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

# System Requirements

> Hardware and software requirements for running Daydream Scope

# Scope System Requirements

Comprehensive specifications for running Daydream Scope locally or in the cloud. Use this reference to verify compatibility, plan hardware purchases, or troubleshoot performance issues.

***

## GPU Requirements by Pipeline

### StreamDiffusion V2 Pipeline

* **Minimum VRAM:** 24GB
* **CUDA Version:** 12.8+

StreamDiffusion V2 provides real-time video generation with streaming capabilities.

### LongLive Pipeline

* **Minimum VRAM:** 24GB
* **CUDA Version:** 12.8+

LongLive enables extended generation capabilities for longer video sequences with consistent quality.

### Krea Realtime Pipeline

* **Minimum VRAM:** 32GB (40GB recommended for higher resolutions)
* **CUDA Version:** 12.8+

Krea Realtime requires significantly more VRAM than other pipelines. You can run it on 32GB GPUs with fp8 quantization at lower resolutions, but 40GB+ is recommended for optimal performance at higher resolutions.

### RewardForcing Pipeline

* **Minimum VRAM:** 24GB
* **CUDA Version:** 12.8+

RewardForcing uses Rewarded Distribution Matching for improved output quality on the Wan2.1 1.3B model.

### MemFlow Pipeline

* **Minimum VRAM:** 24GB
* **CUDA Version:** 12.8+

MemFlow uses a memory bank mechanism for improved long-context consistency on the Wan2.1 1.3B model.

### LTX 2.3 Pipeline (plugin)

* **Minimum VRAM:** 24GB
* **CUDA Version:** 12.8+
* **Python:** 3.12+

LTX 2.3 is a 22-billion-parameter DiT from Lightricks that generates synchronized audio and video. It ships as the [scope-ltx-2](https://github.com/daydreamlive/scope-ltx-2) plugin rather than a built-in pipeline. The model uses FP8 quantization and CPU-resident block streaming to fit on 24GB cards. See the [LTX 2.3 guide](/scope/guides/ltx-2-3) for installation.

<Note>
  Better GPUs generally yield better results with higher FPS and lower latency.
</Note>

***

## Software Dependencies

### Required Software

* **CUDA Drivers:** Version 12.8 or higher
* **UV:** Python package manager (used to run the Scope server)
* **Node.js and npm:** Required for building the frontend
* **Operating System:** Linux or Windows

### Verifying CUDA Installation

To check your CUDA version and GPU status:

```bash theme={null}
nvidia-smi
```

The output should show your GPU model and a CUDA version of at least 12.8.

***

## Cloud Deployment Options

Don't have access to a local GPU? Scope can run on cloud GPU services.

### RunPod

RunPod is a cloud GPU service that supports Scope deployment. We provide a pre-configured template for easy setup.

**Minimum GPU Requirements:**

* ≥24GB VRAM for StreamDiffusion V2, LongLive, RewardForcing, MemFlow, and LTX 2.3
* ≥32GB VRAM for Krea Realtime (≥40GB recommended)
* CUDA 12.8+ support

**Recommended Instance Types:**

* NVIDIA RTX 4090 (24GB) - Good for StreamDiffusion V2 and LongLive
* NVIDIA RTX 5090 (32GB) - Recommended for Krea Realtime

For detailed RunPod setup instructions, see the [Quick Start guide](/scope/getting-started/quickstart#cloud-deployment-runpod).

***

## Need Help?

<CardGroup cols={2}>
  <Card title="Quick Start Guide" icon="rocket" href="/scope/getting-started/quickstart">
    Follow our installation guide for step-by-step setup instructions
  </Card>

  <Card title="Join Discord" icon="discord" href="https://discord.com/invite/5sZu8xmn6U">
    Get help from the community in our #scope channel
  </Card>
</CardGroup>