NovaOtterBack to Solutions

Recommended Starting Point

CPU1x AMD EPYC 9004/9005, up to 192 cores
GPUUp to 4x RTX PRO 6000, A100, H100, or H200
MemoryUp to 2 TB
CoolingAir-cooled CPU and GPU

Configuration Choices

Processor

Select CPU capacity around GPU count, memory bandwidth, and scheduler needs.

Graphics Card

Choose accelerator count and memory class around training, inference, or simulation requirements.

Memory

Size ECC memory relative to GPU VRAM and host-side preprocessing.

Operating System

Choose the deployment image for AI frameworks, orchestration, or your own imaging process.

Primary SSD

Use fast local storage for the OS, models, checkpoints, and active datasets.

Additional SSD

Add a secondary storage tier when local datasets exceed the primary drive.

RAID Controller

Use a RAID controller only when redundancy or specific storage topology requires it.

Network

Plan fabric speed around shared storage, cluster scheduling, and east-west traffic.

Trusted Platform Module

Add hardware-backed platform security if your deployment policy requires it.

UPS System

Add rack power protection when the server will run outside an existing UPS-backed rack.

Warranty/Support

Choose the support window for the server lifecycle.

Why this configuration path

Single-socket AMD EPYC server for four-GPU workloads.
Balances rack density, CPU lanes, and GPU capacity.
Works as a training, inference, or simulation node.