Hall layoutSingle block
Training
Large clusters that run for weeks.
- Deployment
- A dedicated hall or a full site, beside generation
- Cooling
- Liquid cooling at the highest density
Specification
The spec at a glance.
- Design standard
- Tier III, N+1
- Rack density
- High density, for the latest GPUs
- Cooling
- Direct-to-chip or immersion, closed loop
- Network
- Dark fibre to managed IP
- Time to power
- From 6 months, site dependent
Built to your requirements
Sized to the workload.
Tell us the platform and the date. We match the site and size the hall.
Hall layoutSeparate pods
Fine-tuning and research
Smaller clusters that change often.
- Deployment
- Pods within a shared hall
- Cooling
- Liquid cooling at high density
Hall layoutRoom to expand
Inference
Serving models at volume.
- Deployment
- Pods on dark fibre to the major networks, for high-volume and batch serving
- Cooling
- Liquid cooling sized to the pod
Modular delivery
Built in a factory. Installed on site.
Compute, cooling and power arrive as containerised pods, built and tested by suppliers such as Schneider Electric and Vertiv. Less work on site means AI capacity runs sooner, and more pods are added as demand grows.
- 01
Built and tested off site
Racks, cabling, piping, power and liquid cooling are fitted and tested before shipping, which cuts onsite work and delays.
- 02
Site work in parallel
Groundworks and the connection are prepared while the pods are built, so neither waits on the other.
- 03
Staged to demand
Capacity is staged to your date. Hardware goes in last, so a late shipment never holds up the build.
How we deliver
Designed to NVIDIA reference architectures.
- Reference designs
- NVIDIA GB300 NVL72 and HGX B300.
- Partners
- Power and cooling from Schneider Electric, Vertiv and Delta. Servers from Supermicro and HPE.
- AI operations
- Led by engineers who have rolled out GPU services in 70 countries with AWS, Google, Akamai and Fastly.
- Since 2021
- We have run liquid-cooled compute since 2021. It runs today at our R&D centre in Georgetown, Tasmania.
Service scope
Choose where our scope ends.
Every level includes the grid connection, contracted power, land, fibre access and room to expand. Choose how much of the rest DAME runs.
| Powered land | Colocation | Managed | |
|---|---|---|---|
| Building, critical power and liquid cooling | Customer | Included | Included |
| Remote hands, with a service level and credits | Optional | Included | Included |
| Ports, cross-connects, internet and dark fibre | Optional | Optional | Included |
| Monitoring portal and service level reporting | Customer | Included | Included |
| Servers, network fabric and storage, run to the OS | Customer | Customer | Included |
| GPU clusters and orchestration | Customer | Customer | Optional |
"Customer" means the customer provides it. GPU and server finance is available through our partners. Scope is confirmed per site and contract.
Where we build
Operating in Tasmania. Building on the mainland.
2
Operational sites
535MW
In development, five sites
Locations on the map are approximate.
Operational
- Georgetown R&D centreTasmaniaOperational
- Georgetown solarTasmaniaOperational
In development
- SA 1South Australia10 MW
- SA 2South Australia50 MW
- VIC 1Victoria10 MW
- VIC 2Victoria300 MW
- QLD 1Queensland165 MW
How many megawatts,and when?
Tell us the capacity and the timeline. DAME will identify the sites that can supply it.
Ask about capacity