Initiatives / Systems & Infrastructure
Compute Systems
Monash Automation's shared GPU compute platform: automated provisioning, standardised environments and secure remote access, giving researchers reliable, ready-to-use compute.
- Team
- Systems & Infrastructure
- Platform
- Proxmox, Terraform, Packer, NetBird
- Hardware
- NVIDIA RTX 4090 GPUs, rack-mounted
Shared GPU infrastructure
The project began with a collection of NVIDIA RTX 4090 GPUs housed in external enclosures. Rather than running them as separate machines, the hardware was consolidated into a rack-mounted compute environment managed through Proxmox, creating a shared pool of high-performance compute that can be centrally managed, maintained and allocated across the team.
Reproducible compute environments
On top of that sits MACS, Monash Automation's Compute System. MACS uses Terraform to define infrastructure as code so machines are created consistently instead of being configured by hand, and Packer builds standardised Windows 11 and Linux images with GPU acceleration and common tools preinstalled.
Secure remote access
Members connect to compute from outside the lab using NetBird, with access tied into the organisation's single sign-on. Researchers get ready-to-use GPU machines without being physically present, cutting setup time and exposing members to real infrastructure engineering.
