SoftLayerHosting company SoftLayer to offer a new high-performance cloud computing service that combines Intel's ES-2600 Sandy Bridge servers with one or two of Nvidia's Tesla M2090
The top500.org list is compiled by a group from several research institutions around the world and they attempt to keep track of which of these fantastically expensive machines has the most processing power according to a series of computational benchmarks.
More than five years ago, the fastest supercomputers were just a collection of individual CPUs that could be found in your average high-end Sun or IBM server. But then computer scientists began marrying GPUs to these collections and the result was a new breed of hybrid supercomputer. In November 2009, one of the first GPU-style supercomputers made number five on the top500 list, a Chinese design that married Intel Xeon CPU chips with the AMD graphics chips. Interestingly, the GPUs are underclocked to better match them with the CPU clock rates, which isn't what you would initially expect.
Since then, several additional high-performance computing machines have included combinations of both traditional CPUs and graphic processors. For example, on the most current top500 list that was published last November, there are two Chinese supercomputers at numbers two and four and a Japanese supercomputer at number five, all of which combine Nvidia GPUs with Intel Xeon CPUs. A total of 39 out of the 500 systems are using similar technology to combine both CPU and GPU, more than double the number of systems from the list complied earlier last year. To get an idea of how big these clusters are, most have thousands or even hundred of thousands of core processors and are as big as a room. We have returned to the computers of our forebears, to be sure!
But you don't have to look towards multi-million dollar installations to find this CPU/GPU combination. Indeed, the lowly Sony PlayStation 3 has long had a nine-core configuration that combines both kinds of computing, and for a few years it was one of the favorite go-to machines used to assemble very inexpensive computing clusters.Back in 2008, a group of Swiss researchers put together a cluster of 200 actual Sony Playstations to form a primitive supercomputer.
They said at the time: “We have found that one PlayStation 3 game console is equivalent to about 40 modern single core processors. The most computationally intensive part of our method required about three days of work with over 200 game consoles, which is equivalent to 32 years of computing on a typical desktop computer.”
Not bad for a game console that cost a few hundred dollars.
Sadly, Sony updated their OS and disabled the PS3 to run with non-Sony OSs, which quickly ended the PS3 era of compute clusters. But there are other efforts to continue to encourage this hybrid processor design.
One of these that grew out of using PS3s was the OpenCL parallel computing framework that is designed for GPU computing. Now offered by Fixstars, this first began with an effort called Yellow Dog Linux that was acquired by them a few years ago. The OS was originally designed to run on Sony Playstations and since then has expanded to handle a variety of non-traditional processors.
That is a lot of history, but I provide that to set the context of this week's announcement from SoftLayer, a hosting company.
David Strom is one of the leading experts on network and Internet technologies. He has written and spoken extensively on topics such as VOIP, convergence, email, network management, Internet applications, wireless and Web services for more than 25 years.
See here for all of David's Tom's IT Pro articles.
(Shutterstock cover image credit: CPU City)