Five Predictions for HPC Clusters in 2017


2017-loading-1.jpgAs the year draws to a close, we have been thinking about what the HPC industry is likely to face in 2017. How will the role of HPC change in the modern data center? What new uses will HPC be put to? What about HPC in the cloud? We wanted to share our thoughts with you, and ask you to weigh in with your own predictions. Here’s what we think will happen in 2017:

#1 - More HPC in the cloud

Cloud providers will add more HPC capabilities to their offerings. That means more InfiniBand, more multi-core nodes, more GPU nodes, and more Xeon-Phi nodes. We also expect to see machine learning (ML) nodes with intelligent processing unit (IPU) accelerators. More HPC jobs will be offloaded to the cloud.

#2 – Energy efficiency will take center stage

Energy efficiency is already an important goal of most modern HPC environments, and its importance will continue to increase in 2017. Maybe the Green500 list will become as important as the Top500 for HPC bragging rights.

#3 – Machine learning hype will continue and excitement about adoption of IPU accelerators will increase

The increased interest in machine learning will bring with it an interest in ICUs, which are GPUs that are better suited for machine and deep learning. These accelerators will probably all be Peripheral Component Interconnect Express (PCIe). The growing interest in  machine learning will draw in people from outside the traditional HPC family, which means new investors, new expectations, and new capabilities.

#4 – Containerization will be widely adopted in HPC

Containerization will finally become mature enough (more secure, with more support from workload managers) to be widely adopted in HPC. It’s unlikely that  Docker will pay more attention to the high performance computing sector, and  some other containerization technology may well  replace it for HPC. It may be rkt, or Singularity – or maybe something completely new.

#5 – HPC in OpenStack will become commonplace

We think this is the year that running HPC in OpenStack will become routine. Greater integration with container technologies will help drive this trend.

Well, what do you think?

What do we have right? Where do you think we went wrong? What did we miss? Let us know by leaving a comment below.