Facilities, Equipment, and Other Resources Statement

Download PDF The following is a draft Facilities, Equipment and Other Resources statement that researchers can include in research proposals:

Virginia Tech’s Advanced Research Computing (ARC) is a unit within the Division of Information Technology at Virginia Tech and provides centralized research computing resources which are availale to any Virginia Tech faculty member. ARC provides cutting-edge high-performance computing and visualization resources. Currently available high performance computing (HPC) systems include:

  1. TinkerCliffs: a general purpose CPU and high-end GPU cluster. This cluster has approximately 40,000 AMD Rome CPU cores, HDR Infiniband offering 100 Gbps throughput, nodes for high-memory applications, an additional 16 Intel Xeon AP nodes and fourteen nodes with eight NVIDIA A100-80GB GPUs each

  2. Infer: GPU-based cluster made up of 98 compute nodes with a total of 80 NVIDIA Volta V100 GPUs, 18 NVIDIA Tesla T4 GPUs, and 80 NVIDIA Tesla P100 GPUs; Infiniband interconnect

  3. CUI or Protected data: This is a dedicated cluster and storage supporting some types of data which need elevated protections such as ITAR or Export Controlled software/data. This cluster will became available early in the winter of 2021-2022. Is has 12 CPU nodes and 3 dense GPU nodes. The CPU nodes are HPE XL225n servers equipped with 2x AMD EPYC Rome cpus providing 128 cpu-cores and 512GB memory. The GPU nodes are HPE Apollo 6500 Gen10+ with the same AMD EPYC Rome cpus, but have 2048GB memory and 8x Nvidia Ampere A100-80GB GPU accelerators.

  4. (Coming 2024) Owl: high-speed, water-cooled CPU cluster with favorable memory architectures. This cluster has 84 nodes with each with 96 AMD “Genoa” CPU cores, 768GB DDR5 memory (8GB/core), and interconnected with HDR Infiniband running at 100Gbps. Additionally, it has 3 specialty nodes each with either 4TB or 8TB main system memory.

Centralized storage clusters provide over 11 Petabytes of long-term bulk “project” storage and flash-based, cluster-localized “scratch” storage for workflows with demanding i/o needs.

ARC’s Visionarium Lab provides an array of visualization resources, including the VisCube, an immersive 10′ x 10′ three-dimensional visualization environment. In all, the VT Visionarium provides nearly 86 million pixels, 4 billion triangles-per-second and 22 TB/s of GPU memory bandwidth. ARC resources are able to leverage Virginia Tech’s excellent network connectivity, and network. Virginia offers access to advanced national networks, including ESnet, Internet2, and Mid Atlantic Crossroads.