I participated in many interesting discussions around exascale at the Newport HPCC conference.
No argument here. The two biggest constraints we face are funding and power. Significant levels of funding – from government and private industry R&D will be required, and wildly innovative approaches are needed if we are going to solve what some believe is an unsolvable power requirement. It’s too early to tell if either of these requirements can be met.
Price tag? By today’s metrics, it’s estimated the first exascale system will have a price tag of around $200 million, and roughly half of that cost will go to memory.
Who’s on first? And finally, does it matter who is first? A little more than half the attendees were of the opinion that there is indeed a “race” to achieving the first exascale system, and the bragging rights and national pride associated with fielding the first systems should not be taken lightly.
But the other viewpoint frequently expressed is that we are putting too much emphasis on exascale, and that it doesn’t matter who fields the first system because everyone will eventually benefit.
What about applications? The bigger concern I heard was that of how to develop an application portfolio able to take advantage of the early exascale systems. How do we engage with ISVs well in advance so that critical applications are available in the same timeframe as the hardware and system software?
Off to a rough start? A common opinion among attendees was that the IESP (International Exascale Software Project) is not going to be as effective as once hoped for, and the European Exascale Software Initiative (EESI) is only diluting the bigger picture global collaborative efforts. Opinions regarding any timely availability of exascale-class software ranged from pessimistic to weakly hopeful.
For related stories, visit The Exascale Report Archives.
