Scaling up supercomputing power
The fastest machines on the planet can solve a whopping quadrillion (1015) calculations per second, earning them the right to be called petascale systems. With such computational power, they are already speeding through complex problems and detecting patterns from tons of data that no human could ever manually scour through in their lifetime.
But across Asia, researchers and industry leaders are now making strides towards the elusive exascale, capable of a quintillion (1018) operations each second or one exaFLOPS. Winning the 2021 Gordon Bell Prize for HPC, the Chinese team behind the Next Generation Sunway supercomputer recently showcased their machine could reach speeds of up to 4.4 exaFLOPS, rivaling the power of Google’s Sycamore quantum computer.
Scientists at Tsinghua University in China have also developed new memory management technology to bump up Sunway’s capabilities in artificial intelligence (AI) algorithms, which learn from input data and improve over time. Their innovation transformed variably sized data sets and filed them in Sunway’s memory slots, improving access to data without disrupting the machine’s main memory.
Not only can exascale computers solve typical AI problems faster, but they would importantly enable more realistic simulations of processes, from the natural environment to synthetic industries. These would open the floodgates for HPC’s potential to advance scientific discovery and develop impact-driven applications, such as by generating more exact maps of the Earth’s water bodies, climate patterns and any disruptions to them.
With higher computing speed comes greater resolution—referring to seeing systems and patterns in finer detail—and therefore more accurate simulations and predictions. For example, simulations could zoom in to storms in specific cities, looking into microclimates that depend on local situations like the presence of valleys, amount of forest coverage and other structures.
