Question 1

What is GPU compute?

Accepted Answer

GPU compute refers to infrastructure built around graphics processing units, which are well suited to the parallel mathematics behind AI training, fine-tuning and inference. Peregrine provides GPU compute as a service and can also design and commission GPU infrastructure that your organisation owns.

Question 2

What types of AI workloads can run on Peregrine Compute?

Accepted Answer

LLM training and fine-tuning, inference and model serving, generative AI, AI agents, computer vision, robotics and simulation, scientific computing, rendering and GPU-accelerated analytics. Capacity and architecture are matched to the specific workload.

Question 3

Where is Peregrine Compute infrastructure located?

Accepted Answer

Peregrine offers Australian deployment options. Specific locations are matched to project requirements — including data residency, latency and power considerations — and are confirmed during scoping. Architecture and availability are subject to deployment and capacity availability.

Question 4

What is a Compute Blueprint?

Accepted Answer

A fixed-fee assessment that analyses your workloads, utilisation forecasts and constraints to recommend an AI compute architecture, compare buy vs rent vs cloud, and estimate infrastructure economics. Typical engagement: approximately 2 weeks.

Question 5

Do I have to use all four services?

Accepted Answer

No. Assess, Build, Operate and Capacity each stand on their own. Some customers only need GPU capacity; others only want Peregrine to operate infrastructure they already own. Use one service or all four.

Question 6

What GPU architectures are supported?

Accepted Answer

Capacity is built around NVIDIA Blackwell (including B200, B300 and RTX PRO 6000 Blackwell class systems) and NVIDIA Hopper (H100, H200). GPU architecture and availability depend on deployment, capacity and project requirements.

Question 7

Can I reserve dedicated GPUs?

Accepted Answer

Yes. Reserved and dedicated capacity models allocate GPU infrastructure to your organisation over an agreed term, subject to availability. Registering a GPU forecast helps us plan capacity around your requirements.

Question 8

What is the difference between dedicated and on-demand GPU infrastructure?

Accepted Answer

On-demand provides flexible access to shared infrastructure for the duration of a job, subject to availability, and is typically billed per GPU-hour. Dedicated infrastructure allocates single-tenant GPU servers to your organisation for an agreed term, offering isolation and capacity certainty. For sustained workloads, dedicated or reserved infrastructure can provide more predictable economics than purely consumption-based compute.

Question 9

Is registering a GPU forecast a commitment?

Accepted Answer

No. Registering a forecast does not create a purchase obligation and does not constitute a capacity reservation. Peregrine uses forecasts to plan future Australian GPU capacity, and early capacity partners receive priority consideration when new capacity is commissioned.

Question 10

Can Peregrine complement our existing AWS, Azure or university HPC environment?

Accepted Answer

Yes. Peregrine is designed to work alongside the environments you already use. The Managed AI Platform can operate across Peregrine infrastructure, customer-owned hardware and public cloud, and GPU capacity can supplement existing clusters during peaks.

Question 11

What is Private AI Cloud?

Accepted Answer

A dedicated AI environment designed around your organisation — dedicated capacity, data control, a predictable operating model and flexible deployment across Australian data-centre, private or hybrid locations — with cloud-style access for your teams.

Question 12

Can data remain in Australia?

Accepted Answer

Yes. Peregrine's GPU capacity is hosted in Australian data centres, and infrastructure can be designed to support customer security, governance and compliance requirements, including Australian data residency where required.

Question 13

Who operates a Private AI Cloud?

Accepted Answer

You choose: Peregrine Managed, Shared Responsibility or Customer Operated. Peregrine can run the full platform layer as a monthly managed service, share responsibilities with your team, or hand over a documented environment.

Question 14

Can universities use Peregrine for research projects?

Accepted Answer

Yes. Peregrine can supply project or reserved GPU capacity, or design and operate a university-owned cluster. Allocation can follow grants and projects, with quotas, multi-tenancy and the Slurm or Kubernetes interfaces researchers already use.

Question 15

How does Peregrine fit with institutional HPC?

Accepted Answer

As a complement. Peregrine can add GPU capacity, take on platform operations, or design new GPU infrastructure alongside existing HPC and cloud research environments.

Question 16

What is Peregrine ComputeGrid?

Accepted Answer

ComputeGrid is being developed as a distributed AI infrastructure platform connecting GPU capacity across data centres, commercial sites and renewable-energy-enabled locations. It is being progressively deployed as capacity is commissioned.

Question 17

Does ComputeGrid replace data centres?

Accepted Answer

No. Distributed nodes do not replace hyperscale data centres. Centralised facilities suit tightly coupled training; distributed locations are useful for inference, fine-tuning, batch processing and capacity expansion.

Frequently asked questions.

General

Capacity

Private AI Cloud

Research

ComputeGrid

Still have questions?