Launching in early 2026, in partnership with NVIDIA, Dell Technologies, and CDC Data Centres, MAVERIC will be housed at CDC’s Melbourne campus. It features one of the first Australian deployments of NVIDIA’s cutting-edge GB200 NVL72 AI supercomputing platform. This infrastructure integrates NVIDIA’s latest Grace CPU and Blackwell GPU technologies (detailed below), placing MAVERIC at the forefront of global AI capability.
Monash has committed to powering MAVERIC with a carbon-neutral strategy to support its sustainability goals. In addition, the system employs advanced liquid-cooling technology with a closed-loop design that is significantly more water-efficient than traditional data centre cooling. Unlike conventional air-cooled servers (which often rely on evaporative cooling), MAVERIC’s cooling does not consume water, eliminating the need to draw on local water supplies. This design choice not only reduces the system’s environmental footprint but also sets a new benchmark for sustainable high-performance computing. Conventional hyperscale data centres can consume millions of litres of water per day to keep servers cool. MAVERIC demonstrates that it is possible to achieve world-class AI performance without that cost to the environment.
NVIDIA Blackwell architecture and 5th Gen NVLink
The Dell PowerEdge XE9712 is a high-performance AI server platform designed for large-scale artificial intelligence workloads, particularly for training and real-time inferencing of large language models (LLMs). This platform features an industry-leading GPU density with up to 72 NVIDIA Blackwell GPUs and 36 NVIDIA Grace CPUs in a rack-scale design. The 72 GPUs are interconnected using NVIDIA NVLink to function as a single GPU.
The GB200 combines two Blackwell B200 GPUs and one NVIDIA Grace CPU into a single “superchip,” interconnected using fifth-generation NVLink technology for high-speed, unified memory access. Each B200 GPU features 16,896 CUDA cores, 528 Tensor Cores, and 192GB of ultra-fast HBM3e memory, providing a combined total of 384GB memory per superchip and memory bandwidth up to 8 TB/s per GPU.
