Hewlett Packard Enterprise Data Science Institute

The High-Performance Computing (HPC) clusters at the University of Houston (UH) are managed by the Hewlett Packard Enterprise Data Science Institute (HPE DSI). The primary mission of HPE DSI-managed HPC clusters is to support intensive scientific computation and complex data modeling that exceed standard desktop hardware or commercial cloud solutions. GPU-accelerated nodes are specifically optimized for computationally intensive research workloads to maximize performance for high-impact projects.

Users of enterprise Generative AI tools must adhere to the University of Houston Generative AI Guidance https://www.uh.edu/ai/ai-tools/ and all applicable UHS and University policies, which this policy supplements.

Resource allocations prioritize workloads in the following order to manage compute cycles, storage, and GPU hardware: 

• Funded Research: Projects supported by active external or internal grants (e.g., CPRIT, NSF, NIH, DoD, DOE). 
• Fund-Seeking Research: Pilot studies, benchmark generation, or proof-of-concept computations conducted to generate preliminary data for grant applications. These workloads require explicit documentation detailing the targeted grant, sponsoring agency, and expected submission timeline.

HPE DSI HPC resources are prohibited for administrative, operational, or general productivity tasks. Deploying GPU nodes as personal desktop assistants or persistent lab utility servers severely degrades cluster capacity.

Workloads spanning multiple categories should be reviewed with HPE DSI staff to identify the most suitable computing platform.  

Exception: Researchers may use HPC resources for short-duration porting, testing, profiling, and debugging of mature software, where the development scope is limited to adapting it to HPC-specific hardware—such as high-speed interconnects or multi-GPU configurations—and cannot be performed on a single workstation. This includes MPI applications, GPU/CUDA kernels, distributed training, and parallel scientific codes. Such work should run as appropriately sized, short batch or interactive jobs and must not unnecessarily occupy production resources. 

• Scientific AI Research & Model Development: HPE DSI HPC clusters support neural network architecture development, domain-specific dataset training, scientific foundation model fine-tuning, and novel AI/ML research.  
• Large-Scale AI Model Training: Multi-node distributed training or fine-tuning of massive foundation models exceeds HPE DSI capacity and must be directed to national-scale resources (e.g., NAIRR, ACCESS-CI, DOE Leadership Computing Facilities,) via allocation requests. 
• Research Productivity & Interactive AI: Literature reviews, document analysis, research planning, drafting, and general Q&A must use approved enterprise AI tools (e.g., Copilot, Gemini).
• Persistent LLM Services: HPE DSI HPC resources may not host pre-trained, off-the-shelf foundation models as persistent interactive services or laboratory chatbots. These workloads belong on departmental, cloud, or workstation infrastructure.  
• Prohibition of Idle Sessions: Long-running idle interactive GPU sessions and unattended inference services are prohibited and subject to termination.  
• Large Model Deployment Justification: Multi-GPU or large-model deployments require prior consultation with HPE DSI staff and formal submission of:  
  • Scientific Objective: Justification for why enterprise tools or national-scale resources are insufficient and/or inappropriate.  
  • Model Architecture & Footprint: Parameter count, GPU memory requirements, and batch execution strategy.  
  • Funding Linkage: Active grant award or target proposal relying on the execution.  

HPE DSI monitors cluster utilization—including active processes, GPU sessions, node usage, and persistent services—to ensure system reliability, efficient resource utilization, and fair access for all users.

HPE DSI recognizes that inefficient resource utilization may sometimes result from a lack of familiarity with HPC systems, job schedulers, GPU resources, profiling tools, or appropriate scaling practices. Our approach is therefore intended to be both corrective and educational. Researchers are encouraged to take advantage of HPE DSI training and consultation resources to improve their ability to use UH HPC resources effectively.  

HPE DSI provides HPC training at multiple levels, free of charge to all UHS affiliates, including:

UH HPC Clusters Introduction – a short introductory workshop providing an overview of UH HPC clusters and how to get started using them.

HPE DSI 101 – Introduction to Cluster Computing – a more comprehensive certification course covering Linux, shell scripting, queuing systems, and cluster architecture.

Micro-Credential in High-Performance Computing – a structured learning pathway designed to develop broader HPC skills and improve proficiency in the effective use of HPC resources.

These resources are intended to help researchers become more proficient HPC users, including in areas such as job scheduling, resource selection, GPU utilization and profiling, workload scaling, and efficient use of CPU and GPU resources. HPE DSI may also recommend appropriate training, workshops, or one-on-one consultation when resource-usage issues are identified.

Escalation Process for Resource Misuse

Inappropriate or inefficient resource usage is managed through a four-step progressive escalation process:

User Notification: Direct notification to the user and associated Principal Investigator (PI) regarding observed usage issues, with recommendations for corrective actions, training, or other alignment steps.

Consultation & Training: Discussion with the researcher and/or PI to better understand workload requirements and identify appropriate configurations, resource allocations, or alternative platforms. Where appropriate, HPE DSI may recommend relevant training or provide guidance on job scheduling, GPU utilization, profiling, scaling, and other HPC practices.

Temporary Restrictions: Job/session cancellations or temporary resource limits for uncorrected issues (e.g., prolonged idle GPU sessions, unauthorized daemons).

Suspension: Suspension of access to clusters for users, laboratories, or PIs following repeated non-compliance or severe violations (e.g., deliberate policy circumvention, unauthorized personal projects, or misrepresenting administrative tasks as research).

HPE DSI reserves the right to take immediate action—such as terminating jobs or restricting access—without preceding escalation steps if an activity presents an immediate risk to cluster stability, security, or operations.

The goal of this process is not simply to enforce compliance, but to enable researchers to become effective and responsible users of HPC resources while maintaining a reliable and equitable computing environment for the UH research community.