Continuing my features work for Dennis Publishing’s PC Pro magazine, the April 2013 issue sees the publication of The World’s Fastest Computers. A research-heavy look at supercomputers and the high-performance computing (HPC) industry in general, it’s a piece of which I’m particularly proud.
Before getting into the nitty-gritty of the piece, however, I’d like that thank a few people without whom the feature could never have happened: Professor Simon Cox, of the University of Southampton, was a particularly excellent source, speaking to me candidly and at length regarding the realities of running a supercomputing facility and his hopes for the future, and even posing for some photographs to liven up the piece; Nvidia’s Ian Buck, GPU computing general manager and creator of the Compute Unified Device Architecture (CUDA) language, brought years of highly-parallel thinking to the mix, as did Nvidia’s Tesla boss Sumit Gupta; Intel’s Stephan Gillich, director of high-performance computing for the EMEA region, provided a CPU- rather than GPU-led perspective; and finally the Science and Technology Facilities Council was kind enough to provide copyright clearance on several of its historical supercomputing images – including a great shot of a denuded Cray being dismantled at the end of its service, which sadly had to be cut from the piece for space reasons.
The piece is split into three clear sections: a brief history of supercomputing, from the days of the Control Data Corporation 6600 – Seymour Cray’s first HPC design, and the very first system to be described as a supercomputer – to the modern day, followed by a look at what HPC means for education and the industry. The final part, meanwhile, is a look at the future – which, you’ll be amazed to hear, looks very different depending on whether you’re talking to Intel or Nvidia.
Perhaps the most interesting aspect of the piece, for me, is the performance comparison: using data provided by Professor Simon Cox and a great deal of research, I was able to piece together a rough approximation of a performance timeline. Starting with the Ferranti Pegasus in 1956 and working through thirteen other machines – all of which have, at one time or another, been installed at the University of Southampton – I compiled operations-per-second statistics for each. This, more than anything else, demonstrates the runaway nature of high-performance computing: using a linear graph, all but the last two machines – both versions of the University’s current Iridis supercomputer – drew a flat line.
While there’s plenty of information that didn’t make it into the final piece – I compiled nearly 30,000 words of interview material in all – it’s by far one of the most comprehensive I’ve written, and one of which I think I can be justifiably proud.