Member Avatar for happygeek
happygeek 2,411 Most Valuable Poster

File under strange but true: an IBM developed supercomputer based on the same processor technology that powers the Sony PlayStation 3 has just been officially ratified as the fastest computer on the planet. According to the Top 500 supercomputing rankings, published today at the International Supercomputing Conference in Dresden, Germany, the IBM RoadRunner scoops the top spot.

Not only that, but with a benchmarking of 1.026 petaflops it clocks in at more than twice the speed of the previous number one which has held the title since 2004. A petaflop, in case you need reminding, is one thousand trillion calculations per second. Which would make one heck of a fast game of Quake.

Although first seen in the PlayStation 3, those Cell processors were jointly developed by IBM, Sony and Toshiba. Each of them comes complete with a general purpose processor as well as no less than eight co-processors handling high-def graphics and other processor intensive operation functionality.

The IBM RoadRunner can boast no less than 12,240 Cell processors along with its 6,562 dual-core AMD Opteron processors. No wonder, then, that it runs so fast.

No wonder, then, that it cost so much.

How much? Well the US Department of Energy's National Nuclear Security Administration coughed up around $100 million for the beast, to be used within the Los Alamos National Laboratory.

In case you were wondering, IBM remains King of the Supercomputers with an incredible 210 of the 500 fastest machines on the planet. It might surprise you to learn that IBM is only just ahead of HP though, in second place with 183 machines in the list. No surprise at all to discover that most of them, some 75 percent in fact, have an Intel inside...

  • 1 Contributor
  • 0 Replies
  • 189 Views

Dani AI

Generated

As pointed out, Roadrunner’s headline came from applying a console-derived idea at supercomputer scale. Important clarification: the Cell family used in Roadrunner was a server‑tuned variant (PowerXCell 8i) and the machine was built from mixed “tri‑blade” nodes that paired Cell accelerator blades with Opteron control blades. That packaging and the host/accelerator split are what made Roadrunner a true hybrid system and not simply “lots of PS3s in a row.” (powershow.com)

Raw peak numbers only tell half the story — getting real applications to run well required substantial software work. The practical path that succeeded on Roadrunner was: profile to find hotspots; refactor hot kernels into SIMD/vectorized routines for the SPEs (the Cell local‑store processors); move data with explicit DMA and overlap transfers with compute (double‑buffering); and keep MPI/OpenMP on the Opteron hosts for control, I/O and communication. These are the same kinds of steps shown in the Roadrunner performance case studies (for example, Sweep3D) and explained in the Cell architecture literature. (sc08.supercomputing.org)

Perspective: Roadrunner proved large-scale host+accelerator designs could win LINPACK, and its engineering taught the community how critical data motion and toolchains are when working with accelerators. The system was retired in 2013 as the field moved toward more energy‑efficient, easier‑to‑program accelerators, but the lessons about partitioning, memory movement, and close hardware/software collaboration still apply when targeting GPUs or other modern accelerators. (web.ornl.gov)

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.