Fujitsu recently unveiled a supercomputer built for Japan that is so powerful, it outranks its nearest five competitors on the supercomputer list…put together. Intel seems a little put off by all of the news, and threw out some damage control by claiming to prepare for exascale computing in less than seven years.
Before we get there, let’s take a refresher in metric (S.I.) prefixes!
|yotta||Y||1 000 000 000 000 000 000 000 000|
|zetta||Z||1 000 000 000 000 000 000 000|
|eta||E||1 000 000 000 000 000 000|
|peta||P||1 000 000 000 000 000|
|tera||T||1 000 000 000 000|
|giga||G||1 000 000 000|
|mega||M||1 000 000|
So, as you might expect, a supercomputer that starts ranking in exaFLOPS (FLOPS = floating point operations per second; it’s a number used to gauge the raw computational ability of a machine) is one heck of a capable box. The Fujitsu supercomputer recently announced, the superfast Fujitsu K, is currently capable of roughly 8.2 petaFLOPS. That number is expected to rise as the company adds close to 20% more cores before official deployment next year.
Intel jumped to its own defense, trotting out the fact that 77% of supercomputers on the TOP500 ranking list use Intel processors at their core (ha!); that number jumps to 90% when you consider new computers for 2011. The company mentions its goal of making highly power efficient CPUs; as supercomputers require more and more CPUs to grow in ability, so too could grow their power usage.
Current research shows that in just two years, the top 100 supercomputers on the planet will use 1 million CPU cores, doubling to 2 in 2015 and doubling twice more to 8 million cores by 2020. Intel’s models predict the top computer to hit 100 petaFLOPS in 2015, and rising to 1 exaFLOP in 2018.
If Intel has their way, those computers will all be running Xeon processors. It’s easy to forget that a lot of very smart companies produce CPUs that most consumers or even business ever see or hear of, and they’re not sitting back while Intel eats the market whole. The 2010s will be an interesting decade for supercomputing.