Most supercomputers today already use processors developed by Intel – a look at the top ten supercomputing sites in the world show that 40% of them use Intel CPUs (30% are IBM and 30% are AMD). The WSJ reports that of the largest machines, Intel can be found in a whopping 82% – of course, to them, that just means that there’s at least 18% of the market still ripe for the plucking.
The majority of supercomputers use traditional x86 processors like those found in an average consumer desktop, just a large, large number of them. Thousands, in fact. Recent attempts have joined CPUs, which are better at versatility, with GPUs, which excel at highly-parallelizable tasks. One such product, the Chinese Nebulae, pairs Intel CPUs and NVIDIA GPUs and jumped instantly to #2 on the world supercomputer rankings. Experts say its theoretical performance could even put it to number one.
So how does Intel plan on changing the game? Think hybrid chips, which combine the versatile CPU and the fast, but simplistic GPU. Intel’s codenamed Knights Corner microarchitecture will push up to fifty cores onto a single chip with its first iterations. The chipmaker plans to build the supercomputer-inclined processors on its next-generation manufacturing process. Today’s chips use 32nm – tomorrow’s, just 22nm, which is about twice as thick as the membranes which make up the cellular structure of people.
Knights Corner technology, called many integrated core, or MIC, takes advantage of Larrabee, the recently-canceled discrete graphics effort from Intel. Its origins can also be found in an undisclosed Intel research project focused on the creation of single-chip cloud computers.
While Intel does seem to be on top, they can’t simply rest on their laurels. As the Top500 list shows, a substantial portion of the world’s supercomputers are powered by smaller rival AMD – including the #1 unit on the list.