An Interview with Intel’s Extreme Scale Computing Director, Wilf Pinfold

Print Friendly, PDF & Email

wilfpThis article is not meant to be a literal transcript of the audio interview with Pinfold. The author has taken certain liberty to publish this article in the best way possible to capture the essence of the interview.

I met recently with Dr. Wilfred (Wilf) Pinfold, Director of Extreme Scale Computing research with Intel Labs. We talked about a number of issues centered around the race to exascale. Pinfold is cautiously optimistic about exascale R&D, but also warns us that now is the time to get it right. He brings a balanced perspective to the discussion of pursuing the most commercially viable path as opposed to a more risky path of R&D. We pick up the conversation with Pinfold as he discusses the importance of international competition in this global race.

PINFOLD: Regarding HPC and Exascale, we talk a lot about the competition that’s going on – and what we ought to understand is that science and society in general benefits greatly from the ability to use high performance computing to solve problems. And so we should say that what is really interesting is that worldwide – there is an interest in high performance computing. It’s actually very positive that the Chinese see it as important as they do. Basing the building of their society on scientific methods and the use of high performance computing to solve problems is beneficial for all of us.

The Exascale Report: So Wilf – we are coming to the end of 2011. We’re looking at budgets and R&D efforts for 2012. How do you see us wrapping up this year? Positive note – or negative note?

PINFOLD: Mike, I’m very optimistic with where we are around high performance computing. I think a very strong indicator for the future is the level of interest we’ve seen building in the last few years, and not just within the U.S. but worldwide, in the use of high performance computing to solve some of the big challenges we have in science, the environment, and in business. So I’m very optimistic about where we’re headed.

The structure of the industry is under challenge – we’ve had a number of discussions around the changes we see in technology – and I think that puts a challenge on the industry which is also very positive.

I’ve been in this industry for a very long time – as have you – and I think over the last few years we’ve seen a progression established – which is sort of a linear progression. But the disappointment of that is we haven’t seen any significant changes and improvements in technology that are disruptive in the way that we did maybe in the 80s – see some of those discontinuities. I see some of that now and I think there are challenges for us. I think it makes us all look very carefully at what we’re doing. And when we see that kind of disruption in the industry, I see people really interested and concerned in how things move forward. But I also see these as the periods of time that really generate innovation.

So, maybe it’s a mixed message, but I’m actually very optimistic.

I think we have very big challenges ahead. There is much more competition internationally for a leadership position, but I also think that kind of activity leads us to better innovation and faster progress, and I see that as a very good thing.

TER: So we focused a lot over this past year on issues such as power, performance, interconnect, and all the challenges that come with building technology that will drive us to excascale computing. But, we’ve kind of glossed over the budgetary issue. We’ve talked about funding in other countries and we’ve talked about the massive effort on China’s behalf and their ability to throw the kind of money that’s needed at this technology. And we’ve challenged U.S. policy makers with many different editorial pieces. But let’s take it down to a different level. You guys – meaning Intel – are building technology. If you come up with something that is absolutely stunning, breakthrough technology, but the price is just out of this world, don’t you think people will somehow find the money because they need to get to exascale.

PINFOLD: Well let me position where Intel is on this. We certainly have and will continue to invest in commercial wins in the industry, and I think we’ve done a good job of leading that path. I think part of the turmoil that we see in the future is that there may be multiple paths here. I think that the country can bank on Intel following the right commercial path.

The question that’s been raised – and discussed internally and within the community is, “Will that commercial path get us to exascale first?” That’s a very valid discussion to have.

There are several agencies that are forward looking here. They’ve made investments, and we’ve participated in those investments. Not just a dollar investment, but a time and intellect investment, trying to understand if high performance computing will diverge again from where the commercial market goes.

We owe it to ourselves to investigate that further. I think that divergence would be temporary if it is there at all. The enormous benefits of volume manufacturing some of the components in high performance computing have proven their worth. But in times when there are transitions in the industry, and there are discussions that this may be ahead of us, It’s important to have people look out beyond the horizon and do research and investigation that may not be directly in line with the commercial path. The challenge then for companies like Intel and the major stalwarts in this industry are to then bring those back together over time. And I’m very optimistic. I see exactly that going on – both within the company and across the industry with our partners. I see people experimenting. I see investments from DARPA – very forward looking investments. I see investments from DOE that are looking out into the future. Small investments admittedly, but they are trying to lay that path and trying to open up minds as to what the direction should be. In these times of transition there is always that tension between those who need to get their job done now – and those who see the problems that are caused by not doing things with a little more risk involved. And I see that as a healthy tension. I think we are going down both those paths.

Would we like a more stable, more robust funding environment? Absolutely. But I also don’t think that’s simply a high performance computing issue. Some of the instability we’re seeing in the funding environment is a little unnerving. I hope that will level out in 2012 and we’ll see a little more steadiness there. But, back to the optimistic side, I see some very innovative people doing the right kinds of things in trying to win this path.

TER: In earlier issues of The Exascale Report, we had several discussions in which we referred to “the race to exascale”, and a lot of folks came out and said it’s not a race – this is a global collaboration. But now we have this letter that has been signed by 24 U.S. senators appealing for more significant funding and referring specifically to “the race to exascale.”
What’s your personal position on that – and is your personal position the same as Intel’s?

PINFOLD: Well, that’s a really good question. So let me split it into two pieces. As a member of the global high performance computing community, I am really delighted that the Chinese are making significant investments in high performance computing. The alternatives are much worse. As a stockholder at Intel, I am also delighted they are making these investments. They are great customers of ours – across the board in both industry and in government. We’re delighted that this technology is being used to its fullest extent.

I’m an American by choice. I was not born here – I moved here. And I moved here because I believe in the foundations of the American society. And therefore, I view this personally as my team and I want to win. I want us to take the leadership position and push forward to be the first to exascale – to be the first to make these scientific breakthroughs. And I think we’re positioned very, very well to do that. We have the history. We’ve got brilliant people. We’ve got the technology. We have established investments. But I also think we’re moving into a very competitive environment. And I think enormous spoils will go to those who win this race – and I’d like to see them here.
I will do what I can within this community to help move us forward. But as I say, I’m delighted this technology is helping us solve global problems, and I will work with the global community to help the global community move forward here as well. It’s more of a team thing. I feel very positive that these investments are being made. If Asia was turning its back on science and technology, I think we would be in a terrible state. I think it is wonderful to see these significant investments. I think they are helping the industry in general and I think they are helping society in general.

TER: So, the approach that is being taken today, I think by some organizations – on the surface we say we want all this collaboration and we believe this is the right thing to do, but when it gets down behind the scenes there is really some aggressive competition.

Recently in the media, NVIDIA has been slamming Intel and Intel has taken a couple shots at NVIDIA. And it’s kind of fun – we haven’t seen this type of journalism for several years now in this community, but why recently do you think there is such a sparked competition between these two companies?

PINFOLD: I think this is part of the same issue. The path forward is not as clear as it may have been ten years ago. We take different paths – and that makes us see the world differently – and that makes us more energetic in how we try to persuade one party that the other one is wrong. And in a commercial environment – in a good capitalist environment, of course both companies are going to see the other as being on the wrong path. And they may well be.
We have tension with NVIDIA around our products – and rightly so. But actually we have really good collaboration around some of our longer term research. I think both companies see the health of the industry as important, and we see that many of our partners in the industry are common. We have no intention of undermining the value of the industry and the value of our partners by that competition. However, I think it is worth understanding that we do see the world differently, and that we have very different approaches, and we think ours is right. I’m sure they feel strongly that theirs is right – and there will be tension around that. And I think it’s actually quite healthy to have that tension.

TER: So, on the topic of programming. Programming these systems is a mystery – people can’t even describe it on a 30-foot white board at this point. Has there been any progress that you’ve seen? Do you feel encouraged that we will have some set of working applications at a level that makes it practical when we first see exascale-class systems?

PINFOLD: I think there is an interesting balance. We’ve certainly been looking at this closely. Over the last 20 years, since the foundation of the message passing architecture, we seen message passing mature and we’ve got a lot of codes in this space and it’s going to be the stalwart going forward for quite awhile. However, as we move to systems that can exploit more parallelism to get us to exascale, I think we need more. And I think that what that more is – is starting to become a little clearer. Certainly during 2011, I felt a lot more clarity – even between communities that had sort of vying approaches, the next step seems to be clearer.

What I think that is, and I’ll try to generalize this without picking one or the other, I think the idea that there will be more dynamic scheduling of tasks on machines, the idea that asynchronous messaging will become more predominant, and the idea that tasks need to be smaller and there need to be more of them, are fairly non-controversial at this point. When I talk to people throughout the industry, they tend to agree that those items are the things that we need to do. Now that’s a partial part of the problem. That will lead the path. As we move towards trying to both moving our applications to have more asynchronously scheduled tasks and as we move toward developing environments that allow us to do that more easily, I think we’ll see more innovation. And there’s a long way to go. I think things will move relatively slowly because there is such significant investment in where we are today. We have to build environments that comprehend that transition – the challenges that we’re giving our customers by trying to move in a direction that will give more explicit parallelism. And I think that is a challenge. But I’m very optimistic that we’re doing the right kinds of research. I believe we’ll get to the right place – and the issue is speed. And investment does that. I feel confident that we are moving in the right direction.

I wish it were faster. I don’t know how much additional funding would make it faster, versus the amount of intellect that needs to be applied to this. These things do take time. They take time both to incorporate the new methods – and then to use those new methods to move applications.

Where a couple of years ago I would not have felt there was a direction, I now feel that at least there are some common elements to the direction we are going in and they seem to be accepted within the MPI community, within the Open MP community, within the codelet community, within several of the different communities. We need to try to bring those together and make it easy for the application developers who have so much investment in where they are to sort of augment and move their applications in a fairly smooth way towards where we need to be with an exascale system in the 2019-2020 timeframe.

TER: It’s interesting. I had a conversation with a VC recently who pointed out that as we have these ten-year cycles, for example tera to peta, and now peta to exa, so much depends on the economic conditions around the globe at that particular point in time. He pointed out that it’s probably the worst possible time for us to be entering a new era that requires investments that we’ve never even dreamed of in this space. So, the burden of getting that R&D moving forward, while maintaining a business has to be back-breaking, and I think that’s what you were alluding to before – that you have to maintain commercial stability for a company like Intel while at the same time throwing massive amounts of money into R&D. But the point here is that some countries, China for example, seem to get that – or they have the ability to get that, but we don’t seem to have gotten to that point yet as far as the U.S. I see other collaborative efforts, for example, I know there is a massive effort going on in Russia right now with an initiative to go after exascale in a very big way – right down to the silicon level.

So, how do we compete? We’re going through such bitter times as far as politics, and the economics are just undermining everything, how do we really compete in that race?

PINFOLD: This is very important. Overall in the industry, there is an increase in funding because of the Chinese coming in to the market and spending more. So I think overall, if you are a global company, there are significant opportunities here and I think many of us are exploiting those, but I think you are pointing to exactly the right issue. Not only is there instability in the financial environment which makes it very hard to make these sort of risky decisions – if there’s already risk in your environment then you are less likely to take risks that might be technical and might move you forward. And this is true at a time at which it is no longer absolutely clear that America is the winner. We’ve lived with great comfort here. At the end of the day, whether cycles were up or down, it was mostly that we were so far ahead, and we may not have been where we wanted to be – but we were still leading. That’s been a comfort for a lot of people, and I think a lot of the tension that I see in the community at the moment is worry that this is no longer the case. I don’t think we’ve lost at all, but I do feel we have to get our act together. We’re in a real horse race now. This is a serious race. And we’ve got competitors who may be a long way behind, but they are catching up at an incredible rate, and they are making investments that mean their rate of improvement is extreme. If you project out any situation where the rate of one is higher and the rate of another is lower, over enough time, the one whose rate is higher, not the one whose position is better, will always take over. Those lines have to cross at some point or another. And we can all look out into the future and say if the current state stays there, we are not the winners.

I don’t think it will. I don’t think the current environment is sustainable over a long period of time. There are many reasons why we have financial turmoil, but I have faith that we will get through them and we will get back to a stable point and we’ll make the right investments to get our growth curve more aggressive. I think over time it becomes harder and harder for those who are catching up – as they have to take on more and more of the burden of innovation. When you are catching up in any cases, you are catching up by the knowledge of the ground that has been passed before. As you get closer and closer to that line crossing, you have to carry more and more of that burden of the innovation that has historically been America’s to carry – and that slows you down.

So, there’s good reason to be optimistic here, but I think we all want to get through this financial uncertainty period and get back to a state in which we’re making investments because they have real strategic value to our country and not be in the situation where we’re economically strapped and have to cut things we don’t want to.

TER: How about looking forward in 2012. What do you see as one or two of the milestone areas where we might see some definite progress, some impressive benchmarks, or milestones as we move forward in trying to build a good exascale R&D plan?

PINFOLD: I see good research investments planned for next year. I think they are all uncertain because of the financial environment that we talked about. I know there are people fighting for those budgets in this environment and they are the right people – and they are good people, and I feel comfortable that they will do everything they can to get those projects funded and moving forward. That’s really important. If we crimp off the innovative funds – the more risky funds – in an environment in which it’s tempting to do that, the longer term investments, the more risky investments, often are the ones that get cut first. As I said, I think people are trying to do the right thing, but we need to give them support in every way we can. It should not be seen that tension between vendors, or tension between countries indicate that money will be wasted. I think it’s quite the opposite. I think we absolutely need to look at that turmoil as a space in which innovation can happen and a space in which we need to invest at least to the level at which the technologies that are being questioned are tested out and understood and that we are in the lead in understanding which ones will win.

So this is a critical time to make those investments. For 2012, we need to support the community that is trying to do innovation, whether we necessarily agree with the direction or not, we need to make sure that money is going in to test those concepts – test those ideas – and get us to a point at which it is understood what the right direction is for all of us.

So that’s my critical number one for 2012. Let’s make sure that we keep our minds open on innovative ideas and that we test them methodically and that we establish our direction forward.

I think it’s also a time for us to be taking some of the innovations of the last two or three years and get them moving towards more productization. And I encourage the entire industry. Some of these are things that need to be done at the OEM level. Some are things we need to get into our products. Some of these are things we need to get into software products. And I think they are all in question because of the risks that are in the environment and I encourage people not to see it that way but to see that it’s in these down times that we lay the groundwork for the future.

I remember Craig Barrett in the last downturn was well hailed for making significant investments in fabs, and we would be in a much worse position today if he had not done that. It was a visionary thing to do and I encourage people to look at that as an example in these tough times, and make sure we are making investments that will allow us to win in the future.

Comments

  1. exa-preview-1 says
  2. exa-preview-1 says