Hardware

ARC is UC’s large-memory computer cluster. Designed to support data-intensive computing, ARC is particularly well-suited for running software and applications that require large amounts of computer memory and processing. ARC provides specialized deep GPU partitions for researchers with deep learning applications and other applications that require GPUs. ARC currently has 2 major clusters, namely ARC1 and ARC2. The hardware specifications for each are summarized in the table below

ARC 1 (pilot cluster)

Number of nodes Processors per node Cores per nodes Memory per node Omnipath Accelerator card Speed
36 2 Intel Xeon Gold 6148 CPUs 40 192 GB 100 GB/s N/A 2.4-3.7 GHz
1 2 Intel Xeon Gold 6148 CPUs 40 192 GB 100 GB/s 2 NVIDIA Tesla V100-32GB GPU 2.4-3.7 GHz

ARC 2

Node Type Number of nodes Processors per node Cores per nodes Memory per node InfiniBand Accelerator card Speed
RM 95 2 AMD EPYC 7452 CPUS 64 256GB RAM 100 GB/s N/A 2.35-3.35 GHz
GPU 10 2 AMD EPYC 7452 CPUS 64 1024GB RAM 100 GB/s 2 NVIDIA Tesla A100-40GB GPU 2.35-3.35 GHz
Large Mem 1 2 AMD EPYC 7452 CPUS 64 1024GB RAM 100 GB/s N/A 2.35-3.35 GHz

Coming Soon! UC’s newest High-Performance Computing, AI and High-Performance Data Analytics (HPDA) system

UC’s newest high-performance computing and data analytics system, temporarily referred to as ARCC-2, is coming soon with plans for early operations to begin in late summer 2021 and broad availability Fall 2021.  It is funded in part by investments from the Office of Research, IT@UC, colleges and departments and a significant grant from the National Science Foundation’s Major Research Instrumentation (MRI) program (award #2018617). UC is partnering with Hewlett Packard Enterprises (HPE) to architect a purpose-built compute resource for demanding High-Performance Computing (HPC) and Artificial Intelligence (AI) applications. Final specifications may change due to availability, pricing and user requirements.

ARCC-2 will provide transformative capability for rapidly evolving, computation-intensive and data-intensive research, supporting both traditional and non-traditional research communities and applications. The converged, scalable HPC, machine learning and data tools create an opportunity for collaboration and converged research, prioritizing researcher productivity and ease of use with an easy-to-use web-based interface. 

Core Concepts 

  • Converged HPC + AI + Data 
  • Custom topology optimized for data-centric HPC, AI and HPDA (High Performance Data Analytics) 
  • Heterogeneous node types for different aspects of workflows 
  • CPUs and AI-targeted GPUS

Innovation 

  • AMD EPYC 7452 CPUs: 32-core 2.35–3.35 GHz  
  • AI scaling to 20 Tesla A100-40GB GPUs  
  • Mellanox HDR-100 InfiniBand supports in-network MPI-Direct, RDMA, GPUDirect, SR-IOV, and data encryption  
  • Cray ClusterStor E1000 Storage System  
  • Open OnDemand – Web based interface 

Regular Memory 
Regular Memory (RM) CPU nodes provide extremely powerful general-purpose computing, machine learning and data analytics, AI inferencing, and pre- and post-processing.   95 RM nodes will have: 

  • Two AMD EPYC CPUS, each with: 
  • 32 cores 
  • 2.35-3.35GHz 
  • 128MB L3 
  • 256GB of RAM 
  • 8 memory channels 
  • SATA SSD (960GB) 
  • Mellanox ConnectX-6 HDR InfiniBand 100Gb/s Adapter 

Large Memory 
Large Memory (LM) node will provide 1TB of shared memory for genome sequence assembly, graph analytics, statistics, and other applications requiring a large amount of memory for which distributed-memory implementations are not available. 

ARCC-2’s 1 LM nodes will consist of: 

  • Two AMD EPYC CPUS, each with: 
  • 32 cores 
  • 2.35-3.35GHz 
  • 128MB L3 
  • 8 memory channels 
  • 1024GB of RAM
  • Mellanox ConnectX-6 HDR InfiniBand 100Gb/s Adapter 

GPU 
10 GPU nodes provide exceptional performance and scalability for deep learning and accelerated computing.  Each GPU node will contain: 

  • Two NVIDIA Tesla A100 40GB GPUs 
  • Two AMD EPYC CPUS, each with: 
  • 32 cores 
  • 2.35-3.35GHz 
  • 128MB L3 
  • 8 memory channels 
  • 1024GB of RAM 
  • SATA SSD (960GB) 
  • Mellanox ConnectX-6 HDR InfiniBand 100Gb/s Adapter 

Please contact Jane Combs or arc_info@uc.edu with questions or interest in purchasing priority/boosted fair share access to nodes for your time-sensitive research. 

Contribute nodes to the cluster – priority access to your nodes plus access to additional shared resources

Faculty can use their HPC and research computing funding to contribute nodes to the central cluster. Priority access is given to the owner of the nodes, however, when not in use by the owner, the nodes can be shared with others. This is a good option for faculty who need to have full access to their nodes periodically and can take advantage of access to additional shared resources in the cluster. Using the central resource maximizes the amount of compute resources a faculty can purchase because the HPC infrastructure (networking, racks, head/management nodes, support) are provided at no cost. Contact: arc_info@uc.edu

Cost: Nodes contributed to the cluster must be consistent with current cluster hardware configurations. The ARC team can work with you to review your needs and provide an estimate for your purchase.