Costs
For the minimal, starter configuration for GCP, the estimated monthly costs are $544.18. This is for on-demand resources in GCP's Oregon, us-west1 region as of August, 2024. To ensure costs are kept low when users get started, it does not include any GPU based node pools. Customers can add GPU based node pools to the existing cluster as and when required.
Starter Configuration¶
The table below describes the cost breakdown by the various infrastructure components.
| Component | Infra Costs ($/month) |
|---|---|
| GKE Kubernetes Cluster | $479.15 |
| Memory Store for Redis | $35.77 |
| Cloud SQL Database | $9.36 |
| Cloud Storage | $19.90 |
| TOTAL | $544.18 |
Important
Please contact your assigned Rafay customer success person if you need assistance with customizing the infrastructure configuration for your specific requirements.
Users can use this public GCP cost estimator to customize the infrastructure for their requirements.
GPU Node Pools¶
To support the needs of their data scientists and other users, administrators can scale the GKE cluster by adding GPU enabled node pools. Common use cases are deep-learning tasks such as image recognition, natural language processing and compute-intensive tasks such as video transcoding and image processing.
Select the most appropriate machine series if you require specific models of GPUs. For example,
- A3 machine series: H100 GPUs
- A2 machine series: A100 GPUs
- G2 machine series: L4 GPUs
- N1 machine series: All GPUs except A100 and L4
Learn more about GPU pricing on GCP. For example, if you add a GPU node pool with 4 Nvidia T4 GPUs (16GB memory/GPU), you will need to budget for 0.35 per hour per GPU*. This will work out to **1,022 per month* using on-demand pricing approach (i.e. 2x the costs for the minimal configuration)
Important
GPUs are available only in specific GCP regions. Also, please ensure that you have sufficient GPU quota in your selected zone before you create GPU nodes.