The platform
Dedicated GPU capacity, allocated to one customer at a time
ADC is not a public cloud. Each customer's GPU nodes and storage are theirs for the term of the agreement, which is what makes the residency and isolation claims checkable rather than contractual.
| GPU | Form | Interconnect | Typical use |
|---|---|---|---|
| NVIDIA H100 80 GB | 8-GPU HGX nodes | NVLink + 400 Gb/s InfiniBand | Large-model training, fine-tuning |
| NVIDIA H200 141 GB | 8-GPU HGX nodes | NVLink + 400 Gb/s InfiniBand | Long-context training, memory-bound inference |
| NVIDIA L40S 48 GB | 4- and 8-GPU nodes | PCIe Gen5 | Inference, vision, rendering, mixed workloads |
indicative Indicative configurations, subject to availability; specifications and availability are confirmed in writing.
Access
How your team gets in
Access is arranged the way your organisation requires: private connectivity to your network, bastion hosts with your identity provider, or on-site. Never a shared login.
- 1
Tell us what you need
What you are training or serving, how many GPUs, for how long, and the data's classification.
- 2
We scope it with you
A person replies within two business days with what is available and a written quote.
- 3
Onboard
Agreement signed, environment provisioned, access arranged your organisation's way.
- 4
Run
Capacity is yours for the term. Support comes from ADC's team in Canberra.
Scope a configuration with us
Tell us the workload and the data's classification; we come back with what is available and a written quote.