This page contains a Flash digital edition of a book.
Power to the people

How do we achieve maximum performance from hardware resources? Beth Harlen investigates

‘Te rise of commodity processors, specifically those based on the x86 architecture, served to standardise the high-performance computing (HPC) ecosystem,’ says Bob Masson, marketing director at Convey Computer. ‘Te problem, however, is that we’ve run into the laws of physics.’ Tere are many challenges facing the HPC industry right now – many of which have been touched upon in this publication – but one of the most critical is, as Masson explains, the fact that the clock rate of an individual processor can no longer be increased — it simply gets too hot. ‘Tere is a direct relationship between the clock cycle and heat generated, and the latter has been pushed as far as it is physically able to go. As a result, individual processor clock rates are flattening,’ he adds. Tis is no small dilemma and although the

performance of high-performance computing (HPC) systems is increasing, as the Top500 list aptly demonstrates year-on-year, the demand for computational power is rising at an even greater rate. ‘Tere are an unlimited number of research fields which need all the computation power they can get,’ explains Sumit Gupta, general manager, Tesla Accelerated Computing at Nvidia, ‘like the search for new biofuels or for improving the fuel efficiency of engines, or better aerodynamics for cars, trucks or airplanes.’ Even a marginal increase in computational

capabilities can have wide-reaching implications for the type of research work being carried out – the achievable resolution within climate


modelling, for example. And, as Gupta adds, a one per cent improvement in aerodynamics would lead to millions of dollars saved in fuel costs every year for the transportation industry, reducing the cost of consumer goods. In achieving this goal, one option is to

accelerate applications through the use of a graphics processing unit (GPU). ‘GPUs, which were originally used as graphics engines for gaming, have evolved to have 10 times more performance than CPUs alone and are used as accelerators to CPUs,’ explains Gupta who points out that with the advent of GPU-accelerated computing, supercomputing has also become much more inexpensive and accessible. Beyond that, he states that GPU-accelerated computing architecture has democratised supercomputing, leading to countries such as China, Russia, India, Brazil, Spain, and many others using accelerated computing to build some of the fastest supercomputers in the world. Giovanbattista (Giovanni) Mattiussi, sales

(field-programmable gate array), a technology that according to analysts may become more mainstream in the next five years. ‘Companies like Maxeler in the UK are already implementing very specific solutions on FPGAs and we at Eurotech, while leveraging mainstream accelerators like GPUs and MIC, have invested in the technology because we believe that type of acceleration will become even more important in the future,’ he adds. Convey is also making that investment. Te

“Tere is a direct relationship between the clock cycle and heat generated, and the latter has been pushed as far as it is physically able to go” Bob Masson

company’s Hybrid-Core system is a combination of FPGAs that have application-specific logic and an x86 ecosystem, and so fits smoothly into a computing infrastructure. Masson explains that the only difference is that the kernel of the application resides on the FPGAs, which leads to greater performance as well as efficiency in terms of performance per Watt. He also believes that FPGAs provide a solution to another problem: the fact that applications oſten can’t take advantage of multiple cores without being restructured. Te solution, he says, is to utilise application-specific logic. ‘In other words,’ continues

Masson, ‘instead of executing a series of general purpose instructions, it is much more efficient to design gates down in the hardware that perform the application directly.

and marketing, HPC Division of Eurotech, and member of the ETP4HPC office, agrees that in order to get better performance for systems many companies are studying potential new accelerators. Te new Nvidia Kepler and Intel’s Phi are good examples, he says, as are FPGAs

Te only practical way to do that is to use field-programmable gate arrays – load the application’s logic into hardware “on the fly.” Plus, FPGAs are still taking advantage of Moore’s Law, increasing logic density, and that trend appears to be continuing.’

On-demand ‘Building and operating a high-performance server cluster, or even just a few racks, can

Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32