Intel and Cray Build $ 500 Million "Exascale" Supercomputer for Argonne National Lab – TechCrunch


In a way, I have the equivalent of a supercomputer in my pocket. But in another, more important way, this handheld is a joke compared to real supercomputers – and Intel and Cray are putting together one of the largest ever with a half a billion dollar contract from the Ministry of Energy. He will make exaflops!

The "Aurora" program aims to set up an "exascale" computer system for the Argonne National Laboratory by 2021. The prefix "exa" is the prefix indicating the name, in this case 1 trillion floating point or FLOP operations. They are sort of the power rating of supercomputers.

As a comparison, your average modern processor maybe a hundred or more gigaflops. A thousand gigaflops made a teraflop, a thousand teraflops made a petaflop and a thousand petaflops an exaflop. Thus, despite the major advances made in terms of IT efficiency in the manufacture of smartphones and super-powerful workstations, we are talking about several orders of magnitude. (Do not talk about GPUs, it's complicated.)

And even compared to the biggest supercomputers and clusters on the market, you're still thinking up to 200 petaflops (that would be the IBM Summit at the Oak Ridge National Lab) or so.

Why do you need this type of computing power? Petaflops would not do it? Well no. A very recent example of computer limitations in real-world research is this study of how climate change could affect cloud formation in some regions, reinforcing the trend and leading to a vicious circle.

This sort of thing could only be estimated with much ruder models before; IT resources were too limited to allow for the extremely high number of variables involved here (or here – more clouds). Imagine simulating a ball that bounces off the ground – easy – now imagine the simulation of each molecule of this bullet, their relationships to each other, gravity, atmospheric pressure, as well as other – harsh – forces. Now imagine the simulation of the collision of two stars.

The greater the number of available computing resources, the more the number of dedicated resources, as proposed by Intel's press release, "develops cosmological simulations on an extreme scale, uncovers new approaches for predicting the reaction." drugs and discovers materials for the creation of more efficient organic solar cells. . "

Intel says that Aurora will be the first exaflop system in the United States – a major caveat, as China aims to accomplish this task a year earlier. There is no reason to think that they will not reach it either, as Chinese supercomputers have been among the fastest in the world.

If you are curious about why ANL uses its computers, which will soon be built, do not hesitate to browse its search index. The short answer is "pretty much everything".

Source link