The list of the world's 500 fastest supercomputers came out yesterday with a top 10 that was unchanged from the previous ranking issued in June. But further down the list, a familiar name is making a charge: Amazon, with its Elastic Compute Cloud service, built a 17,024-core, 240-teraflop cluster that now ranks as the 42nd fastest supercomputer in the world.
Amazon previously built a 7,040-core, 41.8-teraflop cloud cluster that hit number 233 on the list, then fell to 451st. But Amazon submitted an updated Linpack benchmark test with the addition of a new type of high-performance computing instance known as "Cluster Compute Eight Extra Large," which each have two Intel Xeon processors, 16 cores, 60GB of RAM and 3.37TB of storage. The full cluster on the Top 500 list is Linux-based, with 17,024 cores, 66,000GB of memory, and a 10 Gigabit Ethernet interconnect.
For less than $1,000 per hour, with the purchase of 290 HPC instances, Amazon said customers can create their own 63.7-teraflop cluster, which would be fast enough to make the Top 500 list. We reported before on the building of a 30,000-core Amazon EC2 cluster at the rate of $1,279 per hour. That one was built by a third-party vendor called Cycle Computing for one of its customers in the pharmaceutical industry, and contained 3,809 compute instances, each with eight cores and 7GB of RAM, for a total of 30,472 cores, 26.7TB of RAM, and 2PB (petabytes) of disk space.
Despite Amazon EC2's advances in high-performance computing, it's still a ways off from topping the world's fastest computers. In the most recent list, Japan's K Computer hit 10.51 petaflops, or 10 quadrillion calculations per second, to maintain its position as the fastest in the world, and more than 40 times faster than the Amazon EC2 cluster. Amazon's achievement is less about raw power than it is making supercomputing capacity available to the masses, an hour at a time.
reader comments
30