Will FastForward Be Fast Enough?

Unless you’ve been camped out in a doomsday fallout shelter, you must already be aware of the Department of Energy’s FastForward program awards, most of which were announced over the past few weeks by the vendors receiving those funding awards. Most, but not all. At least not that we’ve seen. The total amount handed out to date as part of the DOE’s FastForward program is $62.5 million. That would be mildly impressive if it was all going to one vendor – or to one specific topic of research. But that amount has been disbursed among five (5) different vendors, covering three research areas, and stretched out over two years.

Yes, we said five vendors. Not just the four vendors as being reported in the other outlets. Whittling down the list from a total of 26 responses to DOE’s RFP, of which 10 focused on processors, 7 on memory and 9 on storage, the funding awards went to (in alphabetical order), AMD, IBM, Intel Federal, NVIDIA, and Whamcloud.

The research to be conducted lines up as follows:

AMD:	processors and memory
IBM:	memory
Intel Federal:	processors and memory
NVIDIA:	processors
Whamcloud:	storage and I/O

There is a really important point that should be part of this discussion – the gem of the RFP that will help ensure a wider spread benefit to all of HPC – not just the emerging exascale community.

The following text comes directly from DOE’s post for the RFP.

The FastForward RFP objective is to initiate partnerships with multiple companies to accelerate the R&D of critical technologies needed for extreme-scale computing. It is recognized that the broader computing market will drive innovation in a direction that may not meet DOE’s mission needs. Many DOE applications place extreme requirements on computations, data movement, and reliability. FastForward seeks to fund innovative new and/or accelerated R&D of technologies targeted for productization in the 5–10 year timeframe.]

Now, first of all, vendors would say pretty much any technology they are researching is “targeted” for productization in the future – but actually knowing which research paths will lead to products 5-10 years out is nothing more than hope and wishful thinking. Still, the sentiment is in the right place.

Will FastForward be fast enough?

Well, it’s a start. But not everyone is thrilled. Politics and posturing are causing frustration among the labs and DOE headquarters, and there is no solid evidence that any sort of roadmap exists to get us to exascale. (See the community response section.) Speaking in very general terms, the issue seems to be that the labs would like to control their own research agendas, and not be handed their direction from the agency level. A number of national lab representatives I spoke with voiced this frustration. And to compound that issue, the vendors / manufacturers want to be the authority on dictating the roadmap for commercial productization and drive the HPC community to use whatever it is the vendors can build. After all, they are product-driven, not research-driven organizations.

Put this all together and you quickly realize, DOE is not getting a lot of love these days.

I suppose we should look at the FastForward program as a first step in helping us to better understand what a roadmap might actually look like. That’s something. No, it’s more than just something – it’s a commendable effort to keep things moving and to not lose any of the momentum started with the UHPC program.

But beyond DOE’s control, if the U.S. is to compete seriously in the race to exascale, funding must be increased at a significant rate – and before the FastForward program expires in two years.

The selections of AMD, IBM, Intel Federal and NVIDIA should not be a surprise to anyone in HPC. The selection of Whamcloud however took a lot of people by surprise. After all, the company is only two years old. But Whamcloud CEO, Brent Gorda, was no stranger to the agency managing the administration of the FastForward program, Lawrence Livermore National Lab (LLNL). Gorda earned his chops in HPC as a former LLNL staff member working on advanced technology projects, and before that as the Lead in the Future Technologies Group at Lawrence Berkeley National Lab. But that’s old news. Today, Gorda is an Intel employee.

Shortly after the FastForward awards were announced, Intel and Whamcloud announced that Whamcloud had been acquired by Intel, giving Intel an even bigger piece of the FastForward research budget. The folks at LLNL had no knowledge of the pending acquisition, which of course didn’t happen overnight, but say that it probably wouldn’t have made a difference in their award selection.

Ah yes, the playing field continues to evolve. Seems Intel is pretty darn serious about owning HPC once again.

What do you think? Is FastForward going to be fast enough? Send us your comments. We’d love to hear from you.

For related stories, visit The Exascale Report Archives.