> ## Documentation Index
> Fetch the complete documentation index at: https://docs.beyondguard.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# BeyondGuard Infrastructure Sizing

> Reference CPU, memory, storage, and GPU sizing for an on-premise BeyondGuard deployment, including database and GPU-worker options.

This page gives a reference sizing profile for an on-premise BeyondGuard deployment. Actual numbers scale with your expected user load and file volume; treat these as a well-tested starting point and adjust with BeyondGuard during onboarding.

## Application servers

| Component | Instances | Resource profile |
| - | - | - |
| **Gateway** | 2 | 8 vCPU / 32 GB / 20 GB disk |
| **Engine** (guard services) | 4 | 16 vCPU / 128 GB / 100 GB disk (file-usage dependent) |
| **Router** (CPU) | 2 | 8 vCPU / 32 GB / 10 GB disk |
| **Total (application)** | **8** | **32 vCPU / 192 GB / 130 GB disk** |

## Databases

Each backing datastore runs in HA and can be a managed service or integrated into the deployment.

| Component | Availability | Resource profile |
| - | - | - |
| **PostgreSQL** | HA (managed or integrated) | 8 vCPU / 32 GB / 80 GB disk |
| **Redis** | HA (managed or integrated) | 8 vCPU / 32 GB / 80 GB disk |
| **Kafka** | HA (managed or integrated) | 8 vCPU / 32 GB / 80 GB disk |

## GPU worker

The security model runs on a dedicated GPU node, separate from the application workers. Choose a tier based on expected concurrency:

| Tier | GPU | Node profile | Capacity |
| - | - | - | - |
| **Minimum** | 2× 24 GB (48 GB vRAM) — e.g. 2× NVIDIA A10 or 2× RTX 4090 | 16 vCPU / 128 GB / 1 TB disk | \~20–30 concurrent requests |
| **High-capacity** | 1× H100-class (\~94–96 GB vRAM) or RTX A6000 class | 16 vCPU / 128 GB / 1 TB disk | \~100–150 concurrent requests |

GPU allocation scales with your intended application and expected throughput. The minimum tier deploys easily on commodity GPUs; scale up to H100-class hardware as capacity demands grow. See [GPU & Performance](/deployment/gpu-performance) for measured throughput across accelerator types.

## Full-scale resource profile

At full horizontal scale, the microservice stack (excluding the GPU node) has been measured at:

* **Requests:** 95 vCPU / 224 GB RAM
* **Limits:** 429 vCPU / 594 GB RAM
* **Recommended worker nodes:** 3, each 32 vCPU / 128 GB RAM
* **Control plane:** a separate control plane or managed Kubernetes
* **GPU node:** kept separate from the microservice worker nodes

## Customer prerequisites

* Network connectivity, including public IPs and internet access as required by your use cases
* SSL certificates
* A load balancer

<Note>
  The operating system and all required third-party software licenses are provided by BeyondGuard, on either physical or virtual infrastructure.
</Note>

## Related

<CardGroup cols={2}>
  <Card title="On-Prem Requirements" icon="server" href="/deployment/on-prem-requirements">
    Backing services, network, storage, and GPU prerequisites.
  </Card>

  <Card title="GPU & Performance" icon="microchip" href="/deployment/gpu-performance">
    Measured throughput per GPU configuration.
  </Card>

  <Card title="High Availability" icon="shield-halved" href="/deployment/high-availability">
    Multi-site active-active model serving.
  </Card>

  <Card title="Installation" icon="download" href="/deployment/installation">
    Deploy with Helm or Docker Compose.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.