1 of 10

ARC Town Hall Meeting

September 23rd, 2026

Advanced Research Computing

Division of Information Technology

Virginia Tech

Sign in

Ben Sandbrook

Director of Research Computing Services

Matthew Brown

Director of Research Software Engineering

2 of 10

Agenda

​

  • ARC “Getting Started” Resources
  • FY26 Owl Expansion
  • Steger Hall Datacenter Maintenance
  • SuperComputing 2026
  • ARC LLM Services Overview
  • Open Q&A

To ensure time to cover all the prepared materials, Please hold questions until the Q&A period.

Sign in

1

3 of 10

3

GETTING STARTED

Faculty Support

  • VT faculty supervise all long-term use of ARC systems
  • Faculty members can add you to their “project”:�https://coldfront.arc.vt.edu
  • Free Tier: 2M compute hours, 640GB in /home, 50TB in /projects

New to HPC or Cluster Computing?

  • Start with OnDemand web interface and work toward batch submission
  • Fall Workshop Series (Zoom):

Wednesdays: 4:00 PM – 5:00 PM

Fridays: 12:00 PM – 1:00 PM

ARC in Your Curriculum

  • Reach out for a consultation to discuss incorporating ARC and best practices
  • Teaching on ARC Guide
  • We will create an “Instructional Allocation” for your course

4 of 10

Useful Info & Links

Office Hours

https://arc.vt.edu/office-hours

  • 1 hour every day on Zoom
  • 1 hour every day in person
  • No appointment needed

Workshops

https://docs.arc.vt.edu/usage/workshops.html

  • Weekly training & skill sessions
  • Wednesdays: 4:00 PM – 5:00 PM
  • Fridays: 12:00 PM – 1:00 PM

Online Documentation

https://docs.arc.vt.edu/

Support & Help Tickets

https://arc.vt.edu/help

Need technical assistance, system help, or account support?

Submit a ticket directly to the ARC support team.

5 of 10

FY26 Owl Expansion

Each year, ARC requests funding for computation resources based on recent usage and expected trends in demand.

​

FY26: CPU expansion and additional high-end GPU resources

CPU: direct-water-cooled AMD EPYC Zen4 nodes expansion

        • 74 additional base nodes (new total 160 nodes)
          • 96 cores, 8 GB/core system memory (identical to existing)
        • 4x nodes x 96 cores (9474F) x 3 TB per node
        • 4x nodes x 128 cores (9554) x 3 TB per node

GPU: 4x nodes with 8-way NVIDIA B200 GPUs

        • New FP4 Tensor Core (NVFP4 LLM quantization)
        • Direct water cooled

​

GPU

A100

H200

B200

Device memory

80G

141G

180G

FP64 Tensor

FLOPS

19.5T

67T

37T

FP4 Tensor

FLOPS

-

-

9000T

INT8

Tensor OPS

1248T

3958T

9000T

Memory BW

2.04TB/s

4.8TB/s

7.7TB/s

5

6 of 10

6

ARC INFRASTRUCTURE

HPC CLUSTERS

​

Storage: ARC mounts a 10 PB projects workspace which is accessible by all clusters

Cluster

Nodes

CPUs

GPUs

RAM

Scratch

Purpose

TinkerCliffs

332

41,600

160

130 TB

1400 TB

CPU and GPU flagship (A100, H200)

Owl

169

16,256

32

141 TB

1200 TB

Water-cooled AMD CPUs and GPUs (B200)

Falcon

106

4,768

298

43 TB

910 TB

Mid-range NVIDIA GPUs (L40s, A30, V100, T4)

CUI

12

768

16

9 TB

-

DoD and ITAR projects (A100)

Biomed

3

192

8

3 TB

-

NIH and NIST 800-171 projects (A100)

Total

622

63,584

514

326 TB

3.5 PB

​

7 of 10

Datacenter Maintenance

​

Background

      • A minor problem has surfaced in the Steger Hall data center’s cooling infrastructure
      • ARC is working with VT Facilities and system contractors on a remediation
      • A short service outage will be needed for the work

​

Impact

      • (primary) The Owl cluster will need to be taken offline for a few hours
      • (secondary) The Tinkercliffs cluster will remain online but will temporarily run on reduced cooling capacity
      • Timing: TBD, but very soon - Likely within the next 1-2 weeks

​

​

2

8 of 10

SuperComputing 2026 Conference

​

      • November 15-20 in Chicago
      • https://sc26.supercomputing.org
      • ARC reprising the “Virginia is for SuperComputing” booth on the exhibit floor
      • Submit research highlights for scrolling display or ask about options to give a short talk

brownm12@vt.edu

2

9 of 10

Q&A

10 of 10

10