Speeches

Michael Dell keynoted SuperCompute 2008 in Austin, Texas on November 18, 2008. He talked about the massive waves of innovation that have shaped high-performance computing over the past 40 years – from specialized processors and proprietary operating systems in the 1970s, to microprocessor-based systems of the 1980s and 90s, to the emergence of high-performance computing clusters built using standardized hardware are software. He also discussed how this wave of standardization, coupled with simplification at every layer of the computing stack, is further democratizing supercomputing and unleashing the capabilities of the world’s greatest minds – even those who don’t have colossal infrastructure budgets.

Additional photos

Michael Dell

SC08 - Unleashing Human Capability:  The 4th Wave of HPC

Austin, Texas

November 18, 2008

 

MICHAEL DELL:  Well, thank you very much.  It's great to be here for this 20th SC conference.

 

You know, most people know that Austin is the live music capital of the world, but did you also know that it's the supercomputer capital of the world?  You might figure that out while you're here.

 

Austin, of course, is home to the University of Texas, and most of the leading microprocessor companies have significant operations in Austin, in fact design some of their most advanced parts here in Austin.  Of course, Intel, AMD, NVIDIA, IBM Cell Processor, and, of course, you've got companies like Dell building systems to support the supercomputing revolution.

 

So, there are many great minds here today from academia, from government, from business, and most of you are Dell customers, so I want to thank you for that.  We appreciate your business very much.

 

I also want to recognize a very special group of high performers:  math and science students from Austin's LBJ High School are here today.  Let's recognize those folks, if you all could stand up.  (applause)   So, these are our next generation of supercomputing leaders.

 

Now, high performance computing touches every life and every human every single day, whether you're at the gas pump or you're getting on an airplane.  It's critical that we continue to innovate in our society, and the pace of that innovation is really accelerating like never before.

 

Now, in some cases, you know, it's okay to slow down a bit, and this is actually a pretty good example.  The Large Hadron Collider at CERN, where Dell servers are hard at work, you know, they've been having a few challenges there, and the scientists have some issues with the collider, and so it's down until this coming spring.

 

Now, I thought that seemed like a long time until I realized what was at stake when they were talking about reducing the planet to goo.  So, you guys, just take your time and make sure you get that right before you turn that thing on.  (laughter)  That would be a real mess.

 

So, seriously, here in this room we have enormous human capability, human capacity.  In computer terms modeling a human brain would require about 20 petaflops of power, computing power. 

 

Now, of course, Japan is busy building a 10 petaflop supercomputer at a cost of about $1.8 billion for about half a human brain.  So, even if we spent $3.6 billion and built a brain, it certainly wouldn't have all the other capabilities of a human brain. 

 

It's also interesting to note that the human brain only uses about 20 watts of energy.  So, it kind of shows you that our industry -- (laughter) still has a long way to go in terms of innovation.

 

But that's not really the point.  The point is that the role of technology is to complement and multiply human capacity and human capability, and really unleash its infinite capacity.  So, our job is to make your job as scientists and engineers easier.

 

So, let's talk about this.  You know, when I think about the great work that you all do, you know, I'm really sort of humbled by the awesome challenges that exist out there in the world, the scientific challenges, and here we've listed some of the great challenges as chronicled by the National Academy of Engineering.  These are really all computational problems at the end of the day.  That's what makes being in our industry so exciting.

 

And so when you look at the various disciplines around computational fluid dynamics, visualization and animation, computational biology, proteomics and genomics, multiscale modeling and simulation, and quantum chromodynamics, these are the kind of disciplines that will help solve these challenges:  like advancing fusion power for new sources of clean, affordable, accessible energy, like nano-filtration to bring clean water.  Only one in six people in the world today have clean water, so that's a big opportunity and a big need.  And nanotech advances in multiplex assays to bring us better medicines.

 

There's one that's actually not on this list that we might kind of look at this list and say, you know, we kind of forgot one.  Maybe at the beginning of the year, it didn't seem like it needed to be on the list but now it does, and that's the economic crisis that we're facing today. 

 

You might say, well, hey, what's science going to have to do with that.  Well, I think there's a lot that the kind of technology that we create can yield to solve some of the economic challenges that we have, but first we've got to realize that budgets are probably going to be affected by the economic crisis, and for me it really highlights the competitiveness and productivity agenda that our country and really the world faces, and really raises this to a level of importance that has never been greater.

 

So, there's good news actually.  The ability to address these challenges is accelerating, and it continues like never before.  In IT there's about a doubling of processing power every year, and costs continue to come down about 50 percent.  Communications technology is doubling in price performance and bandwidth and capacity every year.  This drives costs down and drives innovation up, and it unleashes even greater human capability.

 

Now, the video that we showed earlier talked about some of the evolution of high performance computing, and it's really occurred in waves.  The 1970s was the age of specialized vector processors and proprietary operating systems like the ILLIAC IV, the Cray 1, and the CDC 6600.

 

Now, the 1980s and '90s brought us microprocessor based massively parallel systems.  Fifteen years ago, the number one system on the top 500 was a Thinking Machines machine, 1024 processors with a theoretical peak performance of 131 gigaflops, and a sustained impact performance of 59.7 gigaflops.

 

Now, today, I brought with me our Dell Precision M6400.  This machine has a peak performance of 40 gigaflops; so not quite up to the Thinking Machines capability, but getting pretty close, and it uses a little bit less power and you can kind of take it with you. 

 

It's got a Quad-Core Extreme Edition processor, it's got 1 gigabyte NVIDIA FX graphics, and it can store a half a terabyte of storage, so pretty nice little machine to take with you.

 

Now, advances in graphics technology are actually creating some new opportunities in supercomputing, and this is really coming from advanced graphics workstations and from gaming machines.  So, these GPUs are pretty attractive alternatives.

 

And we're excited to announce today that we're extending our partnership with NVIDIA to advance their CUDA architecture in Dell's Precision workstations.  So, this really is the supercomputer on your desk.

 

And adding one Tesla Card to a Dell Precision workstation delivers a theoretical performance of 1 teraflop.  So, that's seven times higher than that Thinking Machines product back from 1993.

 

Now, this decade has really been marked with the emergence of clusters, and this really is building supercomputers by interconnecting smaller and standards-based servers, the kind that Dell is pretty famous for.

 

The standards-based computing has quickly penetrated the top 500 list, driven by the enormous improvements in hardware, and the wide availability of software components to take advantage of them.

 

And Dell was really at the forefront of standards-based computing starting in 1999 when we, along with the Cornell Center for Advanced Computing, built a 64-node cluster.  Back then, each node was a Quad Pentium 3 running at 500 megahertz, and running Windows NT.

 

This is actually a great opportunity for me to recognize Dr. David Lifka from Cornell, who is here today.  Dr. Lifka, I want to thank you for your partnership.  Thank you.  (applause)

 

And Dr. Lifka and his team were really at the forefront of this whole cluster trend almost a decade ago, and now today 429 of the top 500 supercomputers are based on the X86 microprocessor architecture.

So, now we're entering what we refer to as the fourth wave of high performance computing, and this is marked by continued standardization of high density servers and microprocessors, more scalable and faster access, including to shared pools of storage, datacenter optimization to lower the total cost of ownership, and easier management and administration.

 

One of the things you see here is some of the things from the commercial world in managing large datacenters really penetrating very heavily into the new requirements that we see from customers.

 

This fourth wave is further democratizing the availability of supercomputing, making it more accessible to more organizations in more countries around the world.

 

And there's no better example of this fourth wave than right here in Austin, in the University of Texas, the Texas Advanced Computing Center. 

 

We're proud to announce that TACC's Visualization Wall is running on Dell XPS gaming workstations, and Dell's 30-inch displays, and it will soon be the largest display wall of its kind in the world with 307 million pixels.  Let's take a look.

 

(video segment)

 

(applause) 

 

MICHAEL DELL:  I've got to get one of those.  I don't think it will fit in my office, though.

 

I want to thank and congratulate Dr. Jay Boisseau from TACC and the rest of the Texas Advanced Computing Center team that's here on their great work.  (applause)

 

So, now let's talk about the underlying technology that's behind this fourth wave of high performance computing innovation.  There are many elements in the stack, in the technology stack that kind of make this up, and we'll kind of go through each part of the advancements and kind of take a look at what's happening.

 

So, let's start with the servers themselves, and, of course, one of the first things you see is tremendous increases in density.  Three years ago, using our blade chassis, we put 240 cores in a full sized 42u rack, with 2.8 gigahertz CPUs, and that was 1.34 teraflops of theoretical peak performance per rack.

 

Now today, using our blade chassis we get 512 cores in a 42u rack, with 3.3 gigahertz CPUs, and that's 6.82 teraflops per rack.

 

Now, using our DCS servers, we can actually take that up to around 9 teraflops per rack.

 

So, we continue to increase the performance in available space and lower the power.

Now, blades are a fantastic way to manage a complex large installation with great scalability and also great efficiency.  Our blades, like you see the M1000 series here, use almost 20 percent less power than our leading competitor.  They have integrated switching from partners like Cisco and Brocade, and a 40 gig backplane, the fastest in the industry.

 

We also have blades that are optimized with vary large memory spaces.  So, if you want to put your entire model in memory or your entire database or even we see customers doing virtualization with large memory address spaces, we've got memory address spaces that are 50 percent larger than competitors.

 

Now, this density is really driven by what's going on in microprocessors, and the X86 kind of revolution continues, and you're seeing more and more cores, increased performance, but also without more power required.  AMD has just released a Shanghai processor, which is 50 percent faster than the Barcelona processor, and we started shipping Shanghai in Dell servers yesterday, and we'll also be out in front with Intel's Nehalem processor next year when that arrives.  So, you can stay tuned for that.

 

Now, the past four generations of Dell servers have improved performance per watt by about 500 percent, but they now use 25 percent less energy than they did four generations ago.

 

So, if we go back to 2003, a 2,500 core cluster was 1,250 servers, and it sustained 9.8 teraflops.  Today, a 1,240 core cluster is only 155 servers, and it sustains 10.7 teraflops.  So, it's more teraflops with almost 90 percent fewer servers in just five years.

 

And memory capabilities are improving as well, as I mentioned earlier.  The next generation four-socket servers that we deliver, based on the Nehalem processor, will support memory spaces of up to 1 terabyte of system memory.  This really opens up new possibilities for scientists and engineers in what can be done with this capacity.

 

You're also getting far more for your investment.  In 2003, a million dollars was about 2 teraflops of performance.  In 2005, it was 4 to 5 teraflops.  And what's really happening is pretty amazing; today, about a million dollars is roughly 25 teraflops using quad-core CPUs.

 

So, moving a little bit up the stack into networking and storage, these areas are also accelerating at a pretty dramatic rate, and great examples of this are InfiniBand and 10 gigabit Ethernet.  We'll be the first with quad data rate InfiniBand on our blades.  That's coming very soon. 

 

And in storage, access to I/O in a scalable, reliable way is a pretty major challenge for any high performance computing environment.  And we have a range of storage options, including our alliance with EMC, and also our leadership in the iSCSI space on Ethernet.

 

Back over here on the left-hand side of the chart you see our EqualLogic PS5500 series, which scales to 576 terabytes, and very soon we'll bring 10 gigabit Ethernet to the EqualLogic product line.

So, at Dell we've always had a deep commitment to open standards, and we've also been committed to growing our high performance computing ecosystem along with that commitment to standards.

 

Now, our industry and certainly our company have been developing peta-scale hardware, but more work is needed to ensure that we have peta-scale software to take advantage of all this enormous power that I just described, and all those cores.

 

Now, of course, you've got the normal operating systems, Linux and Windows, and a number of these we, of course, factory install for our customers.  Open Source software packages like Open Cluster Stack from Platform and Rocks+ from Clustercorp, we collaborate very closely with the independent software vendors and also are instrumental in programs like Intel's Cluster Ready Program to make sure with our certification program that we have the broadest set of applications available, and that you're assured that those applications are sized and qualified before they arrive in your center.

 

So, we have a commitment, a strong commitment to the infrastructure and tools to support the ecosystem, so you can focus on the science.

 

We're also committed to driving a revolution in services, and I think this is an often an overlooked area of innovation and opportunity in the high performance computing space.  Tabor reports that 70 percent of the high performance computing budgets go to staffing and facilities, and being able to simplify services mitigates that labor and that cost burden.

 

So, how do we do this?  Well, it's factory integration, so building the entire cluster inside our factories and delivering it in prebuilt racks, turnkey cluster deployments where we take over the entire deployment for customers, and automated cluster design, like our online tools to plan and size clusters, which you can find at Dell.com/HPCC.  And if you go there and you have any feedback, you know, let us know.  We're always looking to improve these tools, and we build them for folks like you.

 

So, I'll also say that you are a pretty demanding set of customers, but you've also taught us a lot.  We've learned a lot about what our environments in the future need to look like by hanging out with the customers that have the toughest requirements, and so we like that.

 

Now, high performance computing is not just about science; it's actually companies of all types. 

 

One of our customers, a company you might have heard of called Facebook, now, I don't think they're going to be solving any great scientific challenges anytime soon, but they do have 120 million users running their service on about 10,000 Dell servers, and they've got some pretty big challenges. 

 

British Petroleum is using cluster power to really accelerate seismic exploration and to look for new energy.

 

China's largest auto group, FAW, is replacing physical component testing with simulation.  We see simulation really expanding dramatically across the world.

 

The Translational Genomics Research Institute is examining molecular profiles to unravel genetic components of disease.

 

And recently, Microsoft announced their new Azure service.  This is a new platform which allows you to scale easily from 10 users to 10 million users without additional coding, and it all runs on Dell.

 

So, the work that all of these organizations do is really very different, and spread out all over the globe, but one thing that they're all concerned with is how do you deal with density, processing power, storage and cost to make these systems really work well.  So, all of these organizations are really doing pretty amazing things.

 

Now, speaking of amazing, I also want to share some great news for the entire high performance computing community.  Dell and Lawrence Livermore National Labs, and nine other vendors, have teamed together to develop Hyperion.  This is a 96 teraflop test bed, which is 100 percent dedicated to tackling your biggest challenges.

 

And hyper-scale computing environments really present some pretty unique challenges.  I mentioned earlier this issue of the need for peta-scale software, applications that can take advantage of this enormous power.  Storage, connectivity, management software:  These are all challenges that we're going to be dealing with as we implement systems of this scale.

 

So, Hyperion is a test bed big enough to really test scale, and it will share those breakthroughs with the entire Open Source community. 

 

So, no single institution really could have done this alone, and I want to thank the Hyperion team that's here today really for their contributions to the entire community. 

 

Now, we've had a chance today to talk about how technology is complementing our human potential to really solve these great scientific challenges, and I believe standardization is going to further democratize IT, and it's going to be critical to ensure that those who invest in it stay competitive.  It's also going to drive higher efficiency and productivity, which is great for our planet.

 

Yesterday, we reported that our OptiPlex desktops -- these are pretty commonly found in business and institutional environments, the most popular business desktop in the world -- through improvements that we've made to those desktops that would be fairly subtle to users, we have saved our customers $3 billion in energy costs since 2005, and avoided 28 million tons of CO2 just in the last three years.

 

So, the point is we can all make a difference.  A hundred years from now when we look back, having solved those grand challenges that were up on the board earlier, our world will be better because of the great work that you all do, and also the role that it inspires in others.

 

So, that's the idea behind something at Dell we call the ReGeneration, and I'd like you to take a look at this little video, and then I'm happy to answer your questions.

 

(video segment)

 

MICHAEL DELL:  All right, so we have time for some questions.  If anyone has a question, let's see, do we have microphones? 

 

MODERATOR:  Anyone who has a question, please stand up, and a person will come to you with a microphone.

 

But before we do, I'd like to present Mr. Dell with a token of our appreciation.  We're presenting him with one of our SC08 committee jean jackets, customized, of course, with the Dell logo on the front.  (laughter)  Again, thank you very much.

 

MICHAEL DELL:  Well, thank you.  Thank you.  (applause)

 

MODERATOR:  Okay, so if you have questions, again please stand up and the people with the microphone will bring the microphone to you.

 

PARTICIPANT:  So, thank you very much for a very nice presentation, and showcasing all the way from the desktop to the high performance to the Hyperion systems.  You alluded to nanotechnologies and also emulation of the brain.  In your mind, how we are going to enable that?  And also given the fact that it requires a lot of research and research funding, where do you see industry in a sense collaborating with government agents to enable such capabilities?  Thank you.

 

MICHAEL DELL:  You know, the point of the slide of the brain was not to say that we should go and build a human brain simulator; it was really to compare the power in a brain to what's going on in supercomputing. 

 

I think really for me the dream and the excitement about computers was always not that the computers were going to replace the brain, but that human creativity and capacity to dream is really tremendous.  And then if you look at computing power, well, that's pretty incredible, too.

 

But if we sit back and think about how these connect, it's a relatively rudimentary process today.  You type a key in and something happens on the screen, and you type another key in, something happens, and maybe that's not the best way to do it, right?  I think there are enormous opportunities to improve that kind of man-machine interface, and really leverage human creativity even further.

As far as industry and government, I think it's going to require partnerships.  I think we have a critical need to increase the funding and focus on basic science and research.  I was pretty disappointed as a citizen to see that Congress passed legislation and funding for science and research, but then not actually provide it; that didn't make a lot of sense.  And I think we have to recognize that we in this country are not the only smart folks in the world.  Ninety-six percent of the people in the world don't live in America, and there are a lot of other folks out there graduating engineers and scientists and working on their own scientific endeavors.

 

So, I think it's going to take a combination of industry and government together and research institutions to ensure that the U.S. stays at the preeminent level in terms of scientific research.

 

Other questions?

 

PARTICIPANT:  Yes.  In practical terms can you describe what the addition of HPC technologies, green IT, and the inclusion of things like the AMD Shanghai chip into your servers and then PCs will do for the average worker and IT departments with their servers?

 

MICHAEL DELL:  Well, if you look at what's going on in computing, there is a pretty massive wave of virtualization.  And virtualization started in servers, but now it's kind of moving into storage and into network and IO virtualization, and into client virtualization. 

 

If you take these large clusters and you say, hey, you know, we learn a lot about how to connect all these together, and really in a scale-out fashion leverage the power of these machines that are connected together, that's kind of how the Internet is being built, that's kind of how the -- we call them private clouds inside large corporations are built.

 

So, just like in just about every sector of technology, university research and research institutions have kind of paved the path for some wave of technology innovation that then comes back to corporations and businesses.

 

So, if you look at businesses today, they're reaping enormous benefits in cost savings by virtualizing their server environments, and also dynamically moving pools of applications around a shared set of servers.

 

So, imagine if you have 10,000 servers but they're all in a shared pool.  Well, in a corporate environment you might have hundreds of applications.  You're not running all those applications at the same time in the same capacity and requirements; you can dynamically move those around at different times of the day, week, month to benefit from that.  So, that's a real tangible example that's absolutely happening right now.  Client virtualization is just starting and, of course, all the networking topology improvements around InfiniBand and 10 gigabit Ethernet, those are foundational for what's happening kind of throughout the whole IT stack.

 

PARTICIPANT:  Mr. Dell, my concern is education, and I'd like to know how we bring the digital promise beyond Facebook and Second Life into the classrooms where students touch the future with the understanding of what's going on here.  We're still back in the Internet age doing not very much of anything.  What is your vision for helping education join and understand what's going on?

 

MICHAEL DELL:  You know, when you look at education in this country or many other countries around the world, what you find is a pretty uneven distribution of computing and other skills and tools.

 

My view is that the public education system in this country or really any country has an obligation to convey 21st century skills upon the people who pass through its doors.

 

What does that really mean?  Well, it means that if you graduate from a public school, you ought to know how to use the tools that are common in today's society, because you basically can't get a job if you don't know how to do that.  And that's true at Dell and just about any other company you'd go to.

 

When I say it's uneven, I mean I think you find some school districts that are really advanced at that, have done a great job, and absolutely do that, and you find others that really haven't done it very well at all.  And I think in our society today that's really a bad situation. 

 

There are other challenges in education.  For sure there are many of them.  One of the things that our family foundation has been pretty focused on is how do you take the best ideas from schools that have really performed well and transfer those around the nation or around the world, and also how do you use measurement and metrics and performance and outcome based metrics to improve performance in schools.  Actually, a number of schools have embraced this, and are seeing some pretty breakthrough results. 

 

So, for the second year, our foundation held a summit on performance management in education, and I think the attendance roughly tripled, so there were several hundred educators from around the country participating in this.

 

I think that's the sort of thing that will help, but I sort of go back to each school district has got to ensure that its students, when they graduate, have those 21st century skills.  It's just as important as reading, writing, and arithmetic.

 

PARTICIPANT:  I would like to know, I have read about ReGeneration and how the Dell company would like to become more green.  One of the biggest flaws in our education system is that many people don't even know what an alternative fuel is, and how we can use that to help our economy.

 

So, what I would like to know is how can Dell reach out to those students and teach them what fuel, just conserving and reaching out to alternative fuels such as the electric car and like other sources such as like hydrogen fuel cells and things like that.

 

MICHAEL DELL:  Well, ReGeneration.org is really a resource to do just that, to share ideas and communicate and educate.

 

We also have another interesting site called Dell.com/Earth, and it has a number of resources that talk about the things that you're asking about.

 

I'll give a basic example.  If you take an average computer that would be sitting around most offices that's, let's say, three or four years old, it uses about $120 of energy, electric energy per year.  We just introduced a whole new series of notebook computers, the E series, and my favorite one is the E4200.  It weighs less than 1 kilogram, and it uses $6 to $7 of energy per year.  So, from $120 to $6 to $7, that's a pretty massive improvement, and those are the kind of things where if you ignite sort of competition among industry to say, hey, who's going to produce the machines that are the most energy efficient, you can yield enormous savings.  Now, that doesn't create new sources of alternate energy, but it certainly lessens the need for energy.

 

You know, we also as a company have been the first technology company on the planet really to become carbon neutral.  We did this several months ahead of our original schedule.  And so, for example, with the help of Austin Energy, all of our energy in our Austin headquarters comes from renewable sources.

 

So, we're taking steps, we're encouraging others to take steps, and I think you'll see companies kind of compete on that basis, and that's also a great way to improve.

 

PARTICIPANT:  I appreciate your comments about the need for developing applications to take advantage of the computing technology we've been developing.  I've sort of seen this myself over the years, that currently we are spending a lot of money on computer hardware, a lot of money on middleware and various infrastructure, often to support computer programs which were developed for computers many generations back, and that --

 

MICHAEL DELL:  Yeah, we need some new software.

 

PARTICIPANT:  -- one of the sad things -- yeah -- one of the sad things is that a lot of the middleware that's being developed now, and a lot of even the thinking is to support benchmarks from applications which were not designed -- I mean, were designed from an older era.

 

My question is that who is going to step up to fund -- who do you see stepping up to fund these future applications, which will be designed from the ground up, for the new hardware and hopefully will begin to induce a response in the middleware community to design things for a whole new world of applications?  Thank you.

 

MICHAEL DELL:  Well, there are application vendors, but, of course, when you talk about apps, there isn't one app, right, there's thousands or tens of thousands of apps, and they get very specialized, and so I think you have to get into really specific industries and companies and institutions that are going to work together to go create those apps.

 

But there's no question that this whole core war came about very quickly, and now we have a huge opportunity to go take advantage of all this power.  And the number of cores that you see in these processors is not going to slow down.  So, if I were out there trying to say, hey, how do I take advantage of this, the folks who figure that out, it's going to be a pretty lucrative opportunity.

 

PARTICIPANT:  I'd like to know what could Dell or other manufacturers do to demand from their vendors to make more energy efficient, low power components that they put in their computers, because Dell makes millions of computers, and they demand electricity, and everything you can do to make that low power, more energy efficient is going to benefit the world in general.

 

MICHAEL DELL:  Absolutely.  That's how we went from $120 to $6 or $7.  We went back to our vendors and we said, okay, guess what, we're going to go build the greenest products in the industry, tell us how you're going to reduce energy consumption.  Even further than that, our supplier partners report to us on their carbon emissions, and we evaluate that as part of how we kind of do a report card for suppliers.

 

So, we're doing that, and you're seeing suppliers kind of step up with all sorts of things, you know, spin down drives, all sorts of lower power improvements to drive energy consumption down.

 

MODERATOR:  I don't think we have any more time for questions, but I'd like to once again thank Michael Dell for joining us at SC08.  (applause)

 

END