be harder to quantify the result, but no one will argue that a high-performance system cannot be described by a single number.
For new record performance to become a reality, we must stop chasing performance.

You may also like