Moore’s Law Might Be Slowing Down, But Not Software Scalability

Based on a recent announcement that ANSYS and Cray has smashed supercomputing records, an editor of a well-known magazine followed up on and asked me whether this achievement might help to compensate the slowdown of Moore’s Law. Although I was able to briefly respond, it was also end of the day and while driving home the question stayed in my head and was the origin of this blog.

What’s Moore’s Law and Why Relevant?

GordonMoore_2_2005For those who don’t know, the law is named after Gordon E. Moore, co-founder of the Intel Corporation and Fairchild Semiconductor, whose 1965 paper described a doubling every year in the number of transistors per integrated circuit. In 1975, Moore revised the forecast doubling time to two years. Because this doubling of the number of transistors led to computer chips that could be packed with increasingly sophisticated circuitry that was both energy efficient and cheap. We all know that this led to the widespread adoption of computers, mobile phones, and the information technology revolution. So you can say (as also confirmed by many economists) that Moore’s law has been a driving force of technological and social change, productivity, and economic growth throughout the late twentieth and early twenty-first centuries.

We all can imagine that it took a tremendous engineering effort, by Intel and others, to make this “law” come true and to keep the two-year cycle. Earlier this year Intel signaled a deviation from this cadence which might imply the slowdown of Moore’s Law. Others claim that due to fundamental physical constraints, the end of Moore’s Law is on the horizon. I will not start speculating here and no one can say exactly when the era of Moore’s Law will come to a close (if ever). I can only pretend that innovations in other areas, such as developments in software and hardware accelerators (like what we’re doing in the area of Intel Xeon Phi and NVIDIA GPUs), will help to pick up the slack in the short term.

In the longer term, there will be fundamental changes in the design of the classical computer required. The next substantive leap forward will be in computers with human-like cognitive capabilities that are also energy efficient. Next generation of quantum computers are also on the horizon and I can mention here the company D-Wave Systems that uses ANSYS engineering simulation for designing and building these computers.

mores law Software Scalability

ANSYS Fluent Parallel – Max Cores Scaled

But let’s go back to how we can help to compensate the slowdown of Moore’s Law short-term. Improving parallel scalability such as what we’ve announced last week is clearly an important way. Also here it took a tremendous effort, by HPC developers and others, to make this happen. And, to repeat what I’ve written in my previous blog, HPC partnerships such as with Cray, NCSA, and NERSC are crucial in this regard.

As a result of this intense focus on HPC software development, our decade-long trend shows that ANSYS Fluent scalability doubles every 14 months by average. In analogy what Gordon Moore recently said in an interview, please consider this “not a law, it is an observation and a projection”.

This entry was posted in High Performance Computing and tagged , , , by Wim Slagter. Bookmark the permalink.

About Wim Slagter

Wim is Director for High Performance Computing (HPC) and Cloud Alliances at ANSYS, Inc. In his role, he is also responsible for the overall design and execution of the global HPC & Cloud partner program within Corporate Marketing at ANSYS. Prior to that, Wim was the Lead Product Manager of HPC. Wim was also Corporate Product Manager for the CFD product line and the Business Manager for Explicit Dynamics and Offshore. Before joining ANSYS in 2003, Wim worked at MSC Software since 1994 in various management positions related to software development, consulting, and sales. Wim holds a Ph.D. degree in Aerospace Engineering from the Technical University of Delft in the Netherlands.

One thought on “Moore’s Law Might Be Slowing Down, But Not Software Scalability

  1. You haven’t though of LED multiplexers ( like 8 or even 16 bits parallel data transmission through 1 fiber optic cable ) or tri-state data multiplexing.

    Just remember who said these things first!

    Art Blackwell

    Scientist, Test Engineer, Network Engineer, Network Security Engineer

    Cray Research, Inc.

Comments are closed.