Configuration

NERSC's newest supercomputer, named Edison after U.S. inventor and businessman Thomas Alva Edison, will have a peak performance of more than 2 petaflops (PF, or 10¹⁵ floating point operations per second) when fully installed in 2013. The integrated storage system will have more than 6 petabytes (PB) of storage with a peak I/O bandwidth of 140 gigabytes (GB) per second. The product is known as a Cray XC30 (internal name "Cascade"), and the NERSC acquistion project is known as "NERSC 7."

Edison will be installed in two phases. 

Phase I

Installation: 4Q 2012
Early User Access: Started in February 2013.  All users were enabled March 2, 2013.

System Overview

• Cray Cascade supercomputer
• 664 computes nodes with 64 GB memory per node
• Two 8-core Intel "Sandy Bridge" processors per node (16 cores per node)
• 10,624 total physical compute cores
• Cray Aries high-speed interconnect (0.25 μs to 3.7 μs MPI latency, ~8 GB/sec MPI bandwidth)
• Scratch storage capacity: 1.62 PB

System Details

• Compute processor: 8-core Intel "Sandy Bridge" at 2.6 GHz
• Compute node: dual-socket Sandy Bridge with 64 GB DDR3 1600 MHz memory (8 GB DIMMs)
• Compute blade: 4 dual-socket nodes
• Number of compute nodes: 664
• "MOM" nodes (execute job scripts): 8 repurposed compute nodes
• High speed interconnect: Cray Aries with Dragonfly topology
• Scratch storage system: Cray Sonexion 1600 Lustre appliance
• Scratch storage maximum bandwidth: 36 GB/sec
• Login nodes: quad-core, quad-socket (16 total cores) 2.0 GHz Intel "Sandy Bridge" processors with 512 GB memory.
• Number of login nodes: 6
• Shared root server nodes: 8
• Lustre router nodes: 7
• DVS server nodes (for interface with NERSC Global File System): 16
• External gateway (network nodes): 4 nodes with 2 dual-port 10 GigE interfaces per node

Phase 2

Installation: second half of 2013

System Overview

• Cray Cascade supercomputer
• Sustained application performance on NERSC SSP codes: 236 Tflop/s (vs. 144 Tflop/s for Hopper)
• Aggregate memory: 333 TB
• 5,200 computes nodes with 64 GB memory per node
• Cray Aries high-speed interconnect (0.25 μs to 3.7 μs MPI latency, ~8GB/sec MPI bandwidth)
• Scratch storage capacity: 6.4 PB

System Details

• Intel multicore processors
• Compute blade: 4 dual-socket nodes
• Number of compute nodes: 5,200
• "MOM" nodes (execute job scripts): 8 repurposed compute nodes
• High speed interconnect: Cray Aries with Dragonfly topology
• Scratch storage system: Cray Sonexion 1600 Lustre appliance
• Scratch storage maximum aggregate bandwidth: 140 GB/sec
• Login nodes: quad-core, quad-socket (16 total cores) 2.0 GHz Intel "Sandy Bridge" processors with 512 GB memory.
• Number of login nodes: 12
• Shared root server nodes: 8
• Lustre router nodes: 7
• DVS server nodes (for interface with NERSC Global File System): 16
• External gateway (network nodes): 4 nodes with two dual-port 10 GigE interfaces per node