Main components of the infrastructure
The Blue Brain workflow depends on a large-scale computing and data infrastructure:
- The Blue Brain Project’s HPE SGI 8600 supercomputer (Blue Brain 5) is comprised of 327 compute nodes delivering 1.06 petaflops of peak performance. The system is equipped with 94 terabytes of memory a memory equivalent of 23,000 laptops and runs Intel®Xeon®Gold 6140 and Intel®Xeon PhiTM 7230 processors as well as NVIDIA®Tesla®V100 graphic processors. The system uses single and dual-rail Mellanox InfiniBand high-performance networks and has 4 petabytes of high-performance storage from DataDirect Networks (DDN), delivering more than 50GB/s aggregated bandwidth, associated with an innovative 80 GB/s Infinite Memory Engine (IME) flash-based burst buffer.With the HPE SGI 8600 system, designed to solve the most complex problems and powering some of the fastest supercomputers in the world, Blue Brain gains scalability to thousands of nodes in an easy-to-manage architecture. The system also features an energy efficient liquid cooling solution that does not exhaust heated air into the data center. The new system has been installed at the Swiss National Supercomputing Centre (CSCS) in Lugano, Switzerland.
In collaboration with EPFL’s Laboratory for Neural Microcircuitry (LNMC):
- State of the art technology for the acquisition of data on different levels of brain organization (multi-patch clamp set-ups for studies of the electrophysiological behavior of neural circuits, Multi-electrode Arrays – (MEAs) allowing stimulation of and recording from brain slices, facilities for the creation and study of cell lines expressing particular ion channels, a variety of imaging systems, systems for the 3D reconstruction of neural morphologies).
The success of the Blue Brain project depends on very high volumes of standardized, high quality experimental data, covering all possible levels of brain organization. Data comes from the literature (via the project’s automatic information extraction tools), from the Human Brain Project, from large data acquisition initiatives outside the project, and from the EPFL’s Laboratory for Neural Microcircuitry (LNMC).
High Performance Computing
The Blue Brain workflow creates enormous demands for computational power. In Blue Brain cellular level models, the representation of the detailed electrophysioloy and communication of a single neuron can require as many as 20,000 differential equations. Even modern multi-core workstations have difficulties of solving this number of equations in biological real time. In other words, the only time to ensure reasonable turn-around times for simulating networks of neurons is the use of High Performance Computing (HPC).
The Blue Brain project’s simulation of the neocortical column incorporates detailed representations of at least 30,000 neurons – in most case several times this number in order to provide valid boundary conditions. A simulation of a whole brain rat model at the same level of detail would represent up to 200 million neurons and would require approximately 10,000 times more memory. Simulating the human brain would require yet another 1,000-fold increase in memory and computational power. Subcellular modeling, modeling of the neuro-glial vascular system and the creation of virtual instruments (e.g. virtual EEG, virtual fMRI) will further expand these requirements.
In the initial phase of its work the Blue Brain project used an IBM BlueGene/L supercomputer with 8,192 processors before moving to a 16,384 core IBM BlueGene/P supercomputer with almost 8 times more memory than its predecessor. Then, it used an IBM BlueGene/Q supercomputer with 65,536 cores and extended memory capabilities hosted by the Swiss National Supercomputing Center (CSCS) in Lugano. In July 2018, Blue Brain announced it had selected HPE to build a next-generation supercomputer for modeling and simulation of the mammalian brain. The new supercomputer, called ‘Blue Brain 5’, will be dedicated to simulation neuroscience, in particular simulation-based research, analysis and visualization, to advance the understanding of the brain.