LiCCAG: Front view

Front view of some of the racks
Rear view of one of the racks

Nodes

ALL

  • A head/login gpu node liccag001 with 2x64 core AMD Epyc-9575F 3.3 GHz CPU-s, 384 GByte of memory, 2.7TB local scratch storage, 2.9TB NVME scratch storage, and 2xNvidia H200 GPUs ;
  • 7x gpu compute nodes, each with 2x64 core AMD Epyc-9575F 3.3 GHz CPU-s, 1.5 TByte of memory, 680GB local scratch storage, 5.8TB NVME scratch storage, and 8xNvidia H200 GPUs ;
  • 1x HGX gpu compute node, with 2x64 core Intel XEON 6767P 2.4 GHz CPU-s, 2 TByte of memory, 680GB local scratch storage, 3.5TB NVME scratch storage, and 8xNvidia B200 GPUs;
  • All nodes are connected by a 2x400 GB/sec Nvidia Mellanox ConnectX-7;
  • SLURM is used for resource allocation and job scheduling;
  • All nodes run Ubuntu Linux 24.04 LTS.

Login, Interactive, and Data Transfer Nodes

A head/login gpu node liccagpu-li-01.rz.uni-augsburg.de :

  • 1x gpu node (liccag001);
  • Processor type AMD Epyc-9575F;
  • Processor base frequency 3.3 GHz CPU-s;
  • Cores per node:  2x64;
  • Main memory (RAM) per node 768 GByte;
  • 8 NUMA domains with 16 physical cores each;
  • 2.7TB local scratch storage;
  • 2.9TB NVME scratch storage;
  • GPU type Nvidia H200;
  • GPU-s per node: 2;

GPU Access

GPU Access only via Slurm allocation.


Test GPU Nodes

  • 1x gpu node (liccag002);
  • Processor type AMD Epyc-9575F;
  • Processor base frequency 3.3 GHz CPU-s;
  • Cores per node:  2x64;
  • Main memory (RAM) per node 1.5 TByte;
  • 8 NUMA domains with 16 physical cores each;
  • 680GB local scratch storage;
  • 5.8TB NVME scratch storage;
  • GPU type Nvidia H200;
  • GPU-s per node: 8;

General Purpose GPU Nodes

  • 6x gpu nodes (liccag[003-008]);
  • Processor type AMD Epyc-9575F;
  • Processor base frequency 3.3 GHz CPU-s;
  • Cores per node:  2x64;
  • Main memory (RAM) per node 1.5 TByte;
  • 8 NUMA domains with 16 physical cores each;
  • 680GB local scratch storage;
  • 5.7TB NVME scratch storage;
  • GPU type Nvidia H200;
  • GPU-s per node: 8;

Special Purpose GPU Nodes (HGX  Blackwell)

  • Nvidia HGX Platform;
  • 1x gpu node (liccag009);
  • Processor type Intel XEON 6767P;
  • Processor base frequency 2.4 GHz CPU-s;
  • Cores per node:  2x64;
  • Main memory (RAM) per node 2 TByte;
  • 4 NUMA domains with 32 physical cores each;
  • 680GB local scratch storage;
  • 3.5TB NVME scratch storage;
  • GPU type Nvidia B200;
  • GPU-s per node: 8;

I/O Subsystem

  • 1 PByte of shared disk space

Queue Partitions

HPC Resource

Name


Partition


Timelimit


number of Nodes


Nodes

Per-Node


Purpose

CPUCoresRAMRAM/CoreGPU
LiCCAG-h200-devh200-dev30 minutes1liccag0012×AMD Epyc-9575F2×64768GiB6GiB2×Nvidia H200short queue for development on the login node
LiCCAG-h200-testh200-test2 hours1liccag0022×AMD Epyc-9575F2×641.5TiB12GiB8×Nvidia H200queue for testing and short calculations
LiCCAG-h200h2002 days6liccag[3-8]

2×AMD Epyc-9575F

2×641.5TiB12GiB8×Nvidia H200general purpose GPU nodes
LiCCAG-b200b2002 days1liccag0092×Intel XEON 6767P2×64

2TiB

16GiB8×Nvidia B200HGX Blackwell special purpose GPU nodes (code must use multiple GPUs)
 ∑-9--

1152

13Tib-

58×Nvidia H200, 8×Nvidia B200

-

Special Systems