Roadrunner: 130,536 cores break the Petaflop barrier

Armonk (NY) – IBM is prepared to deliver the Roadrunner supercomputer to the Department of Energy’s National Nuclear Security Administration (NNSA). The system was development over the past 18 months and is not only the first hybrid supercomputer using Cell processors, but also the first commercial system to exceed a performance of 1 PFlops.

Roadrunner is the NNSA’s third IBM-built supercomputer, adding to two IBM-built supercomputers at the Lawrence Livermore National Laboratory: BlueGene/L, completed in 2005, has 131,072 p5 processors and delivers a sustained performance of 280.6 TFlops. The 12,208 processor ASC Purple is estimated to provide a sustained performance of 75.8 teraflops.

The new Roadrunner, which is currently disassembled and will be shipped to the NNSA within 30 days, according to IBM, was originally described to deliver about 2.8x (3.5x peak: 1.6 PFlops ) the combined processing power of BlueGene/L and ASC Purple. Back in 2006, IBM said that the system would bet more than 16,000 Opteron processor cores and more than 16,000 Cell BE CPUs to deliver on the Petaflop goal.

The final system ended up with less processors – 13,896 Opteron cores (6948 dual-core processors) as well as 12,960 Cell BE chips with a total of 116,640 PPE and SPE cores for a combined 130,536 cores. The original design called for about 176,000 cores.   

According to IBM, Roadrunner has the floating point performance of about 100,000 notebooks. To describe the horsepower of the system, IBM said that Roadrunner can do in one day what would take the entire population of the world 46 years on handheld calculators, assuming a pace of one second per calculation.

Other interesting facts of the supercomputer, which IBM said cost about $100 million, include a power consumption of about 3.9 Mwatts, as well as a structure that includes 80 TB of memory, 576 miles of fiber optic cable as well as 3456 tri-blades which can deliver 400 GFlops each.

Also noteworthy is that Roadrunner was supposed to become the flagship product of AMD’s Torrenza platform, a project which was pitched as an open AMD x86 platform that takes advantage of Direct Connect Architecture and the Hypertransport interface. However, we haven’t heard much about Torrenza anymore and we don’t expect AMD to make a big deal of it.