Systems Overview Table
NERSC Computational Systems
System Name | System Type | CPU | Computational Pool | Node Interconnect | Scratch Disk | Avg. Power (KW) | |||||||
Type | Speed (GHz) | Nodes | SMP Size | Total Cores | Flops per Core (Gflops/sec) | Peak Performance (Tflops/sec) | Aggregate Memory |
Avg. Memory/core |
|||||
Hopper | Cray XE6 | Opteron | 2.1 | 6,384 | 24 | 153,216 | 8.4 | 1287.0 | 211.5 TB | 1.41 GB | Gemini | 2.2 PB (local) + 1.1 PB (global) | 2,200 |
Carver | IBM iDataPlex | Nehalem, Westmere, Nehalem-EX | 2.67, 2.00 | 1,202 | 8, 12, 32 | 9,984 | 10.68, 8.00 | 106.5 | 35.75 TB | 3.67 GB | QDR InfiniBand | 1.1 PB (global) | 266 |
Dirac* | NVIDIA GPUs on IBM iDataPlex | Tesla C2050 (Fermi) and C1060 with Nehalem | 1.15 or 1.30 for GPUs and 2.4 for CPUs | 56 GPU nodes on 50 CPU nodes | 448 or 240 for GPUs, and 8 for CPUs | 23,424 GPU cores and 400 CPU cores | 1.15 or 1.30 for GPU cores and 9.6 for CPU cores | 25.4 for GPUs and 3.8 for CPUs | 176 GB for GPUs and 1,344 GB for CPUs | 7.7 MB for GPUs and 3.36 GB for CPUs | QDR InfiniBand | 1.1 PB (global) | - |
PDSF** | Linux Cluster | Opteron, Xeon | 2.0, 2.27, 2.33, 2.67 | 194 | 8, 12 | 1,844 | 8.0, 9.08, 9.32, 10.68 | 17.6 | 6.4 TB | 4 GB | Ethernet | 34.9 TB for batch nodes and 184 GB for interactive nodes | 92 |
Genepool*** | Various vendor systems | Nehalem, Opteron | 2.27, 2.67 | 547 | 8, 24, 32, 80 | 4,680 | 9.08, 10.68 | 42.8 | 33.7 TB | 7.36 GB | Ethernet | 1.1 PB (global) | - |
* Dirac is a testbed experimental system and is not considered a NERSC production system.
** PDSF is a special-use system hosted by NERSC for the High Energy Physics and Nuclear Science community.
*** Genepool is a cluster dedicated to the DOE Joint Genome Institute's computing needs.
See more information at NERSC computational systems.
NERSC File Systems
The following table shows NERSC file systems, their size and availability on computational systems.
File System | Size | Hopper | Carver | Dirac | PDSF | Euclid | Genepool | Data Transfer Nodes |
Global homes |
246 TB |
Y | Y | Y | N | Y | Y | Y |
Global scratch | 1.1 PB | Y | Y | Y | N | Y | Y | Y |
Global project | 3.9 PB | Y | Y | Y | Y | Y | Y | Y |
Global projectb | 2.6 PB | Y | Y | Y | N | N | Y | Y |
Local scratch | - | Y (2.2 PB) | N | N | Y (34.9 TB for batch nodes) | N | Y | N |
Others | - | N | N | N | N | Y (2.1 PB for /house and /ifs) | N |
See more information at NERSC file systems.
NERSC HPSS (High Performance Storage System) Mass Storage Systems
Storage System | Usage | Maximum Capacity | Disk Cache | Tape Drives | Maximum Aggregate Bandwidth* |
Archive (archive.nersc.gov) | Storing user files | 150 PB | 250 TB | 102 (Oracle STK 9840D: 26; Oracle STK T10KB: 50; and Oracle STK T10KC: 26) | 12 GB/sec |
Regent (hpss.nersc.gov) | Computer system backups | 90 PB | 50 TB | 42 (Oracle STK 9840D: 8; Oracle STK T10KB: 26; and Oracle STK T10KC: 8) | 4 GB/sec |
* An estimated total aggregate bandwidth of the disk subsystems only, not all the interface in the system.
See more information at NERSC data archive systems.