This page contains a Flash digital edition of a book.
than a normal CPU, because of their higher computational efficiency, they propel such systems up the list,’ adds Cameron. He notes that, while the Green500 is


basically a re-ordering of the Top500 list by power, that method limits what can be recognised. To address this limitation, he and his partners established the Little Green500, an exploratory list which broadens the definition of a supercomputer to being as fast as the 500th-ranked computer on the Top500 list 18 months prior. It thus allows for smaller machines, and yet another list includes novel machines. For example, he points to the first cluster run on solar power at the Tokyo Institute of Technology as being ‘in the spirit of things we want to encourage’. For the Little Green500 List, the average efficiency is 199 Mflops/W, but this increase over the Green500 comes at the expense of lower overall performance.


How cool are supercomputers?


With HPC centres consuming megawatts, operators are looking for every opportunity to trim their energy


budget. As Paul Schreier discovers, manufacturers are making progress in reducing power consumption


W


e’re all aware of energy efficiency for desktop computers, but HPC is a whole new ball game. In PCs, EnergyStar power-management


features emphasise placing monitors and computers into a low-power ‘sleep mode’ after a period of inactivity, however, expensive HPC systems are running at high capacity as much as possible so other approaches are needed.


The Green500 illustrates progress Advances in energy-efficient HPC move at a rapid pace; to get a snapshot of where we’re at, it’s always useful to check the latest version of the Green500 list of the most energy-efficient supercomputers. The leader by a wide margin is a prototype IBM Blue Gene/Q system at the TJ Watson Research Center (1684.20 Mflops/W), followed by a system at the GSIC Center of the Tokyo


22 SCIENTIFIC COMPUTING WORLD


Institute of Technology based on the HP ProLiant SL390s (958.35 Mflops/W) and then a hybrid cluster core at the National Center for Supercomputing Applications (NCSA) at the University of Illinois , Urbana-Champaign (933.86 Mflops/W). According to Kirk Cameron, co-founder


of the Green500 and director of the Scalable Performance Laboratory at Virginia Tech, digging a bit deeper reveals an interesting trend; in fact, one might argue that this year’s list was ‘The Year of the Accelerator’, whether commodity GPUs from Nvidia and AMD or the custom PowerXCell 8i processor from IBM. Accelerator-based supercomputers take eight of the top 10 slots on the list. This class of computer on the Green500 List has an average efficiency of 756 Mflops/W, compared to the others on the list with an average of 211 Mflops/W. ‘Although the local power consumption of a GPU is higher


Tops in both performance and energy The Blue Gene/Q leads the Green500 by an enormous factor, almost 2:1. Exactly how did IBM achieve that? According to Alan Gara, lead architect of the Blue Gene systems, a very efficient water-cooling system can explain 10 to 20 per cent of the system’s energy efficiency, but the biggest part comes from the use of a new processor optimised for energy efficiency rather than thread performance. ‘Earlier,’ says Gara, ‘the philosophy in supercomputers was to take the most powerful processors and couple them together. Our approach is different in that we work with energy-efficient processors that are scalable.’ This new processor, dubbed the A2, allows


for more energy efficient processing for the highly parallel codes commonly encountered in HPC. Also, the 16-core chip contains a memory controller and network adapter and achieves 200 Gflops, consuming roughly 80W including memory. The A2 is making its premier in a


supercomputer in this prototype BlueGene/Q system; that supercomputer has roughly 1000 chips per rack for 200 Tflops, and a full system has 96 racks. Gara notes that IBM is building this prototype to investigate preproduction chips. The machine is the basis for the Sequoia system which will be delivered to Lawrence Livermore National Lab, likely next year – and Gara claims that this computer might be both the world’s fastest and the most energy efficient at the same time, something he adds that was similarly true with the original BlueGene/L system. He predicts with confidence that


production computers based on the A2 will have even better energy numbers. ‘Usually,


www.scientific-computing.com


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  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44