# Hardware requirements

> Plan the physical infrastructure for a baseline production on-premises deployment

Use this page to estimate the physical hardware for a baseline production on-premises deployment. Actual sizing depends on user volume, workload concurrency, transformation complexity, storage footprint, and your availability requirements.

> **Note:**
>
> This page applies to customers who supply their own physical or datacenter hardware. If you deploy to an existing managed Kubernetes platform, use the Kubernetes resource profiles on [Overview of the infrastructure](./infrastructure.md) to determine sizing. The profiles (small, medium, large) specify Kubernetes resource allocations rather than physical hardware. Use the medium profile for production by default.

## Typical infrastructure profile

A baseline production deployment requires approximately the following capacity:

* 88 vCPUs
* 400 GB RAM
* Supporting database, storage, observability, and internal platform services

## Recommended hardware for production

For production, high availability, and spare capacity, provision the following hardware:

| Item                         | Description                                                         |
| ---------------------------- | ------------------------------------------------------------------- |
| Compute servers              | Three enterprise rack servers, each with 24–32 cores and 256 GB RAM |
| Storage                      | 8–12 TB enterprise NVMe or all-flash usable capacity                |
| Networking                   | 10/25 GbE switching, firewall, and cabling                          |
| Backup and disaster recovery | Backup appliance or backup integration                              |
| UPS and power protection     | Rack UPS and related power equipment                                |
| Optional GPU                 | NVIDIA L4 or equivalent,, where transformation workload requires it |

## Sizing variables

Adjust the hardware profile based on the following factors:

* User volume: the number of concurrent users who access Unity Asset Manager
* Workload concurrency: the number of simultaneous transformations and operations
* Transformation licensing: the number of concurrent transformations your Unity Asset Transformer SDK license permits; licensing can limit throughput independently of GPU and CPU capacity
* Transformation complexity: the size and complexity of the 3D assets you transform
* Storage footprint: the total volume of assets you store on the platform
* Availability requirements: your high availability targets and redundancy needs