Thursday, November 17, 2011

GPUs, Multicore Processors Drive Up Performance of Top Supercomputers

The latest version of the Top500 list of the world’s most powerful supercomputers came out this week, and processing power continues to grow at a Moore’s Law pace.Japan’s K Computer maintained its No. 1 position on the latest list of the world’s Top500 supercomputers, which was published this week. K’s top spot was assured, thanks to an upgrade that makes it four times as powerful as its nearest competitor. Installed at the RIKEN Advanced Institute for Computational Science (AICS) in Kobe, Japan, the system has a peak benchmarked performance of 10.51 petaflops. This is the first system to ever reach the 10-petaflop range; 10 petaflops is 10 quadrillion calculations per second.
The K Computer uses 705,024 SPARC64 processing cores, does not use graphics processors (as many of the systems on the Top500 list now do), and is one of the most energy-efficient systems on the list.

Japan’s K Computer retained first place on the new Top500 list.
Following the K Computer on the list are the Chinese Tianhe-1A system, with 2.57 petaflops performance, and the U.S. Cray XT5 system called Jaguar, installed at the Oak Ridge National Laboratory, with a benchmarked performance of 1.75 petaflops. Incidentally, the Oak Ridge system is undergoing an upgrade that, once completed in 2013, could deliver 20 petaflops of processing power.
The Top500 list has been published every six months since 1993. This current list is the 38th edition. Peak performance is measured by running a standard Linpack benchmark application.
Although a single benchmark can never truly reflect how a system will run a particular application, the use of a common benchmark over the last 18 years allows for some analysis of high-performance computing trends when viewed in context of the entire list.
One notable trend is that the top 10 supercomputers remain unchanged since June 2011. This is the first time the top performers maintained their ranking from one list to the next in its entire history.
And although the top rankings did not change, the newest list does highlight a number of other developments. For example:
1. China has increased its number of systems on the list to 75. It is now the clear No. 2 country in terms of high-performance computing (HPC), ahead of Japan, the U.K., France and Germany.
2. The Chinese systems, at Nos. 2 and 4, and the Japanese Tsubame 2.0 system at No. 5, are all using Nvidia GPUs to accelerate computation.
3. Thirty-nine systems use GPUs as accelerators (up from 17 six months ago); 35 of these use Nvidia chips, two use Cell processors, and two use ATI Radeon chips.
4. Sixty-two percent of the systems use processors with six or more cores.
5. Intel continues to provide the processors for the largest share (76.8 percent) of Top500 systems.
With every list, the entry-level of performance just to claim the 500th spot increases. In the latest list, the barrier for entry moved up to 50.9 teraflops, compared with 39.1 teraflops six months ago. The last system on the newest list would have been ranked 305th in the previous list.
Other points of interest include the following:
1. A total of 384 systems (76.8 percent) are now using Intel processors. This is down slightly from six months ago (386 systems, 77.2 percent).
2. Intel is now followed by the AMD Opteron family, with 63 systems (12.6 percent), down from 66.
3. The share of IBM Power processors has stabilized for now with 49 systems (9.8 percent).
4. Gigabit Ethernet is still the most-used internal system interconnect technology (223 systems, down from 230 systems), due to its widespread use at industrial customers, followed by InfiniBand technology with 213 systems, up from 208 systems.
5. IBM kept its lead in systems and now has 223 systems (44.6 percent), compared to Hewlett-Packard with 140 systems (28.0 percent).

No comments: