Features:

INTEL SHOWCASES BREADTH, RANGE AT SC04

What's on Intel's radar this year at SC04? Richard Dracott, director of Intel's Enterprise Platforms Marketing group, shares his perspective with HPCwire contributing writer Jan Rowell.

Rowell: The Supercomputing conference is always a great way to gauge the state of high-performance computing. What are people going to see this year?

Dracott: They're going to see that HPC is vibrant and growing, and that Moore's Law – along with advances in scalability, operating systems and other HPC technologies – has brought us to a very exciting place that's transforming computational science. Look at something like NASA's Project Columbia system, where NASA and SGI have integrated 20 Altix computers, each with 512 Intel Itanium 2 processors, and created the world's largest single-system, shared- memory supercomputer. By uniting these 10,240 very powerful processors, NASA is getting roughly a 10-fold increase in its computing capability, which is giving NASA and its industrial collaborators a quantum leap in their research capabilities. They're able to approach problems in whole new ways, and get answers in days that would have taken six months or a year previously, or simply been impossible to do.

What's especially significant about Project Columbia is that it's not a stunt machine. It didn't take years to deploy. It isn't a one-of-a-kind, look-at-me, academic platform. It's an easy-to-replicate, workhorse system that came together very quickly, and within a couple of months was supporting groundbreaking research like NASA's digital astronaut simulation, where life sciences teams model human organs in very high fidelity. It provides outstanding price/performance that helps NASA do more science per dollar, and it's expected to deliver tremendous value as NASA works on its return to flight program.

Rowell: Supercomputing is also a place to see the scientific progress that's being made on high performance computers. What are some of the highlights for Intel this year?

Dracott: There are so many. One very telling example is the Vital-IT Center in Lausanne, Switzerland, created by the Swiss Institute of Bioinformatics, an innovative alliance formed with some leading life sciences institutions along with Intel and HP. Switzerland has a thriving biotech industry and Swiss developers have created some of the world's most widely used biotechnology programs and databases. The Swiss Institute of Bioinformatics got a 64-CPU Itanium 2-based system this year, and the first week it was operational, researchers made scientific discoveries that may one day lead to more rational vaccine design. Another great example is the work Boeing is doing on its Computational Fluid Dynamics (CFD) code, the results of which is the optimized Boeing 777 Nacelle, the outer engine cover, that can be seen hanging above the Intel booth.

Rowell: With respect to Intel, what will people see at the show this year?

Dracott: I think they'll see the tremendous breadth of Intel's involvement in supercomputing. Intel is working with a wide range of companies, academic institutions and end-user organizations, as well as in our own research labs and product development teams, to help drive advances on a broad range of HPC technologies. Our goal is to deliver higher performance at lower price points and to bring the "Intel goodness" – the performance, scalability, stability, reliability and breadth of choices of the Intel platform and the Intel developer community ? further into the HPC space. In our booth, people will see a 10 teraflops-sized system built around the latest Intel Xeon processor with Intel EM64T technology, a 10 GB/s InfiniBand interconnect and InfiniBand switch fabric. We're also showing off the newest, fastest Intel Itanium 2 processor, which we're announcing at the show, along with software tools from Intel that can help developers maximize performance of HPC applications. And of course they'll also see the great innovation in the Intel community, from both users and vendors, including SGI's interactive visualization demos and the University of Washington's interactive high-definition video demos.

Rowell: You introduced 64-bit technology to the Intel Xeon processor family this year. Does the world really need two 64-bit processors from Intel? What's the impact for HPC users?

Dracott: You know, HPC is all about 64-bit computing, so Intel's introduction of Intel EM64T technology is very positive for HPC and other market segments. We don't believe a one-size-fits-all approach addresses the unique computing needs of HPC users (and other users), so we're offering different flavors of 64-bit computing, each of which is well suited to different application workloads. For example, the Intel Xeon processor, with Intel EM64T, DDR2 memory support and Hyper-Threading technology, excels at applications like seismic modeling, computational fluid dynamics and global climate research. The Intel Itanium 2 processor, with its large cache and IO bandwidth, is outstanding for tasks such as reservoir modeling, biosimulation and finite element analysis. This means that whether you want the Intel Itanium processor family's EPIC instruction set, or the flexibility of a dual-core solution, you can get it from Intel, along with the outstanding performance, price- performance and scalability that Intel is known for. You can benefit from all the work that Intel and others in the industry are doing to create superb compilers and tools and to optimize application performance and drive progress on a variety of HPC building blocks. And you can count on Intel's product road maps to deliver higher performance and advanced capabilities in the future.

Rowell: HPC users are early adapters. What are you doing to help them take advantage of the Intel EM64T technology?

Dracott: Intel is providing tools, technical support and expertise for organizations that are incorporating Intel EM64T into their operating systems, tools, device drivers or applications. There's a comprehensive software developer guide and other information at http://developer.intel.com/technology/64bitextensions/. Intel and the development community are also providing ongoing training to support this migration.

Rowell: What's new from Intel in the tools space?

Dracott: Clusters continue to be a focus of activity in HPC, and we're supporting this trend with our new Intel Cluster Tools, which are designed to help developers get maximum performance from cluster applications. We're also announcing the release of Intel MPI, a fabric-independent MPI library based on MPICH2. In addition to providing top library-level performance, the Intel MPI library enables an application to run on a variety of popular interconnects, including Ethernet, shared memory fabrics, and InfiniBand architecture, all with a single executable.

We'll also be demonstrating new features in the Intel Trace Analyzer and Intel Trace Collector – tools that work together to provide insights into the performance profile of MPI applications and allow developers to quickly focus on the desired level of detail and tune for better performance. Since last year's conference, we've also introduced the Intel Cluster Math Kernel Library, a highly optimized math library that uses MPI and includes ScaLAPACK, BLAS and LAPACK Libraries, Discrete Fourier Transforms and Vector Math Library.

Rowell: Intel has created an Advanced Computing Center. What does that mean for the HPC community?

Dracott: The Advanced Computing Center is part of Intel's research focus on high-performance computing systems, aimed at making optimal use of multi-core and multithreaded CPUs to help get the performance leaps that will be needed for future workloads. The ACC funds R&D projects across the spectrum of advanced computing – from CPU and platform architectures to system software, programming environments and application analysis. In addition, ACC funds research projects with key universities worldwide to study emerging platform technologies and help accelerate innovation in mainstream computing. The ACC is an exciting initiative, not only because it's pushing the envelope for the next generation of high-performance computing, but also because it will drive HPC capabilities into smaller form factors and lower costs. We expect to eventually see HPC systems that are affordable enough for the desktop and provide the ease of use of today's desktop applications.

Rowell: Power is a challenge for the entire semiconductor industry. What's Intel working on with respect to the power envelope?

Dracott: Our R&D teams and product development engineers are exploring a variety of novel, power-saving techniques – everything from new transistor structures, to advanced packaging materials, to software optimization techniques – all with a goal of increasing performance and reducing power consumption, current leakage and heat. For example, future Itanium processor family products may offer even more capabilities at an even more favorable power picture. For example, the current Itanium 2 processor with 6MB of cache, has a Thermal Design Power of 130 watts. When we launch the Montecito product (currently planned for next year), it is expected to run at a higher clock speed, provide dual cores with multithreading, have over 26MB cache and deliver a significant performance increase over previous generations. And the cool thing – literally – is that Montecito is expected to have a Thermal Design Power of 100 watts, 30 watts less than the Itanium 2 processor with 6MB of cache. We're also planning new power-related technologies – our Foxton technology will monitor power and temperature and will help optimize performance to get an even greater productivity boost. At the data center level, demand-based switching will look at processor utilization and help smooth out the peaks and valleys, and is expected to get another overall reduction in power used by the servers. That's a lot of added value for HPC users.

Rowell: HP pulled its Itanium 2 processor-based workstation line a few months ago. Was that a blow to Intel?

Dracott: Not at all. As a matter of fact, this is a good example of the comment I made earlier about this not being a one size fits all world. There was a bit of negative speculation, but the press missed the important point, which is that our family of Intel Xeon processors provide optimum performance and price/performance for the workstation market segment. They've grown significantly in popularity over the last few years and are rapidly displacing RISC architectures. Intel and HP are in strong agreement in targeting the Intel Xeon processor to the workstation market segment and the Itanium 2 processor to the mid- and high-end server market segment, and we continue to displace RISC platforms on both fronts and to penetrate into the heart of the RISC data-tier applications environment.

Rowell: There's been a surge in grid computing over the past year. What's driving that momentum, and what does it mean for Intel?

Dracott: We're seeing three worldwide market segments being established. The first was the national, government and academic HPC segment, which is already well-recognized. Just look at the National Science Foundation's TeraGrid in the US, the European Union DataGrid, the China Grid, as well as many others. These are the true pioneers, thought-leaders and early adopters, and they're driven by HPC's traditional appetite for more MIPS and its need for global collaboration.

The second segment, which is beginning to gain momentum, consists of commercial enterprises in fields like financial services, pharmaceutical and manufacturing, where there's an early adopter HPC community that recognizes that grids can help give their companies a competitive edge. These companies are using grids to drive cost-efficiencies around things like introducing new and more sophisticated financial instruments, or accelerate the process of drug discovery, or bypassing physical prototypes altogether and moving straight from design to production.

The third segment is also primarily in the commercial space, but it includes sharing and managing all enterprise compute and information assets ? handhelds, laptops, appliances, applications, databases, servers and more. This market segment is in the very early stages of development, and achieving its vision of a fully distributed infrastructure will depend heavily on well defined standards, flexible building blocks, end-to-end solutions, and a broad range of services.

What do grids mean for Intel? We view grid computing as a very positive and important evolutionary step along the path from proprietary infrastructures to commercial-off-the-shelf building blocks and solutions that ultimately give the end user maximum value, choice and flexibility. The real opportunity for Intel is to offer end-to-end architectural solutions that include mobile devices, laptops, workstations, networks, servers and large-scale SMP products. Grid is going in a direction that is very consistent with Intel's vision of the service-oriented enterprise based on provisioning of data and resources across the firewall in a fully distributed global infrastructure. As you would expect, Intel is committed to driving standards toward a convergence that enables volume economics and drives global proliferation.

HPCwire contributing writer Jan Rowell (janrowell@comcast.net) is a freelance writer and 20-year observer of the high-performance computing scene.