China's Tianhe-1A supercomputer still has a lot to prove
As international battles of ego go, getting a supercomputer to the top of the prestigious Top 500 List can look like a benchmark way to measure a country’s computing prowess.According to chip maker Nvidia, China has now overtaken the US...
According to chip maker Nvidia, China has now overtaken the US using a hybrid design based on mixing 14,336 Intel Xeon chips with 7,168 Nvidia graphics chips in the 2.507 petaflop Tianhe-1A supercomputer. The full list will be published in the next week or two.
In the last ranking, China made it to number two, so many will view the moment as symbolic. For a number of reasons I’d argue that the symbolism is misleading.
It is worth pointing out that the US will still hugely dominate the top 500 supercomputer list, having accounted for 282 in the last one in June. That’s unlikely to have shifted downwards by much.
That should matter on some level. It’s not which computer is number one, or number three or number nine that matters but how much overall supercomputing power a country can muster in total and in this the US is still way ahead.
But Tianhe-1A raises a larger question of what all this computing power is for. Supercomputers still need software and applications to run, and a suitable pipeline of tasks to run these against.
Nvidia’s press release lists a number of applications for the hybrid architecture that helped propel the Chinese machine to the number one spot, including “drug discovery, hurricane and tsunami modelling, cancer research, car design, even studying the formation of galaxies.”
It’s not clear which of these Tianhe-1A will be used for, but suspicions abound that some of what the world is getting with the new machine is really an engineering exercise driven by a hybrid architecture married to a special interconnect developed to boost throughput.
The real measure of any supercomputer is the uses to which it is put, the knowledge it helps the scientific and engineering community to cement. The Top500 List does not measure such a thing because such a thing is almost impossible to measure objectively.
Supercomputers need to be more than powerful boxes rolled out to impress the media. Their power needs to be accessible to the scientific community, and that probably means distributing the resources in smaller chunks across more institutions. On that score at least China still has something to prove.
ShareTwitter Facebook Google Plus
We explore the tech giants making waves in artificial intelligence
Ground-breaking research into supercapacitor technology could soon transform battery limits for smartphones and laptops too