IBM packs 128TB of flash into brain-simulating supercomputer

To accommodate all the data needed to model the 70 million neurons that make up a mouse brain, Big Blue is using scads of the same type of memory used for PC solid-state drives.

Stephen Shankland Former Principal Writer
Stephen Shankland worked at CNET from 1998 to 2024 and wrote about processors, digital photography, AI, quantum computing, computer science, materials science, supercomputers, drones, browsers, 3D printing, USB, and new computing technology in general. He has a soft spot in his heart for standards groups and I/O interfaces. His first big scoop was about radioactive cat poop.
Expertise Processors | Semiconductors | Web browsers | Quantum computing | Supercomputers | AI | 3D printing | Drones | Computer science | Physics | Programming | Materials science | USB | UWB | Android | Digital photography | Science Credentials
  • Shankland covered the tech industry for more than 25 years and was a science writer for five years before that. He has deep expertise in microprocessors, digital photography, computer hardware and software, internet standards, web technology, and more.
Stephen Shankland
2 min read
Felix Schuermann of EPFL is scheduled to describe a flash-infused IBM supercomputer the June supercomputing conference in Germany.
Felix Schuermann of EPFL is scheduled to describe a flash-infused IBM supercomputer at a supercomputing conference in Germany next week. IBM

Drawing from engineering ideas that are revamping personal computers, IBM and two Swiss universities are using flash memory to improve performance of a supercomputer designed to simulate an actual mouse brain.

But there's a lot more flash memory than you'll find in the latest laptop.

A PC's solid-state drive may come with something like 128GB to 512GB of flash memory. The mouse-brain project's specially upgraded version of a Blue Gene/Q supercomputer has 250 to 1,000 times as much flash memory -- 128 terabytes.

The supercomputer project, in conjunction with the Ecole Polytechnique Federale de Lausanne (EPFL) and the Eidgenoessische Technische Hochschule Zurich (ETH Zurich), aims to model the 70 million or so neurons that make up a mouse brain. The work dovetails with the Blue Brain project and the European Union's Human Brain Project.

IBM announced the development in conjunction with the International Supercomputing Conference 2013 that takes place next week in Leipzig, Germany. Felix Schürmann, general manager of the Blue Brain Project, is scheduled to present results at the show.

Flash memory is cheaper, more capacious, and slower than DRAM, the type of memory used to store data that a computer processor needs pronto. Flash memory also is faster but more expensive than hard-drive storage. For that reason, it often occupies an intermediate tier between hard drives and DRAM (dynamic random access memory) in the memory hierarchy.

"The major reasons are both cost and speed," said IBM Research computational scientist Alessandro Curioni of the decision to use flash memory. "Detailed brain simulations in particular have unique requirements that make it desirable to use a different ratio of compute [power] to memory [capacity] than today's systems."

The flash memory will store a variety of data, including data from the mouse brain that's under study, simulation results that are awaiting further analysis, and checkpoint data to keep track of what's going on in the simulation, Curioni said.

Flash memory in PCs today is usually connected with the SATA interface used to hook up hard drives, but the Apple's new Mac Pro puts flash memory a step closer to the processor by linking it with the PCI Express interface. IBM wouldn't say how it's hooking up its flash memory, citing "patent discussions," but said it is "deeply embedded."

Computer scientists have long looked at brains for inspiration, and in recent decades, one avenue has been in self-learning neural networks that behave in some ways like a brain's teeming interconnected neurons. The Blue Brain project, though, isn't a neural network. Instead, it seeks to simulate the physical reality of real nerve cells, including their shape, size, and electrical behavior, IBM said.