AI Computing Power Empowerment A Thai Research University Builds an Intelligent Cloud Foundation with ZStack

A public research university in Thailand had long relied on baremetal GPU servers to support its teaching and research. As courses related to artificial intelligence and 3D rendering became more widespread, computing demands surged and the traditional model began to show clear limitations: rigid resource allocation, complex operations and maintenance (O&M), insufficient data isolation, and underutilized hardware investments.

To address these challenges, the university adopted the ZStack Cloud platform to enable flexible scheduling of GPU resources in both GPU passthrough and vGPU modes. This significantly improved resource utilization and reduced total cost of ownership (TCO). At the same time, finegrained tenant isolation enhanced data security, and O&M efficiency improved dramatically, effectively resolving the highperformance computing resource management challenges in teaching and research scenarios.

Key Challenges: Traditional GPU Management Under Pressure

As a leading university in Thailand, the institution regards information technology as a core pillar of its teaching and research. However, its existing baremetal GPU deployment model exposed multiple pain points as demand increased:

Rigid resource allocation:Under the baremetal model, GPU resources could not be flexibly scheduled among different user groups such as students, faculty, and research teams. During peak periods—for example, at the beginning of the semester when courses are concentrated or during critical phases of research projects—GPU resources were in severe shortage. In offpeak periods, however, many devices sat idle, wasting hardware capacity. This rigid supply model could not match highly variable demand.

Growing O&M workload:A large number of standalone physical devices required manual, onebyone maintenance. Configuration, monitoring, and troubleshooting were all handled manually, resulting in heavy workloads, low efficiency, and slow response to issues.

Data security and cost control pressures:There was no effective isolation between different users’ coursework and research data, leading to risks of interference and potential data leakage. At the same time, simply stacking more hardware to expand capacity drove up procurement costs without fully leveraging existing assets.

The university therefore needed a way to achieve elastic allocation, precise management, and secure isolation of GPU resources—without compromising performance—while also maximizing return on existing hardware investments. After extensive evaluation and rigorous testing of multiple solutions, the university selected the ZStack Cloud platform for its strengths in GPU resource management, heterogeneous hardware reuse, and multitenant isolation.

Solution: “One Cloud, Multiple Capabilities” for GPU Management

Based on the university’s actual needs and existing IT environment, ZStack worked closely with the institution to design a tailored cloud platform solution built around four pillars: reuse, adaptation, isolation, and operations. The core objective was to create a flexible, efficient cloud environment supporting both GPU passthrough and vGPU virtualization.

1.Hardware Reuse and Integration to Protect Existing Investments

ZStack Cloud supports the reuse of multibrand, multimodel, and multigeneration servers, fully compatible with the customer’s existing hardware. In this project, the university’s ASUS servers and NVIDIA L40S GPUs were consolidated under a single ZStack Cloud management plane. Leveraging ZStack Cloud’s heterogeneous hardware management capabilities, all resources were pooled and centrally managed.

This eliminated the need to retire legacy equipment or purchase large quantities of new servers, protecting prior capital investments. Through resource reuse and unified management, the university significantly reduced TCO and aligned its IT strategy with sustainability goals.

2.DualMode GPU Support for Diverse Scenarios

To meet the distinct requirements of teaching and research, ZStack Cloud provides dualmode GPU support—GPU passthrough and vGPU virtualization—enabling “one cloud, multiple capabilities”:

· GPU passthrough:
Delivers nearnative performance for highintensity workloads such as AI model training, complex 3D rendering, and scientific computing. This ensures that performancesensitive research projects can progress efficiently.

· vGPU virtualization:
Virtualizes a single physical GPU into multiple independent vGPU instances, enabling many students or standard users to share GPU resources concurrently. This model is ideal for classroom teaching, basic experiments, and largescale shared usage, significantly boosting overall resource utilization.

3.Tenant Isolation for Secure and Fair Resource Usage

With ZStack Cloud’s tenant management capabilities, the university can create independent resource spaces for different departments, research groups, classes, and even individual users. Each tenant is allocated dedicated CPU, memory, and GPU/vGPU resources based on actual needs.

Workloads and stored data are logically isolated, ensuring that research data remains secure and private. At the same time, resources can be distributed transparently and fairly, eliminating conflicts associated with adhoc, mixed usage of shared hardware.

4.Efficient O&M to Reduce Management Burden

ZStack Cloud offers powerful capabilities for batch deployment and lifecycle management of virtual machines, improving the efficiency of IT resource scheduling. The university’s IT team no longer needs to configure each physical server manually. Instead, all compute, storage, and network resources are centrally managed and monitored via a unified cloud management interface.

This dramatically reduces O&M complexity and daytoday workload, allowing the IT team to shift from repetitive manual tasks to highervalue technical support for teaching and research.

Customer Benefits: A Benchmark for IT Modernization in Higher Education

By deploying the ZStack Cloud platform, the Thai public research university successfully addressed multiple GPU resource management challenges and achieved a coordinated upgrade of its teaching, research, and IT infrastructure.

1.Significant Cost Optimization

Through hardware reuse and resource pooling, the university substantially reduced TCO. The savings can now be redirected to support teaching innovation and research, including the development of cuttingedge courses and funding for new projects.

2.Dramatically Improved Resource Utilization

GPU idle time has been effectively eliminated. Elastic allocation capabilities now match fluctuating demand across the academic calendar. Whether during peak periods of intensive usage or during more evenly distributed daily workloads, computing resources can be precisely matched to actual needs.

3.StepChange in O&M Efficiency

The unified management platform has replaced manual, devicebydevice operations. O&M complexity is greatly reduced, while response times and issue resolution efficiency are significantly improved, providing a more reliable and agile IT backbone for the university.

4.Enhanced Teaching and Research Experience

From AI model training to handson classroom exercises, diverse computing needs can now be met quickly and reliably. Faculty and students enjoy a stable, highperformance IT environment that supports improved teaching quality and accelerates the translation of research into tangible outcomes.