Last week we talked about the upcoming release of Intel’s Xeon E5 processor family. This week, we have some even more important announcements regarding Intel MIC and the strategic direction that Intel is headed regarding high performance computing.
Die shot of 'Aubrey Isle,' the silicon chip included in the Intel MIC 'Knights Ferry' development platform
This is a continuation of our look at future trends in high performance computing. In part 1 we covered the first five of the top ten trends. In this installment we’ll wrap up with the remaining five.
As we near the end of 2011, we take a moment to reflect on the past year. It’s been a busy year for IT across virtually all verticals, from mobile and search to enterprise servers and cloud computing. When we attended HPC360 a few weeks ago, we had the pleasure to attend a keynote presentation by Addison Snell, CEO of Intersect Research in which he discussed the most important trends in high performance computing (HPC).
HPC is an exciting and growing industry that ICC has been moving into the past couple years. The traditional HPC space revolved around high-end research facilities particularly in science and engineering. However, with each year technological innovations and tailored systems such as our Supermicro GPU Simcluster have brought the realm of HPC closer to reality for many small/medium-sized business and organizations.
In this 2-part series we will look at the top 10 future trends in HPC from Intersect360′s research, coupled with our own analysis and thoughts. No better way for us computer nerds to close the year right? Let’s get started.
Last month we announced our new line of GPU supercomputing products powered by NVIDIA. These recent releases were only the build up to the real GPU highlight coming out – ICC NovaServ™ Tesla GPU SimCluster solutions.
Integrating cutting-edge technology from our industry-leading partners, ICC’s NovaServ™ Tesla GPU SimClusters are ready-to-deploy cluster solutions off the shelf, delivering the power of CPU-GPU parallel processing. The 1U 6016GT-T is a powerful stand-alone system. But the real power in high performance computing and GPU supercomputing comes from the ability to scale upward through computer clusters. By transforming the base 6016GT-T into such a configuration you gain an asset with incredible performance potential that draws upon the combined AND clustered power of CPU and GPU.
We’ve just launched two new GPU products, along with our new GPU supercomputing solutions section! GPU computing is the use of a graphics processing unit (GPU) in computing purposes, from general-purpose to supercomputing tasks. With a constantly increasing demand for greater computing performance across the board, supercomputing is more and more drawing upon a hybrid model which integrates the roles of GPUs and CPUs.
If you are looking to gain a performance boost in your cluster or grid computing, ICC’s GPU solutions may be the perfect fit. The scalability requirements involved especially in grid computing make GPU solutions an ideal IT asset.
At ICC, we integrate our top technology with NVIDIA’s® CUDA™ GPU architecture, the simplest way for you to purchase, utilize, and manage a GPU-based cluster. GPU supercomputing has never been easier than with ICC NovaServ™ solutions, providing optimal value by minimizing time spent dealing with the technology and allowing you to focus on what you do best.
Wow, this is the first update in a while on the ICC blog. We have been working on several web-based projects that have been keeping us busy, and I would like to highlight some of them (and other news) in this post.
First of all, as you may have noticed, our HPC by ICC section of the site launched earlier this month which describes ICC solutions for high-performance computing clusters. There is an outline of the Platform Computing HPC Suite, an industry-leading cluster management software package, as well as a diagram which explains common cluster components. Our goal is to open up high-performance computing to industries that have been slow to adopt it, even though HPC may save them a lot of money in the long term and help them stay competitive. If you think you could benefit from an upgrade to your IT infrastructure, feel free to contact us for a free consultation.
At SC10, Supermicro unveiled their GPU SuperBlade server modules (SBI-7126TG) that will offer perhaps the highest density CPU-GPU computing power available on the market. A 42U rack, comprised of six 7U blade enclosures – each with ten GPU SuperBlade modules – can carry 120 CPUs and 120 GPUs. For comparison, a rack with 42 standard dual-processor 1U servers would have 84 CPUs and no GPUs. We will have these high-density server products available on our site soon after they are released. Read more about them in the Supermicro press release or on this podcast interview with Tau Leng, GM of Supermicro, by insideHPC.
After almost a year-long run, the Jaguar supercomputer at Oak Ridge National Laboratory in Tennessee has relinquished its title as the world’s fastest computer. This honor now belongs to the Tianhe-1A supercomputer located in the National Supercomputing Center in Tianjin, China.
Tianhe-1A is expected to officially become the leader of the TOP500.org list of the world’s fastest supercomputers sometime in mid-November. It clocked an impressive 2.507 petaflops on the LINPACK scale, which is about the sum of the performance of supercomputers #6 to #10 on the Top 500 list, according to insideHPC. Jaguar, now the second most powerful supercomputer in the world, had a peak performance of about 1.75 petaflops.
Although Tianhe-1A may re-ignite the anxiety in the West that usually accompanies news of great achievements from East Asia, this is not the first time that America or Europe had lost the #1 place on the Top 500. In 2002, Japan captured the top spot with their Earth Simulator (ES) supercomputer, which remained the world’s fastest until September of 2004 when IBM’s Blue Gene/L cluster at Argonne National Laboratory surpassed it. The quasi-geopolitical competition for computing power is far from over, but China’s ascendancy is actually one of the less interesting things about Tianhe-1A.
Tianhe-1A can potentially usher in a new era in “personal supercomputing”. It is the first leader of the Top 500 to make extensive use of GPUs (Graphics Processing Units). In fact, it is comprised of 7,168 NVIDIA Tesla M2050 GPUs and 14,336 Intel CPUs. In comparison, Jaguar has 37,376 AMD CPUs and no GPUs.
For the past few years, processing technology has been steadily trying to break out of the typical x86 CPU mold that has been the standard since the 1970s. Since GPUs (graphics processing units) have emerged, traditional processing methods have been augmented with the vector computing approach of those units working in conjunction with CPUs.
Now, new types of processors are being developed that will take computing even further. As HPCWire reports, a company called Lyric Semiconductor is launching a line of products that will tackle computation in an entirely different way than the current linear method with Boolean gates that standard x86 processors employ. In other words, Lyric Semiconductor is chucking the 0s and 1s, the fundamentals of computer programming for the past half-century, out the window.
The method Lyric has been developing is based on probability processing. Michael Feldman of HPCWire describes how current computing applications have outgrown the linear model:
The goal is to construct hardware circuitry and software purpose-built for probability applications. With conventional digital technology, processing has to follow a strictly linear path. This is fine for software like operating systems, spreadsheets, word processing, and database transactions, where the computing consists of straightforward calculations or data movement. “But most of the interesting things happening nowadays don’t really fit into that model,” says Reynolds. . .That encompasses a wide range of applications including Web searching, financial modeling, genome sequence analysis, speech recognition, climate modeling, credit fraud detection, spam filtering, and financial modeling, among many others. People tend to associate these probability-based applications with human-like intelligence, and this is clearly where software, in general, is moving.
Lyric is ambitious, and this new form of processing could eventually overtake traditional CPUs in supercomputing applications. As the above article describes, new computing languages are being written that are customized for probability processing. As computers are programmed to think more and more like humans, the hardware that goes into them will continue to evolve beyond the linear modes of x86 computing.
Multicore microprocessors, which emerged after CPU manufacturers hit the power wall and could not sufficiently cool the ever-shrinking transistors on single-core processors, hold a lot of promise for the future of computing. But one major obstacle prevents them from revolutionizing CPU speeds: the challenge of programming for parallel and multicore computing.
In a fascinating article called “The Trouble with Multicore” on IEEE Spectrum, David Patteron describes the history of how processor production shifted from single-core to multicore and the unfulfilled promises of the latter technology. Patterson refers to the advent of multicore processors as a “Hail Mary pass” on the part of the integrated circuit manufacturers when Moore’s Law, which accurately described since the 1980s that CPU speeds would double every 1.5 years, hit the power wall described above.
At that point, Intel and AMD began building processors with multiple cores and Moore’s Law, at least the part about the number of transistors in a single processor, continued to hold true. Multicore processors, in rough terms, are several processors bundled into one. So instead of measuring the progress of CPU technology solely in terms of transistors counts and clock rates, now core counts have become equally important for increasing computing performance.
But their promise is limited by software, not hardware. Programming for parallel computing, as Patterson describes, has been one of the greatest challenges in computer science for decades. No computer language created can effectively handle a diverse amount of parallel applications. Patterson writes, “It’s much easier to parallelize programs that deal with lots of users doing pretty much the same thing rather than a single user doing something very complicated. That’s because you can readily take advantage of the inherent task-level parallelism of the problem at hand.” So, programming for parallel computing (which is a requirement for full utilization of multicore processors) has only been successful in a few heavily-funded applications, such as bank ATM software, online tracking of airline ticketing, computer graphics, GPUs, and scientific computing. No general language or method has been found to apply to them all.
The moral of the story is that software is now one of the major bottlenecks in improving CPU performance. As Patterson forcefully notes after contrasting programming for single-core (which was relatively simple, given that increased transistor counts guaranteed programs ran faster no matter how they were written) and multicore processors, “The La-Z-Boy era of program performance is now officially over, so programmers who care about performance must get up off their recliners and start making their programs parallel.”
When Platform Computing began as a company in 1992, computers were used by large organizations much like they had been for several previous decades. If a complex computing job needed to be run, it would be run on one machine, often only during certain times in the day. There was no widespread use of clusters or computing clouds. Computational research that now takes several days or weeks to complete used to take many months or years.
The first commercial project that Platform Computing undertook, according to an interview on HPCintheCloud.com with Platform CEO Songnian Zhou, was to help the engine manufacturer Pratt & Whitney design the engines for the new Boeing 777 airliner. Zhou describes how supercomputers were used back then:
At that time, they were using one Cray supercomputer rather than IBM mainframes to do it — and every night they would run one job. One job! Per night! Using that one Cray they had to explore all the parameters — how big or small, how many blades, and so on — all the design alternatives; that takes dozens and dozens of runs. They had to run half a year, which is of course a big problem for their product cycle to serve the airlines and their customers.
Platform Computing sought to change the way organizations such as Pratt & Whitney used supercomputers. Instead of running one job on one computer at a time, Platform pioneered the use of software to break up complex jobs to be performed on many computers connected together in one computer cluster. The airline and automotive industries, according to Zhou, were the early adopters of this technology and used it to speed up and simplify their design simulations.
Today, cloud computing and clustering have become industry standards, and there are now many companies that offer software and other services to facilitate them. Platform Computing still specializes in private clouds (a network of computers in the same facility most likely owned by the same organization) and community clouds (a network of computers owned and shared exclusively by a few organizations) which means that they cater towards large corporations and organizations that can afford purchasing large clusters.
Public clouds, the type most people think of when they talk about cloud computing, is facilitated mostly by other companies, although Platform has made an effort to reach out to smaller businesses and organizations in 2010.
If you are interested in Platform Computing software for clustering or private clouds, feel free to contact ICC and we can talk to you about the different options they have available to take full advantage of your computing hardware resources.