In our last issue, The Exascale Report broke the news that Richard Dracott, formerly the GM of Intel’s High Performance Computing organization, was taking a temporary position as the Acting Director of Intel’s European exascale labs.
Under Dracott’s leadership, Intel’s presence, or maybe we should say, saturation, among the Top 500, grew from 262 spots in 2006 to the current number of 406 entries. We are pleased to bring you this feature interview with Richard Dracott.
We recently noticed your recruitment ad for an exascale lab director. How many people does Intel anticipate employing in Europe as part of your exascale efforts?
Richard Dracott: Our plans this year are relatively modest, with a handful of Intel people at each lab. This would be in addition to researchers also coming from our partners at each lab. The plan is to roughly double the number of Intel people at the HPC labs in Europe over the next year (in the August issue of The Exascale Report, we reported that Intel has three exascale labs in Europe).
Do you feel there is more / appropriate talent in Europe than in the U.S.? Is this part of the reason Intel has invested so heavily in Exascale labs in Europe?
Richard Dracott: There is excellent experience and talent on both sides of the Atlantic, as well as in Asia. The exascale labs in Europe are just one of several vehicles we are using to engage the HPC research community around the globe. Other examples are the recently announced collaboration on UHPC with DARPA, or targeted research projects with specific customers or academia, which are not always made public. We will continue to look for additional such opportunities given the breadth of technical challenges that need to be addressed to not only reach exascale, but to do so at manageable cost and power levels.
Since you can’t hire someone with actual exascale experience, what do you look for when hiring Exascale development talent?
Dracott: The primary focus of all three labs is on how to scale today’s established and emerging technologies to overcome hurdles that will inevitably arise as we move to supercomputers three orders of magnitude more powerful and more complex than those we have today. So we are looking for people who not only have established HPC expertise, but in particular have experience at scaling recent solutions in relevant research areas. For example, tuning MPI libraries to scale up to tens of thousands of processes, or tuning applications and software tools to work effectively in machines with tens of thousands of nodes. The reality is that for the most part, the industry doesn’t even have many solutions that can even scale to today’s PetaScale — let alone exascale, so any research in this field could realize meaningful benefits long before we get to exascale.
What specific roles are you currently looking to fill in Europe?
Dracott: We have a couple of immediate management roles — one is to replace me as the European Lab Director when my temporary assignment wraps at the end of the year. Another is to be the Intel technical lead at the ExaTec collaboration lab in Paris. Both of these positions are posted publicly at: www.intel.com/jobs.
As you might expect, we are looking for people not with strong HPC experience, but a strong network within the European HPC community. In the near future you will also see openings for the individual researchers in both the Paris and Flanders labs
How would you describe where we are today in terms of building out a research roadmap for exascale?
Richard, Dracott, Intel
Dracott: I think we’re off to a good start, but this is something that will continue to evolve and grow. In large part this starts with identifying known obstacles and challenges and focusing the research on those. Beyond the software scaling already mentioned, examples include alternative fabric technologies, new approaches to memory subsystem design, such as multi tiered memory architecture, and more power efficient memory devices, improved processor and subsystem reliability and of course, fundamental processor performance.
As we march towards Exascale, we believe that hardware and software need to be designed in greater collaboration. Intel’s aim is for the labs to research the concept of co-design in order to ultimately help us develop machines that are more tailored towards the high end applications, and run them at higher levels of efficiency.
What do you have to say to the critics who think Intel should be creating these exascale labs in the U.S.?
Dracott: Like I said before, the exascale labs are just one vehicle we are using — UHPC is an example of an alternative specific to the USA. We’ll continue to deploy a number of alternative approaches in each geography according to what seems to be working best.
The European Union — primarily through PRACE — has been much more aggressive and committed to Exascale research and development than what we are seeing from any other countries. I make a point of using the words, “than what we are seeing” because I’ve been getting strong signals — but nothing I can verify — that China is investing heavily in this area. So, my question is, if the U.S. announces a national exascale initiative, will this spark some competition between the U.S. and Europe or will it bring them closer together?
Dracott: PRACE has done a good job at helping the EU member states to start working more in concert so that their collective investment effort might have greater impact, more comparable to that of the USA, or the growing effort in the PRC and other parts of Asia. PRACE has not so much focused on Exascale per se, although the recent IDC recommendations to the EU for FP8 include the suggestion that evolving the existing PRACE initiative would be a good starting point for an EU level exascale collaboration and exascale systems are anticipated within the FP8 time horizon. I am sure you will see similar announcements more specific to exascale in the USA. The timing of the announcements though is less significant than when real research and investment is occurring to move in the direction of exascale, and that is already well underway in multiple countries.
Over the years, Intel has been hot and cold when it comes to a commitment to HPC. Everyone knows this. Right now, most people are not sure where Intel stands with regards to HPC. What is Intel’s commitment to exascale — and why is exascale important to Intel?
Dracott: Intel is very serious about HPC and committed not just to helping enable exascale but growing the adoption of HPC from the workgroup to large datacenter scale deployments. Intel’s overall investment in HPC has increased steadily for more than a decade, and continues to grow, as demonstrated by the recent announcement of the MIC architecture — which represents an incremental investment of multiple hundreds of millions of dollars over the next several years.
For more than 10 years, Intel’s focus for HPC has been as a technology supplier to the HPC market segment rather than selling complete systems solutions directly. The late 90’s saw the emergence of clusters as a more cost effective and broadly available means of deploying HPC capabilities. Intel has continued to invest significantly in both the underlying processor capabilities, as well as the software tools and expertise required for HPC applications not only to migrate to cluster style deployments but also to run best on Intel processors. For example, the latest (and future) generations of Intel’s Xeon processor family all include features specifically added to improve the performance of a wide range of HPC applications, which is why you have seen a steady increase in the use of Intel processors in the Top 500 and beyond. At the same time we have continued to work closely with multiple companies building SMP class machines of varying sizes. We have also put a lot of effort into not just helping enable the highest end of supercomputing but also help broaden the adoption of HPC techniques in the commercial sector. Recent examples include the Intel Cluster Ready program, and our extensive involvement in a Missing Middle program addressing the manufacturing sector.
Exascale is important to Intel only in that it’s an inherent element of continuing to be a key player in the HPC business. It’s not an end in itself, rather it’s just another milestone in HPC advancement. HPC is important to Intel for many reasons. It is a strong growth segment that represents a steadily increasing proportion of our overall Data Center Group business. HPC users have always been the earliest adopters of many new technologies and by virtue of the increasing machine size, represent an increasingly large part of the early ramp of any new processor from Intel. They are also our most demanding users for increasing performance whilst maintaining power efficiency and programmer productivity. This means they have been, and will always be, a great source of feedback on how to improve our products. The longer lead HPC research not only informs our product and architectural definitions, but we believe HPC also represents opportunities to field new emerging technologies to customers who can both take best advantage of them and provide valuable feedback to evolve those technologies for future deployment in the broader high volume market segment.
As we march towards exascale, we believe that hardware and software need to be designed in greater collaboration. Intel’s aim is for the labs to research the concept of co-design in order to ultimately help us develop machines that are more tailored towards the high end applications, and run them at higher levels of efficiency.
One other thing to remember is that Intel is already a large user of HPC in its own right — some 85% of Intel’s servers are using HPC techniques to help design our processors and other products. The continuing increase in processor complexity is one example of what drives the continuing increase n performance demanded by HPC customers.
You recently stepped down from your position of GM for HPC at Intel. Who’s taking up the reins to continue Intel’s advances in HPC?
Dracott: The new HPC GM at Intel is Raj Hazra, is a 15 year Intel veteran. Prior to joining the HPC team in 2009, he ran the Systems Technology Lab in the Intel Labs under Justin Rattner (intel’s CTO). Since then he has been focused primarily on driving Intel’s exascale agenda and some of our external strategic government engagements, including the recently announced collaborative exascale labs in Europe. I was very fortunate to have Raj on the team and ready to take over the GM role from me, and the same strong team of experienced HPC veterans backing him up, such as Stephen Wheat, Bill Magro, Bill Camp and Dave Fenwick, to name just a few. He has obviously been ramping up in the new role over the last couple of months, and meeting many of our HPC customers. You are certainly going to be seeing a lot more of Raj in the near future!
If you are interested in talking to Richard about immediate opportunities, check out the postings at www.intel.com/jobs; specifically the HPC Exascale Lab Manager Paris (#584556) and the HPC Exascale Labs Director, Europe (#484553).
