Top500 List

There are computers that are far larger than any laptop. They fill entire halls, consist of thousands of interconnected individual machines, and often cost several hundred million euros. Such machines are called supercomputers. The Top500 list is a ranking of these machines: it lists the 500 fastest installations that are publicly known. Whoever holds first place operates, according to this measurement, the fastest known computer in the world. The list has been published twice a year since 1993, in June and in November, and is maintained by a group of researchers from Germany and the United States.

What a place in the ranking says about a country

Supercomputers are expensive state projects. They are needed for weather and climate models, for drug development, for simulations in materials research, and for military purposes. A country at the top thereby demonstrates that it can build, pay for, and operate such installations. That is why the list always also has a political dimension.

This becomes especially clear in the relationship between the USA and China. Both countries have traded places at the top for years. Since the USA restricted the export of certain high-performance chips to China, China has conspicuously registered very few new systems with the list. Experts assume that powerful computers exist there that simply are not being entered. The list therefore does not show the world, only the voluntarily reported part of it.

For the financial world, the list is also an indicator of which manufacturers are currently in demand. If most of the new top-ranked machines are equipped with graphics chips from a particular vendor, that is a signal of that vendor’s market position. Data center operators also watch the list, because it shows which technology has proven itself in practice.

The Linpack test and the flops unit

In order to compare computers at all, all of them must solve the same task. This task is called the Linpack benchmark. The computer must solve a very large system of linear equations, that is, many equations with many unknowns simultaneously. What is measured is how many computing steps it manages to perform per second.

The unit for this is called flops, short for floating-point operations per second. Roughly speaking, one flop is a single calculation with decimal numbers. Today’s top systems operate in the range of exaflops. That is a trillion calculations per second, a one followed by eighteen zeros. A good laptop manages roughly a hundred-billionth of that.

The test has a well-known weakness. It measures only a single type of computational work, one that happens to distribute very well across many processors. Many real-world tasks behave differently, because they constantly have to shuffle data back and forth between components. A computer can shine in the Linpack test and still be noticeably slower in everyday use. That is why there are supplementary lists such as HPCG, which pose more realistic tasks, and the Green500, which does not measure speed but computing performance per watt of power consumption.

The list in the news and in the debate over AI data centers

Twice a year, reports go through the press about who currently has the fastest computer in the world. Germany is regularly represented, among others with the Jupiter system at the Jülich Research Center, the first European computer in the exaflops class. The Leibniz Supercomputing Centre near Munich and Stuttgart have also appeared in the top ranks for years.

What is interesting is what is missing from the list. The large AI data centers of companies such as Google, Microsoft, or Meta mostly do not register their installations. To do so they would have to run the Linpack test and disclose details, and neither brings them any benefit. According to expert estimates, several of these private installations would otherwise be found in the top ranks.

A common misconception is therefore to read the Top500 list as a complete overview of the world’s computing power. It is a ranking of the participants, not of everything that exists. Read correctly, it should be understood as a snapshot of publicly funded research computers and as a trend indicator for the technology being used.