A year ago, we tried to predict what will happen in the HPC field in 2016. How did we do? Not that bad, actually.
(Image: Fuse/Thinkstock)
HPC in 2016: What we got right – and what we didn’t...

Forecasting is always hard, especially that of the future. However, we did give a try a year ago in our blog posts (part I, part II). Not all of the predictions hit the spot, but it does not prevent us from giving a new try.

We'll publish a new blog post for 2017, but let us begin with some quick comments on what we wrote one year ago:

Interconnect wars heat up

This prediction was not far-fetched, just too early, as the mentioned technologies entered the stage only towards the end of 2016. And yet one more year with no Ethernet for HPC. (We’ll get back to this topic later in our next post.)

Many-core processors making an impact

As for many other predictions, we were a bit too optimistic on the timeframe. In 2016, the first small test systems were installed, and developer platforms were made available, but the first large installations are only now becoming available. Early results show some promising results, but also proves code modernization is essential to take full advantage of the new architecture. This is work that we will also continue next year at CSC as an Intel Parallel Computing Center.

First steps toward practical quantum computing

Progress here did not take place as rapidly as anticipated: There were no new non-disclosed D-Wave systems installed in 2016. They introduced however a new 2048-qubit chip, and interest is steadily increasing also in Europe towards adiabatic quantum computing. For instance, the quantum computing workshop CSC organized in October was fully booked.

Containers make custom environments easy

This prediction was somewhat on target. Interest in, and experimentation with container technologies continued, but they are not in wide production use. The future still looks promising, but it may be that containers will first find traction in cloud HPC like environments.

Increased application I/O via burst buffering

This one failed; only few of the 2016 installations featured burst buffering in large scale. One of the most important ones, the Cray XC40 "Cori" at NERSC, did, however.

Alternative processors inching to the mainstream

No ARM-based HPC chip entered the market, and whilst the OpenPOWER effort is still very much alive, it was not the thing of 2016. However, the prediction itself was not too far-fetched, Intel's dominance is not as pronounced as it looked like one year ago.

On our next blog post we'll take a closer look at the HPC trends in 2017. Stay tuned...

Pekka Manninen

Pekka Manninen

Dr Pekka Manninen is a senior HPC specialist at CSC. He works with support and optimization of end-user applications on CSC supercomputers and does a lot of HPC training, and in general provides opinions in all things related or not related to HPC. Twitter:@pekkamanninen

Sebastian von Alfthan

Sebastian von Alfthan

Dr Sebastian von Alfthan is a senior HPC specialist at CSC with a talent for deep optimization of codes for extremely high-performance as well as exploring novel processor architectures. He has also worked at the Finnish Meteorological Institute developing the space plasma code, Vlasiator. Follow him on Twitter: @SvAlfthan

To front page
Trackback URL:

comments powered by Disqus