Encanto Supercomputer May Be ‘Sold in Pieces’
January 2nd, 2013 By: Rich Miller
A supercomputer that was the third-fastest machine in the world in 2008 has been repossessed by the state of New Mexico and will likely be sold for parts to three universities, local media reported last week.
The state has been unable to find a buyer for the Encanto supercomputer, which was built and maintained with $20 million in state funding. Last summer the administration of Gov. Susana Martinez took control of the supercomputer, saying the non-profit New Mexico Computing Applications Center could no longer afford to maintain the machine.
This week state officials told the Albuquerque Journal that no single buyer has emerged. As a result, the state is likely to divide the machine up and sell it by the cabinet to state universities that could repurpose the hardware.
“Barring someone offering to buy the whole machine, we can still get piecemeal use from it,” state Information Technology Secretary Darryl Ackley told the paper. “The universities have proposed to cannibalize it to put some of the assets back into service.”
Encanto is composed of 28 racks of processors. Each rack has 500 processors, for a total of 14,000 cores that together can perform 172 trillion calculations per second. UNM wants to take 10 of the computing racks. New Mexico State University wants four, and the New Mexico Institute of Mining and Technology would take two racks.
Encanto is housed at an Intel data center in Rio Rancho, New Mexico. It features a water-cooled SGI Altix ICE 8200 cluster with a power load of 32 kilowatts per cabinet. It runs over an Infiniband network connected to the National Lambda Rail network.
The supercomputer had the enthusiastic backing of Gov. Bill Richardson, who saw the project as an economic development tool for New Mexico. But the commercial projects did not materialize, and Martinez did not support the project.
“I think that the supercomputer is a symbol of excess,” Martinez said after taking office in early 2011. “We have analyzed the situation. We have seen what the previous administration has invested in it. And we have looked to see what we have gotten back in revenue. It doesn’t match up. It’s unacceptable to the taxpayer, so we’ve got to stop investing good dollars into that project because it’s not proving itself to be useful whatsoever.”
Here’s a technical overview of Encanto from Intel in happier times for the project. This video runs 3 minutes, 30 seconds.
RyanPosted January 3rd, 2013
I’m a little confused. Are they unable to sell time on the cluster to pay for the electricity (i.e. its losing money)? Why can’t they keep it in that facility and provide remote access to it for the universities?
It may have been a mistake to build it in the first place but it seems like dismantling a fully functional machine is also likely a waste of time and money.
Oh, and 172TB. Haha. I know guys with more storage than that in their basements these days. In 2009 the company I worked for was stuffing 10 times that into a single rack…
HPCmonkeyPosted January 4th, 2013
you are correct that the computational capacity of a 2008 supercomputer is not enough to justify the electricity and cooling costs. If you were to operate such a system for about two years, it would likely turn out to be cheaper to buy new equipment to match the capacity of the old – simply because you are saving floor space, cooling, power and engineer time on a massive scale. The new system would be roughly a third of the size of the old one, way more powerful from a single thread point of view, and require a fraction of the power/cooling. It would also require a lot less replacement parts and running maintenance.
As to your storage claims, I think you need to check your units, since 1.7PB on 2009 spindles would not physically fit inside two racks, let alone one. Also, I doubt that any private person without a small hadron collider or some other high data volume instrument would be able to fill more than a few dozen TB over the lifetime of said drives.
CJPosted January 4th, 2013
3 fastest in 2008 in the supercomputer world equates to actually reaching 3 place on the top500 list which it didn’t and its 172.0 TFlop/s is Rpeak not Rmax. Rpeak is Teoretical and Rmax is what they actually produced.
Its 172.0 TFlop/s Ryan as in calculations not storage.even though that gets easier to run at home with GPU:s as well.and some ppl like to have 43+ harddrives spinning in their basements i guess.