Campus-Wide AI Centers of Excellence (CoE)
Transform your university into a premier AI research destination. We architect, deploy, and manage massive, multi-tenant AI supercomputing clusters that serve as a sovereign cloud for your entire campus, powering diverse faculties and corporate incubation projects simultaneously.
Software Ecosystem & Turnkey Installation Support
Every CoE deployment ships with enterprise-grade cluster orchestration, identity management, and high-speed parallel file systems, fully configured and validated by our engineers.
We Validate Your Sovereign Cloud
Skip the cluster orchestration nightmares. Kirti Telnet delivers fully integrated multi-rack deployments featuring Slurm job scheduling, LDAP/Active Directory user authentication, and high-speed parallel file systems, backed by complete IT Facility Management Services (FMS).
Cluster Orchestration & Queuing
Fair-Share Multi-Tenant Compute
Best for: Fair-Share Multi-Tenant Compute
Kubernetes — Container orchestration platform for scalable, distributed microservices and AI inference workloads
NVIDIA Base Command — Enterprise AI management platform for orchestrating multi-node GPU cluster workloads
Identity & Security
Campus-Wide User Authentication
Best for: Campus-Wide User Authentication
Active Directory — Enterprise identity platform for Windows-integrated campus authentication & group policies
Keycloak — Open-source IAM with SSO, federated identity, and fine-grained role-based access control
Parallel File Systems
Eliminating I/O Bottlenecks
Best for: Eliminating I/O Bottlenecks
VAST Data — All-flash disaggregated shared storage with universal namespace for mixed AI & analytics
Lustre — Open-source parallel file system powering world’s largest supercomputing clusters
Ceph — Software-defined distributed storage with block, object, and file interfaces at scale
Supported Operating Systems
Secure Enterprise Datacenters
Best for: Secure Enterprise Datacenters
Red Hat Enterprise Linux (RHEL) — Enterprise-grade OS with long-term support & certified driver stacks
Cluster Architecture Matrix — Campus-Wide AI CoE
A Center of Excellence is not a single server; it is an orchestrated symphony of specialized nodes. Our CoE architectures integrate lightweight login nodes for user access, massive multi-GPU worker nodes for heavy execution, and ultra-fast parallel storage fabrics to feed the clusters.
All specifications are representative and subject to change based on campus requirements and component availability.
User Access & Job Submission
User SSH access, script compiling, and Slurm job submission gateway for all campus researchers and students.
RAM: 256 GB
Storage: Redundant Boot Drives
Campus Core
Standard Compute Tier
Serves the bulk of undergraduate and master’s level JupyterHub workloads with quad professional GPUs.
CPU: Dual CPUs
GPU: 4x NVIDIA L40S / RTX 6000 Ada
RAM: 1 TB
RoCEv2 Fabric
Heavy Deep Learning Tier
Dedicated to massive Ph.D. foundation model training and faculty research with 8-GPU HGX baseboard architecture.
GPU: NVIDIA HGX H100 / H200 (8-GPU)
RAM: 2 TB
NVIDIA Quantum-2
Parallel Storage Layer
High-speed centralized dataset repository accessible by all compute nodes with petabyte-scale capacity.
Software: WEKA / Ceph
Scale: 100TB to Petabyte
Storage Fabric
Management & Monitoring
Runs LDAP, monitoring dashboards (Grafana/Prometheus), and cluster telemetry on a dedicated out-of-band network.
CPU: Single CPU
RAM: 128 GB
Storage: High-Endurance SSDs
Out-of-Band
Architectural Features & Performance
Multi-Tenant Slurm Job Queuing
In a campus-wide CoE, hundreds of students and faculty need GPU access simultaneously. We deploy Slurm Workload Manager to create fair-share priority queues. A bioinformatics student can request 2 GPUs for 5 hours, while an engineering professor reserves an entire 8-GPU HGX node for a 14-day run—all orchestrated automatically without manual IT intervention.
The Sovereign Data Imperative
Public clouds charge exorbitant egress fees and pose severe privacy risks for proprietary research. A university-owned CoE functions as a Sovereign AI Cloud. All medical imaging, demographic datasets, and custom IP remain physically secured within the university datacenter.
Unbottlenecked Parallel Storage
When 20 different compute nodes request access to the same massive ImageNet dataset simultaneously, standard NAS systems crash. We deploy enterprise-grade parallel file systems (like WEKA or Lustre) over 400Gbps network fabrics, ensuring every GPU is fed data instantaneously, maximizing your hardware ROI.
Engineering Disclaimer: Configuration specifications are representative and subject to change based on availability. GPU performance in research workloads depends on software optimization, dataset size, and concurrent user load.
Frequently Asked Questions
How do we bill different departments for using the Center of Excellence?
Our cluster management orchestration (via Slurm and specialized billing modules) meticulously tracks exactly how many GPU-hours and CPU-cores each user, department, or external corporate tenant consumes. This allows the university to implement accurate chargeback models to recoup infrastructure costs.
Does Kirti Telnet provide physical datacenter preparation for a CoE?
Yes. An AI Center of Excellence requires robust MEP (Mechanical, Electrical, and Plumbing) engineering. We consult on and provide high-density smart racks, 3-phase high-amperage PDUs, hot-aisle containment, and In-Row cooling systems necessary to support multi-kilowatt supernodes.
How is the system maintained after deployment?
We offer comprehensive Annual Maintenance Contracts (AMC) and provide dedicated IT Facility Management Services (IT-FMS). Our engineers can actively monitor your cluster’s telemetry, manage software patches, and replace hardware to ensure 99.9% academic uptime.
Can we start small and scale the CoE over several years?
Absolutely. We design the initial network fabric (InfiniBand/Ethernet switches) and storage arrays with massive headroom. You can launch your CoE with just three or four worker nodes and seamlessly add 8-GPU HGX supernodes in future academic years as grant funding is released.
Consult a Center of Excellence Architect
Talk to KTPL’s infrastructure engineers for a sovereign cloud assessment and multi-rack supercomputing architecture for your campus-wide AI Center of Excellence.
Get in Touch Today for Expert Solutions and Support