Warning
INTERNAL USE ONLY — NOT FOR PUBLIC DISTRIBUTION This document has not been approved for public release. It must not be committed to a public repository or shared externally until this notice is explicitly removed by the document owner prior to merge.
This document provides hardware specifications, supported OS images, onboarding verification, sample benchmarks, and best-practices for OCI deployments using the <GPU_FAMILY> GPU shape.
- Shape:
<shape name> - GPU configuration:
<gpu model and count> - Recommended OS baseline:
<minimum supported OS> - Recommended software baseline:
<primary driver/CUDA or ROCm/NCCL or RCCL> - Primary verification command:
<hello world or device visibility command> - Operational profile:
<single-node, scale-out, topology-sensitive, etc.>
Use this shape when you need <brief workload fit>.
It is best suited for <single-node vs scale-out guidance> and is a strong fit when <topology, memory, or networking differentiator>.
- Launch the shape with one of the approved images from the Provided Images table.
- Run the primary verification command and confirm all GPUs are visible.
- Run the hello-world or container smoke test from Hello World Verification.
- Confirm the expected RDMA or interconnect interfaces are up before moving on to benchmarks.
- Hardware Specifications
- When To Use This Shape
- First 15 Minutes
- Recommended Operating Systems
- Performance Benchmarks
- OKE GPU Getting Started
- Troubleshooting
- Further Reading & Support
| Shape Name | GPU Model | GPUs/Node | GPU Memory (GB/GPU) | GPU Memory Total | CPU | # of CPUs | System Memory | Local Storage | Host NIC | RDMA (ROCe) NICs |
|---|---|---|---|---|---|---|---|---|---|---|
See the OCI Compute Shapes Docs for up-to-date details.
To build your images using packer clone the OCI HPC Images repo and run the commands found there OCI HPC Images GitHub Repo.
| OS Version | Image Packer Build Details | OCI Platform Image Link | Driver Versions | Build & Dependency Status |
|---|
Here you can find suggested troubleshooting methods.