Amazon takes supercomputing to the cloud
If you have $1,279, you can buy an hour of supercomputing time on Amazon's cloud.
You may not need to use the 42nd fastest supercomputer on Earth, but if you want to, you can for just $1,279 per hour.
As reported by Wired, Amazon Web Services latest salvo into the computing on demand landscape is a platform known as the Elastic Cloud Computer, which at $1279 per hour, or $11 million a year if run full time, is probably on par in comparison to the time, effort and expense of procuring the same level of compute power in your own data center.
Amazon's virtual super computer is capable of running 240 trillion calculations per second, or 240 teraflops on 17,000 cores. While undoubtedly impressive, this pales in comparison to Fujitsu's K Computer, which hit 10 petaflops in November 2011, equating to 10 quadrillion calculations a second.
Admittedly there is a big difference, but I'm pretty sure that K Computer wasn't also running much of the tech world's infrastructure, nor was K Computer also selling books and every other retail item on the planet at the same time.
So what does it take to run at 10 petaflops? According to some public specs, 864 racks, holding a grand total of 88,128 interconnected CPUs, all-in estimated at $20 million. Each processor is linked with 16GB of RAM, bringing the memory in the machine to 1,377 terabytes. All that compute consumes 9.89 megawatts of power -- the same as 10,000 suburban homes. It costs Fujitsu $10 million per year just for the power bill.
Oddly enough there is more news this week in the land of supercomputers as ExtremeTech reported that Lomonosov Moscow State University--the oldest university in Russia--will house a 10-petaflop supercomputer created by T-Platforms, an up-and-coming Russian high-performance computing (HPC) company.
The real question is if you have the need for such computing power, do you need it run full-time 24x7 or is there enough variability in your workload needs that the Amazon offering can solve your problem. I suspect that most supercomputing users will stick with their own data centers but as more applications are developed that can take advantage of large numbers of CPUs for intermittent periods of time, the Amazon offering will start to look very attractive.