====== Goethe-HLR ======

Goethe-HLR is a general-purpose computer cluster based on several CPU/GPU architectures. The system was installed in Industriepark Höchst in 2019 and extended with GPU nodes in 2020.


**Cluster performance:**
  * Cluster performance peak: 1,449.98 TFlop/s (with 472 nodes)
  * Cluster performance HPL (2018): 958.67 TFlop/s (with 472 nodes)
  * GPU cluster performance HPL (2023): 9,087.00 TFlop/s (with 105 nodes)

**Hardware:**
  * 730 compute nodes in water-cooled racks
    * initially 472 at the end of 2018 
  * thousands of CPU cores
  * 864 AMD Radeon Instinct MI210 GPUs
  * 192, 768, 128 or 512 GB of RAM per node
  * EDR and FDR InfiniBand interconnects (2:1 blocking fat tree network)
  * a parallel scratch filesystem with a capacity of 2.4 PB and an aggregated bandwidth of 27 GB/s

^  Nodes  ^  CPU  ^  Cores \\ per CPU  ^  Cores \\ per Node  ^ Threads \\ per Node ^  RAM \\ per Node  ^ IB ^ GPU ^
|  412 |2x Intel Xeon Gold 6148 (Skylake) |  	20 |  40|  80|	192 GB | EDR |n/a |
|   72 |2x Intel Xeon Gold 6148 (Skylake) |  	20 |  40|  80|	768 GB | EDR |n/a |
|  138 |2x Intel Xeon E5-2640 v4 (Broadwell) |  	10 |  20|  40|	128 GB | FDR |n/a |
|  108 |2x AMD EPYC 7452                  |      32 |  64|  128|  512 GB | EDR |8xAMD Instinct MI210 64 GB|

\\