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

# Hardware requirements

> This topic describes hardware requirements and recommendations for production deployments. Actual resource requirements may vary based on your specific workload.

## Supported architectures

RisingWave can run on the following hardware architectures:

* x86\_64 (64-bit Intel/AMD CPUs)
* ARM64
* Mac with Apple Silicon

<Note>
  **SIMD requirements for Docker images**

  The docker images for `x86_64` are built with AVX2 SIMD extensions, and the images for `aarch64` (ARM64) are built with NEON SIMD extensions. These must be supported by your machine. If not, build the docker image with the build-arg `simd_disabled=true`.

  * To build the images, run the following from the project root: `docker build . -f docker/Dockerfile`.
  * To build the images without SIMD vector extensions, run the following from the project root `docker build . -f docker/Dockerfile --build-arg simd_disabled=true` and run any subsequent docker commands on the resultant image.
</Note>

## Kubernetes resource sizing guidance

When deploying on Kubernetes (via Helm or the RisingWave Kubernetes Operator), set resource `requests` and `limits` to the **same values** for each component. Matching them places pods in the [`Guaranteed` QoS class](https://kubernetes.io/docs/concepts/workloads/pods/pod-qos/), which improves scheduling predictability and makes pods less likely to be evicted under resource pressure.

RisingWave is memory-intensive. Use a **1:4 CPU:memory ratio** when sizing components (for example, 1 CPU core to 4 GiB memory). See [Optional: Resize a node](/deploy/risingwave-k8s-helm#optional-resize-a-node) for an example `values.yaml` snippet.

## Compute Nodes

Compute Nodes (including Streaming Nodes and Serving Nodes) handle query processing and state management. More resources enable higher query throughput. For cost efficiency, machines with high memory-to-CPU ratios (4:1 or higher) are recommended due to RisingWave's memory-intensive nature.

* Minimum:
  * 2 CPU cores
  * 8 GiB memory
* Recommended:
  * ≥4 CPU cores
  * ≥16 GiB memory

## Compactor Nodes

Compactor Nodes perform background compaction jobs to optimize state storage. Insufficient resources for compactors will result in slow state access and impact overall performance.

* Minimum:
  * 1 CPU core
  * 4 GiB memory
* Recommended:
  * ≥ 2 CPU cores
  * ≥ 8 GiB memory

## Meta Nodes

Meta Nodes manage metadata and coordinate the cluster. It is advisable to deploy 2 Meta Nodes (primary and backup) to ensure faster recovery and avoid single points of failure (SPOF).

* Minimum:
  * 1 CPU core
  * 4 GiB memory
* Recommended:
  * ≥ 2 CPU cores
  * ≥ 8 GiB memory

## Frontend Nodes

Frontend Nodes serve SQL queries from clients, parse and plan queries, and route requests to Compute Nodes. They are stateless and typically do not require as many resources as Compute or Compactor Nodes, but they should still be sized to handle peak client concurrency.

* Minimum:
  * 1 CPU core
  * 4 GiB memory
* Recommended:
  * ≥ 2 CPU cores
  * ≥ 8 GiB memory

## Storage

RisingWave offers support for multiple storage systems as storage backends. For the complete list of supported storage systems for Kubernetes deployments, see [Set up a RisingWave cluster in Kubernetes](/deploy/risingwave-kubernetes#deploy-a-risingwave-instance).

Please notice that storage performance can **significantly** impact RisingWave's performance. We recommend using high-performance cloud storage systems such as AWS S3. For self-managed storage systems such as MinIO or local file system, please ensure to use high-performance SSD disks.