When you look at the Top500 supercomputers in the world, the list that is updated biannually, you can see that there really hasn’t been much movement over the course of this year. For an easier comparison, I recommend you download the excel file that compares all of the 500 different supercomputers in the world.
If you look at the list, you will notice that the Chinese are ahead of the US by quite a margin (Green bar) and it will likely be some time until the US or Europeans catch up. Even though the Chinese are using an all Intel solution, it is interesting to see that the AMD + Nvidia solution is the fastest supercomputer in the US. Out of the top 10 of the top 500, there are only 2 academic supercomputers. The rest are for ‘research’ which can be anything you imagine it to be as long as it requires a question to be answered. Many of these questions can be defense related, but the truth is that most supercomputers original purpose is rarely what they get used for once they’ve accomplished their initial task.
You can also see that Nvidia definitely has the most efficient supercomputer in the top 10 with the Swiss supercomputer using Tesla K20X compute cards. Interestingly enough, even though Nvidia has announced their K40 compute card, there weren’t any supercomputer build outs using it. Which means we will probably have to wait until 2014 to see how much of an impact, if any, K40 will have, especially with Maxwell coming next year. The K-Computer from Fujitsu in Japan is by far the least efficient and that is because they are using very high power consuming Sparc chips while the rest are using x86 Intel, AMD or IBM Power PC.
The U.S. is clearly the leading consumer of HPC systems with 265 of the 500 systems (253 last time). The European share (102 systems compared to 112 last time) is still lower than the Asian share (115 systems, down from 118 last time). Dominant countries in Asia are China with 63 systems (down from 65) and Japan with 28 systems (down from 30). In Europe, UK, France, and Germany, are almost equal with 23, 22, and 20 respectively.
The Supercomputing List is likely to change next year with new GPUs and CPUs from all of the major vendors, however, I wouldn’t expect much movement in the February list, but rather the November list after all of the new hardware has been deployed.