History of Supercomputing

Succession of Supercomputers at the RZG Garching (now: MPCDF) since 1962
January 01, 1962

1962-1969: IBM 7090

0.1 MFlop/s, 128 kB RAM
© MPCDF
January 01, 1969

1969-1980: IBM 360/91

15 MFlop/s, 2 MB RAM
© MPCDF
January 01, 1979

1979-1986: Cray-1

80 MFlop/s, 8 MB RAM
© MPCDF
January 01, 1986

1986-1991: Cray XMP/2

400 MFlop/s, 32 MB RAM
© MPCDF
January 01, 1991

1991-1997: Cray YMP/4

1.3 GFlop/s, 512 MB RAM
© MPCDF
January 01, 1991

1991-1998: nCUBE2/64

141 MFlop/s, 256 MB RAM
© MPCDF
January 01, 1995

1995-1997: Cray T3D/128

19 GFlop/s, 8 GB RAM
© MPCDF
January 01, 1996

1996-2002: Cray T3E/816

0.47 TFlop/s, 104 GB RAM
© MPCDF
January 01, 1999

1999-2005: NEC SX-5/3C

24 GFlop/s, 12 GB RAM
© MPCDF
January 01, 2002

2002-2008: IBM p690

5 TFlop/s, 2 TB RAM
© MPCDF
January 01, 2007

2007: IBM BlueGene/P

54 TFlop/s, 8 TB RAM
© MPCDF
January 01, 2008

2008: IBM p575 Power6 System

125 TFlop/s, 20 TB RAM
© MPCDF
January 01, 2013

2013-2018: Hydra

Peak Performance:    2.7  PetaFlop/s (aggregated)

Compute  nodes:      > 4000  (Intel Xeon Sandy + Ivy Bridge processors)

Accelerators:  676 NVIDIA K20X GPUs, 24 Intel Xeon Phi (Knights Corner)

Total main memory:   ~ 250 TeraByte

Interconnect:   Infiniband FDR14 (56 Gb/s)

© MPCDF
January 01, 2016

2016: Draco

Peak Performance:    1.0 PetaFlop/s

Compute  nodes:      875 (Intel Xeon Haswell processors)

Accelerators:  204 NVIDIA GTX970 GPUs:      

Total main memory:   ~  100 TeraByte

  769 x 128 GB, 4 x 256 GB, 4 x 512 GB

Interconnect:   Infiniband FDR14 (56 Gb/s)

© MPCDF
January 01, 2018

2018: Cobra

Peak Performance:  ~ 10 PetaFlop/s

Compute nodes:    3188 (each with 2 Intel Xeon Skylake processors with 20c @ 2.4 GHz)

Compute Cores:  ~ 130,000

Total main memory:   ~ 0.5 PetaByte

Main memory:  1272 x 96 GB RAM, 1808 x 192 GB RAM, 8 x 768 GB RAM

Interconnect:    Intel OmniPath (100 Gb/s)

File system:   IBM Spectrum Scale

Mass storage:   Automatic tape access (via GHI to HPSS)
© MPCDF
January 01, 2018

2020: Raven

Peak Performance:  ~ 7.5 PetaFlop/s
Compute nodes:    1592 (each with 2 Intel Xeon Icelake CPUs with 72 cores)
Compute Cores:  ~ 115,000
Total main memory:   ~ 0.375 PetaByte
GPU-accelerated nodes: 192 (each with 4 Nvidia A100 SXM4-40GB GPUs)
Interconnect:  InfiniBand HDR (100 - 400 Gb/s)
File system:   IBM Spectrum Scale
Mass storage:   Automatic tape access (via GHI to HPSS)
© MPCDF
Web-View
Print Page
Open in new window
Estimated DIN-A4 page-width
Go to Editor View