B07.01 HPC Centers Presentation HITS-U III Industry Day 20260624 Slide Deck 2.pptx

HPCMP Centers Overview HITS-U III Industry Day Frank Witzeman, Associate Director for Centers 24 Jun 2026

Approved for Public Release

Page-‹#›

How is HPC Different Than Typical IT?

HPC Systems are very large containing different node types (user login, data transfer, data visualization, AI/ML, compute) ~1,000’s of nodes!

Nodes are like individual computers that contain processor cores (very many!), memory (large!), data storage – information shared among nodes through high-performance interconnect network Integrated high-performance data file systems (very large!) are used for user working files, data input & output from HPC applications HPC Systems also connect to global networks for access to external systems, other HPCs The core is the most basic unit of processing power For HPC Systems, the number of node types, number of cores & GPGPUs inside each node, & total number of nodes, vary widely!

Standard Node

96-192 Cores Memory 200-2,000+ GB/Node

GPGPU Node

96-192 Cores Memory 200-700+ GB/Node

1-8 GPGPUs Parallel File Systems for Data Input/Output

Interconnects w/ other nodes & external systems/networks Interconnects w/ other nodes & external systems/networks HPC Systems operate round-the-clock & continuously manage & process 1,000’s of use cases/workflows & require significant power & cooling to operate

HPCMP Centers At-A-Glance Today DSRCs host 27 HPC systems (Unclassified & Restricted Access/Classified) ~2.5M cores, ~2.2K GPUs, ~145 PetaFLOPS

Coming 2026 – 8 HPC systems (Unclassified & Restricted Access/Classified) ~0.7M cores, ~0.4K GPUs, ~47 PetaFLOPS Cross-Centers/Enterprise Team (across multiple states) Budget & Staff for DSRC Operations HPC Technology Insertion (TI) HPC System Administration & Maintenance HPC Help Desk HPC Benchmarking HPC Authentication Services Advanced Visualization Center (DAV-C) Air Force Research Laboratory (AFRL) DSRC Wright-Patterson AFB, OH

Army Research Laboratory (ARL) DSRC Aberdeen Proving Ground, MD

Engineer Research & Development Center (ERDC) DSRC Vicksburg, MS

Navy DSRC Stennis Space Center, MS

Maui HPC Center (MHPCC) DSRC Kihei, Maui, HI DSRC = DoD Supercomputing Resource Center

HPCMP Centers Mission Operate DSRCs, Deliver HPC Capabilities, Help Users Mission Area Elements DSRC Operations Unclassified HPC Classified HPC Specialized HPC Enterprise Services

HPCMP Centers Mission Elements/Functions DSRC Operations (per DSRC) DSRC Directors, Deputies, Tech Leads Ktr Site Leads, PMs Office Admin Support Program Mgrs/ Financial Mgrs/ CORs/COTRs Cybersecurity Physical Security Advanced Tech Dev Facility Infrastructure Support HPC Infrastructure Support DSRC Operations & Maintenance DSRC Travel, Training, Supplies Unclassified HPC (per DSRC) HPC Sys Integration HPC Sys Admin HPC Sys Maintenance Application Support Workload Mgmnt Help Desk Liaison/ Customer Support Data Storage Secret-Level HPC (per DSRC) HPC Sys Integration HPC Sys Admin HPC Sys Maintenance Application Support Workload Mgmnt Security above Ops Customer Support Data Storage Above-Secret HPC (per DSRC) HPC Sys Integration HPC Sys Admin HPC Sys Maintenance Unique App Support Access Support High-Level Security Unique Cust Support Specialized HPC (per DSRC) Sys Integration Sys Admin Sys Maintenance RDTE Proj Support Enterprise/Cross-Centers (across all DSRCs) AD Centers Technical Support Technical Assessments Project Management Financial Management Contract Management HPCMP Liaisons Enterprise Baseline Ops Outreach/Web Dev Enterprise S/W Management Technology Evaluation Benchmarking (ERDC DSRC) Enterprise Serv/Sys Support Procurements Performance & Business Management (“Back Office”) User (& Other) Support Help Desk (AFRL DSRC) Authentication Services (AFRL DSRC) DAV-C Adv Viz (ARL & ERDC DSRCs)

HPCMP Centers Team (Core, Cross-Centers, DSRCs) AD for Centers Frank Witzeman Tech Advisor(s)/Project Mgr(s) HPCMP Cybersecurity Liaison*

HITS-U COR*

HPCMP Contracts Lead* Budget/Finance Analyst HITS-U Program Manager Enterprise Services, User Experience, Management/Back-Office,… HITS-U Cross-Centers Team (HITS-U CC Team Leader)

AFRL DSRC

Bryon Foster

ARL DSRC

Sophia Paros (Acting)

ERDC DSRC

Ben Parsons Navy DSRC Christine Cuicchi Help Desk & Auth Serv

DAV-C

DAV-C & Benchmarking

MHPCC DSRC

Scott Pierce HITS-U + Local CTR HITS-U + Local CTR HITS-U + Local CTR HITS-U + Local CTR Local CTR (No HITS-U reqmnts except Enterprise)

* Provided in-kind by HPCMP

* Provided in-kind by HPCMP Notional HITS-U organizational structure for Cross-Centers

AFRL DSRC

Floor space – ~32,000 ft2 Power – ~21,000 kW Cooling – ~4,800 Tons Focus areas Unclassified HPC Classified HPC HPC Help Desk & Authentication Services Outreach to new users in acquisition engineering Hosting of both HPCMP & AF special-purpose systems “Warhawk” (HPE/Cray EX) unclassified 139,776 cores (AMD 7H12) & 104 GPUs (NVIDIA V100) “Raider” (Penguin TrueHPC) unclassified 189,440 cores (AMD 7713) & 152 GPUs (NVIDIA A40, A100) Coming 2026: “Flyer” (HPE/Cray XD) unclassified 172,800 cores (AMD 9654) & 80 GPUs (NVIDIA L40, H100) Adding 168 GPUs (NVIDIA GH200) Coming 2026-27: 1 unclassified & 3 classified HPC systems ~240,000 cores & ~100 GPUs total 6 classified HPC systems ~170,000 cores & ~140 GPUs total Support for PDUs, UPS, Generators, Chillers, Cooling Towers

ARL DSRC

Floor space – ~26,000 ft2 Power – ~15,000 kW Cooling – ~4,000 Tons Focus areas Unclassified HPC Classified HPC Dedicated AI Systems (new) Persistent Data Services Data Analysis Support & Visualization (DAV-C) “Jean” (Liqid Composable) unclassified 55,392 cores (Intel 9242) & 294 GPUs (NVIDIA A100) “Ruth” (HPE EX4000) unclassified 133,632 cores (AMD 9654) & 88 GPUs (AMD MI250)

Coming 2026: “Fran” (HPE EX4000) unclassified 173,184 cores (AMD 9654) & 88 GPUs (NVIDIA L40, H200)

4 classified HPC systems ~260,000 cores & ~380 GPUs total Coming 2026-27: 1 classified AI system TBD GPUs

ERDC DSRC

Floor space – ~22,000 ft2 Power – ~18,000 kW Cooling – ~5,300 Tons Focus areas Unclassified HPC Classified HPC Data Analysis Support & Visualization (DAV-C) Benchmarking Hosting of both HPCMP & special-purpose customer systems 1 classified HPC system ~80,000 cores & 16 GPUs “Wheat” (Liqid Composable) unclassified 85,440 cores (Intel 9242) & 528 GPUs (NVIDIA A100) “Carpenter” (HPE EX4000) unclassified 315,136 cores (AMD 9654) & 8 GPUs (NVIDIA A40)

“Barfoot” (HPE EX4000) unclassified 212,736 cores (AMD 9654) & 12 GPUs (NVIDIA L40)

Coming 2027: 1 classified HPC system TBD cores & TBD GPUs

Navy DSRC Floor space – ~13,000 ft2 Power – ~8,400 kW Cooling – ~4,300 Tons Focus areas Unclassified HPC Classified HPC Highly available HPC Large-scale, long-term data storage management “Narwhal” (HPE/Cray EX) unclassified 308,480 cores (AMD 7H12) & 112 GPUs (NVIDIA V100) 1 classified HPC system ~34,000 cores & 8 GPUs “Nautilus” (Penguin TrueHPC) unclassified 186,368 cores (AMD 7713) & 144 GPUs (NVIDIA A40, A100)

“Blueback” (HPE/Cray EX4000) unclassified 256,512 cores (AMD 9654) & 152 GPUs (AMD MI300A, NVIDIA L40)

Coming 2026-27: 1 unclassified & 1 classified HPC systems ~170,000 cores & ~190 GPUs total

MHPCC DSRC

Floor space – ~2,400 ft2 Power – ~1,200 kW Cooling – ~240 Tons Focus areas Emerging Technology (Exploratory HPC systems & services) HPC Portals Customer-reimbursed HPC Non-allocated/specialized HPC systems “Reef” (Aspen Systems) unclassified 400 cores (Intel 6248) & 22 GPUs (NVIDIA V100, RTX 8000) “Coral” unclassified Hybrid cluster w/ x86 & ARM nodes & various GPUs “Crux” (Dell) unclassified 40,320 cores (AMD 9634) & 40 GPUs (NVIDIA A16, H100) 2 classified systems ~3,100 cores & 60 GPUs

“Makau” (HPE XD) unclassified 3,840 cores (AMD 9534) & 78 GPUs (NVIDIA L40, H100) MHPCC Operations not included in HITS-U requirements; enterprise hardware & software support may include MHPCC

HPC Technology Insertion (TI) Investments Annual Procurement of Major Systems Rotates among DSRCs; usually ARL, ERDC, & MHPCC in even years, AFRL & Navy in odd years Requirements generation, acquisition strategy development, & overall process execution by government (not HITS-U) Delivery, installation, & acceptance testing by TI OEMs (penalties associated w/ late acceptance, poor performance) Integration w/ site infrastructure (power distribution, cooling distribution, local networks) by HITS-U & others Sustainment Operations System administration & maintenance (hardware & software break/fix not covered by warranties) usually executed on contract options w/ TI OEMs or through third-parties (penalties associated w/ poor performance) Security requirements, management of third-party application software, tuning of job scheduling/orchestration software, integration w/ workload characterization capability, & user account management accomplished by TI OEMs, government staff, HITS-U, & other contractors (significant interdependencies & interrelationships) Expected Service Life Each system targeted for 5-7 years of continuous service under sustainment operations Systems “decommissioned” at EOL & made available for reutilization or removed for destruction

HPC Infrastructure Elements (Historical Reference) HPC Portal Web browser-based user interface to access HPC systems & run virtual/web applications Unclassified & classified instances maintained on appliances at each DSRC Center-Wide File System (CWFS) “Near-online” 120-day data storage (~10s PB) Within each DSRC between HPC online file systems (30-day) & tape archive storage (indefinite) Mass Storage Archival System (MSAS) “Offline” data storage to tape media (~100s PB) Upgrade planned for FY26 Local Network Enclaves Unclassified (DREN), classified (SDREN & others), isolated/closed (environmental monitoring) User workstations, local servers, enterprise servers Replaced w/ Open OnDemand Decommissioned Early 2026; replacing w/ enterprise data fabric

Cross-Centers/Enterprise Functions Enterprise Operations Services Outward: Help Desk Ticketing System, User Authentication Services, Web Services, HPC System Workload Characterization, Advanced Reservation Services, Commercial S/W License Manager, GitLab (general use), Advanced Visualization Services (DAV-C) Internal: Centralized User Account Management System, Operational S/W Support (incl GitLab), Database Support, Cybersecurity Support (Lead ISSO), HPC System Benchmarking, Studies, Analyses & Tech Projects Some enterprise services managed & operated within DSRCs (Help Desk, Authentication Services, Benchmarking, DAV-C) User Experience Services How-to Documents & Videos, HPC User Guides, HPC Application Support “Back Office” Support Services Procurements (hardware, software, maintenance agreements), Project Management (studies, analyses, major projects), Financial Management, Contract Performance Management

Summary of HITS-U Roles (1/2) DSRC Operations Support for 24/7/365 operations of AFRL, ARL, ERDC, & Navy DSRCs Support for: facilities (system integration w/ power & cooling), cybersecurity, networking, system administration, database development & administration, software development & management, maintenance agreements, configuration management, procurements Delivery of HPC Support for integration of HPC systems w/ infrastructure Management of certain features & various applications on HPC systems (may include system administration & maintenance) Support at multiple classification levels User Assistance Help Desk; basic enterprise, advanced assistance, local DSRC support (some classified) Advanced visualization application development & services Secure authentication services development/sustainment, HPC application support, outreach

No HITS-U direct support to MHPCC DSRC in these three areas; however, MHPCC may benefit from some of them

Summary of HITS-U Roles (2/2) Enterprise/Cross-Centers Contract management, project support Support for enterprise software (outward-facing & internal) development & sustainment Support for enterprise services (account management, software license management, repositories, databases) Support for enterprise procurements (hardware, software, maintenance agreements) While MHPCC DSRC is outside the direct scope of the HITS-U contract, it may be included as an authorized beneficiary under enterprise-wide software, services, & procurement actions

AFRL DSRC

ARL DSRC

Navy DSRC

ERDC DSRC

MHPCC DSRC

Closing Remarks HITS-U is critical to HPCMP mission Successful performance on HITS-U = Successful HPC Centers HITS-U budget significant driver of total program budget execution HITS-U provides opportunities for enterprise focus Reduce stove-piped approaches across DSRCs, improve information sharing & lessons learned, close communication gaps Overall goal is to increase effectiveness, exploit efficiencies, leverage innovations HITS-U includes scope & ceiling for base & option periods Base PWS & DSRC Annexes intended to capture broad range of requirements w/ “such as, but not limited to…” Individual Task Orders will contain more specific requirements

HITS-U required to support very complex environment ranging from routine operations of 4 DSRCs to highly specialized HPC capabilities in multiple classification domains https://centers.hpc.mil image1.gif image3.png image11.jpeg image12.jpeg image13.jpeg image14.jpeg image15.jpeg image4.jpeg image5.jpeg image6.jpeg image7.png image8.png image9.png image10.png image24.png image25.png image16.jpeg image17.png image18.jpeg image19.jpeg image20.jpeg image21.png image22.png image23.png image26.jpeg image27.jpeg image28.jpeg image29.jpeg image30.jpeg image31.jpeg image32.jpeg image33.jpeg image34.jpeg image35.png image36.jpeg image37.jpeg image38.jpeg image39.jpeg image40.jpeg image41.jpeg image42.jpeg image2.png