thumb made from computer parts

How to build a supercomputer (without breaking the bank)

Posted on

Historically supercomputers have been incredibly expensive. As a result, only the largest organisations have been able to buy one.

But as processors have evolved, and the cost of critical components like CPUs and RAM has fallen, so too has the headline price for a supercomputer. It is now possible for almost any business to buy a supercomputer – if they need one.

How is this possible?

In the past, supercomputers were custom-built by firms like Cray, costing millions of dollars each. But advances in technologies like grid computing and software-defined storage have radically transformed system design.

Take the Owens Cluster (named after Olympic athlete Jesse Owens) that was deployed in the Ohio Supercomputer Center in 2016. The system has a peak theoretical performance of 750 teraflops, allowing it to perform millions of calculations per minute, thanks to its 23,392 CPU cores.

But as the name “cluster” implies, the computer is not just a massive magic box. It is actually constructed from several hundred high specification off-the-shelf Dell PowerEdge servers deployed in a cluster as detailed on the Owens Cluster homepage.

Building your own supercomputer

Taking a lead from the Owens Cluster project, you can relatively easy begin building your own supercomputer. Given that processing power is provided by the clustered capabilities of the component servers, you don’t even need to use the latest technologies – but you will need more lower-specification servers to provide similar performance.

Theoretically, you can build a supercomputer from second user Dell Blade servers to further reduce the headline cost.

There are two potential problems however. First the cluster is configured and controlled by a custom platform to allocate resources and collate results. This software is vital to ensuring the underlying servers are used as a single device.

Second, running hundreds of servers, even in the blade form factor, is incredibly expensive. The cost of powering and cooling the data centre housing your supercomputer will cost thousands of pounds, every month – well beyond the IT budgets of most businesses.

A useful insight

So, although you could (theoretically) build your own supercomputer, the reality is that most companies never need that level of processing power. The cluster technologies underlying modern supercomputers is important however – they show the value and potential of using off-the-shelf servers. And they also show how second user servers may have a role to play in your future IT strategy.

If a 13th Gen Server is good enough for the Owens Super Computer, it’s more than suitable for fulfilling your computing needs.

To learn more about second user servers and what they offer your business, please give us a call.