Last week, Southern Methodist University in Dallas, Texas unveiled the “ManeFrame” – a 120 teraFLOPS supercomputer valued at $6.5 million. Previously named Mana and stationed at the Maui High Performance Computing Center (MHPCC), this system made the top 500 list of worldwide supercomputer sites as recently as the end of 2012. Needless to say it’s a very exclusive list that’s really difficult to earn a spot on.
SMU is no stranger to supercomputing – its existing $14 million datacenter has already helped cancer drug testing and physics, including research leading to the Higgs boson discovery. This in turn helped earn François Englert and Peter W. Higgs the 2013 Nobel Prize in Physics.
Equally amazing was the price SMU paid for the ManeFrame – just $50,000. The U.S. Navy graciously asked only for shipping costs in exchange for this incredible machine. This is strong recognition for SMU’s contributions to science and will no doubt help the university continue to advance the greater good. As proud members of the High Performance Computing community here in Dallas, we at Nimbix congratulate SMU for this exciting new addition to the family.
What if this type of computational power was available to anyone on demand, and how much would it cost? Recently we wrote about HPC on demand enabling innovation for the masses. The more people who can get their hands on this type of capacity, the more breakthroughs we are likely to see and benefit from as a result. In fact, even a fraction of this power is generally enough to solve many of the toughest problems we can think of. Most supercomputers, just like computing clouds, are multi-tenant. It’s unlikely that a single user can consume the entire system at any given time.
At $6.5 million, ManeFrame’s equipment alone costs over $54,000 per teraFLOPS! An example electromagnetic simulation job consuming 7.8 teraFLOPS and running for 12 hours would require a $600 “chunk”. And that’s not the half of it (literally). It’s no accident that SMU’s datacenter is valued at $14 million. Supercomputers are hot blooded beasts with an insatiable thirst for cooling, electricity, and computing skills (read: humans) to keep running. If you were to assemble one of these yourself, you would need to either build or lease a datacenter large (and cool) enough to fit it in. Then there’s the electric bill. It’s neither cheap nor a “solo project”.
Back to our HPC on demand idea… why not pay only for what you need, when you need it? Using our JARVICE supercomputer, we can deliver 120 teraFLOPS for less than $200 per hour. The same electromagnetic simulation costing close to $600 on the ManeFrame (in slice of CAPEX cost alone) can be yours for only $156 total on the Nimbix cloud. And unlike the ManeFrame, it shuts off automatically when it’s done, so you don’t keep paying even when it’s idle. The $156 includes space, cooling, power, and humans to help make sure it runs smoothly as well. Don’t forget that you can’t just “slice off” $600 of ManeFrame – you still have to invest $6.5 million, plus the operating costs. With JARVICE, the $156 is the total amount you pay for the example job with no other strings attached.
How can we do this? First, JARVICE is a bit more modern than ManeFrame, and utilizes the latest GPUs. This means much higher core density and considerably lower overall costs to operate. JARVICE is “green”, saving not only energy, but also money, which we pass directly onto you. Second, JARVICE is open to the public and constantly in use, running many different kinds of jobs at once. Privately accessible supercomputers face the challenge of staying utilized enough to make their behemoth costs worthwhile. JARVICE, on the other hand, never rests. This drives its marginal cost down low enough for us to make it available to you at such an attractive price.
ManeFrame’s 120 teraFLOPS join a proud program at SMU that ultimately helped two scientists win the Nobel Prize. What can you do with 120 teraFLOPS of HPC on demand? We can’t wait to find out!