Supported configurations - Poolside
Supported environments
| Deployment | Description |
|---|---|
| Amazon EKS 1.29+ | Amazon Elastic Kubernetes Service, with IAM Roles for Service Accounts (IRSA) and an Application Load Balancer |
| OpenShift 4.16+ | Red Hat OpenShift environments |
| Upstream Kubernetes 1.29+ | Self-managed Kubernetes environments such as RKE2 or Charmed Kubernetes |
| On-premises | Single-node Kubernetes (RKE2) on customer-provided or Poolside-provided hardware |
Inference requirements
Poolside models have different minimum requirements. Use this table to size your inference nodes. For concurrent-agent capacity and developer-seat estimates, contact your Poolside account team.
| Model | Quantization | Minimum GPU memory | Minimum CPU | Minimum host memory |
|---|---|---|---|---|
| Laguna M.1 | FP8 | 384 GB | 128 cores | 1 TB |
| Laguna XS.2 | FP8 | 96 GB | 44 cores | 512 GB |
For local development on your own hardware, see How to run a Poolside model locally. For cloud deployments, storage requirements depend on whether S3-compatible storage is colocated on the inference node. For on-premises deployments, see Storage requirements for guidance on sizing and configuring storage.
GPU memory reference
Poolside model inference targets the following NVIDIA GPU families: RTX 6000 Blackwell, H100, and H200. The memory per GPU is listed here for reference against the per-model minimum GPU memory table above:
| GPU | Memory per GPU |
|---|---|
| H100 | 80 GB |
| RTX 6000 Blackwell | 96 GB |
| H200 | 141 GB |
Which GPUs suit a given model is model-dependent. The per-model minimum GPU memory table is a floor, not a GPU selector: context length, batch size, and the number of GPUs all affect what actually serves a model. Confirm the right combination of model, GPU type, and GPU count for your workload with your Poolside account team.
Support and compatibility
For questions about integration with specific enterprise tooling or deployment workflows, contact Poolside.