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

# Hardware Requirements

<Warning>
  Cloud-based environments are not officially supported.
</Warning>

## Requirements

All requirements are the same between validators (consensus participants) and full nodes
aside from bandwidth:

* **CPU**: 16 core CPU with 4.5 GHz+ base clock speed, e.g. AMD Ryzen 9950x, AMD Ryzen 7950x,
  AMD EPYC 4584PX, etc.
* **Memory**: 32 GB+ RAM
* **Storage**:
  * 2TB dedicated disk for TrieDB (Execution)
  * 500GB Disk for MonadBFT / OS
  * PCIe Gen4x4 NVME SSD or better for both
* **Bandwidth**:
  * 300 Mbit/s (Validators)
  * 100 Mbit/s (Full Nodes)

<Warning>
  Hard drive performance can vary dramatically by manufacturer. Below are results from internal testing:

  **Ranked performance**

  1. Samsung 980 / 990 Pro - PCIe 4.0, top class performance
  2. Samsung PM9A1 - PCIe 4.0, pretty good performance and stable performance under load
  3. Micron 7450 - PCIe 4.0, pretty good performance BUT has weird random slowdowns under a lot of load

  **Known unreliable**

  1. Nextorage SSDs - can become unresponsive under load due to overheating, requiring a system reboot.
</Warning>

A community-driven set of hardware recommendations and notes can be found [here](https://monadhcl.xyz/#recommended-hardware).

## Why Bare Metal?

Monad nodes must operate on bare metal servers rather than virtualized or cloud-based environments (e.g., AWS EC2, GCP, Azure) due to the system's strict performance and timing requirements.

A bare metal server gives the node direct, stable access to hardware, ensuring smooth operation and synchronization with the network:

* Monad’s consensus protocol enforces tight time windows—blocks are proposed and voted on in sub-second intervals, and the network assumes nodes can validate and execute blocks within this budget. In such a situation, cloud-based environments may introduce latency and unpredictability, which can cause nodes to miss deadlines, fall behind in block processing, or become unstable during high-throughput periods.

* Even when resources appear sufficient on paper, virtualization adds an additional layer of software between the node and physical hardware. This layer introduces context switching overhead and restricts direct I/O access to SSDs and network interfaces. These effects are negligible for average compute tasks but become significant when sustained high-throughput and low-latency operations are required, as in Monad’s consensus and execution loops.

In summary, a bare metal server provides predictability and determinism, which are crucial for maintaining synchronization and throughput across the network. Cloud-hosted VMs may work under light loads, but they cannot guarantee consistent real-time performance required by Monad consensus at scale.