Compute profiles — CPU and GPU
A compute profile is the CPU, memory, and (optionally) GPU allocation attached to a notebook or workload. Profiles are chosen when you create a notebook and are enforced by the platform’s admission and scheduling layer.
CPU profiles
Section titled “CPU profiles”- CPU notebooks run under your project’s CPU and memory quota.
- Workloads that exceed your project quota are not admitted until capacity is available; they queue rather than fail.
GPU profiles
Section titled “GPU profiles”GPU profiles are implemented on the platform and are scheduled through fair-share queues so everyone gets a fair turn on accelerator capacity. Two rules apply:
- GPU requires an entitlement. A GPU entitlement is attached to your account only by Aidin (per your plan or an enterprise agreement). It is not granted automatically with signup or workspace creation, and there is no self-service path to obtain one today.
- GPU capacity is limited. Workloads requesting GPU may queue during peak demand. Stopped notebooks release their GPU reservation.
Choosing a profile
Section titled “Choosing a profile”| Need | Profile |
|---|---|
| Data exploration, light training, pipelines authoring | CPU |
| Fine-tuning or training with CUDA frameworks | GPU (entitlement required) |
| Serving or long-running jobs | See Model serving and Cloud resources |