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

# Hardware requirements

> Two badges, two bars — AI READY for running models locally, AI OPTIMAL for earning by sharing GPU power.

export const Win = () => <span style={{
  display: 'inline-block',
  padding: '2px 10px',
  borderRadius: '999px',
  fontSize: '0.82em',
  fontWeight: 600,
  background: 'rgba(0,120,212,0.18)',
  color: '#7CC4FF',
  border: '1px solid #0078D4',
  whiteSpace: 'nowrap'
}}>Windows</span>;

The AI App has two distinct use cases, and they have different hardware bars:

* **AI READY** — run AI models locally (chat, image, video, voice) on your own GPU.
* **AI OPTIMAL** — also accept network earning workloads to your machine and get paid in GUSD.

<Frame>
  <img src="https://mintcdn.com/gamerhash/xVp-mW9DIyqS08n6/images/hardware-tiers.png?fit=max&auto=format&n=xVp-mW9DIyqS08n6&q=85&s=ae78f2704d09bca18965c4ef7749456c" alt="VRAM tier ladder: 8 GB and 12 GB qualify for AI READY (local models only); 16 GB and 24 GB+ also qualify for AI OPTIMAL (earn from network workloads). Each tier shows what models unlock — entry chat models and image generation at 8 GB, 14B chat and FLUX at 12 GB, 20B chat and video replace at 16 GB, full Model Center at 24 GB+." width="1600" height="736" data-path="images/hardware-tiers.png" />
</Frame>

Check your exact rig against both badges in the official [GamerHash profit calculator](https://gamerhash.com/en/calculator) before you buy.

## AI READY — run models locally

Lower bar. If you can run modern AAA games at 1080p, you can run local AI.

* **OS** — <Win /> 10 (22H2+) or 11
* **GPU** — NVIDIA RTX with **8 GB+ VRAM** (RTX 3060, 4060, 5060 and up)
* **RAM** — 16 GB
* **CPU** — modern x86-64 (Intel 8th gen / AMD Ryzen 2000 or newer)
* **Driver** — recent NVIDIA studio or game-ready driver
* **Disk** — \~30 GB free for downloaded models

The more VRAM you have, the bigger the local model you can load:

| VRAM       | Modules unlocked                                                                                                                                |
| ---------- | ----------------------------------------------------------------------------------------------------------------------------------------------- |
| **8 GB**   | AI Chat (7B/8B quantized), Sana Sprint, FLUX.1 Schnell, voice (TTS + Whisper transcription), OCR, image upscale, background removal, embeddings |
| **12 GB**  | 14B chat models, FLUX dev / FLUX.2 / Z-Image-Turbo at higher resolutions, image edit, LTX-Video 480p, video upscale, music generation           |
| **16 GB+** | 20B-class chat models, high-res FLUX, longer LTX-Video clips, video replace / animate                                                           |
| **24 GB+** | Full Model Center access — every model the app ships                                                                                            |

The Model Center inside the app shows live fit per model on your specific hardware.

## AI OPTIMAL — earn from GPU sharing

Higher bar. Earning starts at **16 GB VRAM** — anything below that is AI READY only.

Everything in **AI READY**, plus:

* **GPU** — NVIDIA RTX with **16 GB+ VRAM** (RTX 4060 Ti 16 GB, RTX 4070 Ti Super, RTX 4080 / 5080, RTX 4090 / 5090, Blackwell pro)
* **RAM** — **32 GB minimum**, more is better. Concurrent workloads benefit from larger system memory.
* **Network** — stable broadband (50 Mbps down comfortable). Earning workloads stream small payloads — consistency matters more than peak speed.
* **Uptime** — leave the machine on. Most pay comes from being online when a workload arrives.

The more VRAM, the higher-paying workload tiers you qualify for:

| VRAM       | Earning workload pool                           |
| ---------- | ----------------------------------------------- |
| **16 GB**  | Entry tier — video and 20B-class chat workloads |
| **24 GB+** | Full earning pool, top-paying tiers             |

The [profit calculator](https://gamerhash.com/en/calculator) gives you a per-GPU yield estimate based on current network demand.

## What's not supported

* **AMD GPUs** — NVIDIA only in the AI App today.
* **macOS / Linux** — AI App is Windows-only. The [Mobile app](/apps/mobile) on iOS/Android handles account management, not local AI.
* **iGPUs** — integrated graphics don't have enough VRAM.
* **Multi-GPU per machine** — the AI App uses a single GPU per process. Multi-machine is fine; GHXP and earnings consolidate per account.

## Upgrading the GPU

Swap GPUs without recreating your account. On next start, the AI App detects the new card and re-runs its benchmark — typically a couple of minutes. GHXP, account level, and earning history all carry over. Workloads matched to the new card take effect from the next dispatch.

## Tips before you buy

* **Run your exact rig through the [profit calculator](https://gamerhash.com/en/calculator)** before spending. Both badges (AI READY / AI OPTIMAL) are checked there.
* **For AI OPTIMAL, prioritize VRAM over clocks.** 16 GB is the entry threshold — a 16 GB 4060 Ti qualifies for the earning pool, an 8 GB 4070 doesn't.
* **Used 30-series cards** can be cheap entry to AI READY (3060/3070 at 8–12 GB). For AI OPTIMAL on used hardware, only the **RTX 3090 (24 GB)** clears the bar — and check warranty, sustained load is harder on cards than gaming.
* **Check live network demand** at [network stats](/ai/stats). A bigger GPU only pays more if there are workloads in the queue for it.

## Next

<CardGroup cols={2}>
  <Card title="Modules" icon="layer-group" href="/ai/modules">
    Match modules to what your GPU can run.
  </Card>

  <Card title="Earning model" icon="coins" href="/ai/earning">
    What workloads pay more, and how to read the queue.
  </Card>

  <Card title="Live network stats" icon="chart-line" href="/ai/stats">
    Current GPU count, VRAM, and capacity across the network.
  </Card>

  <Card title="Install the app" icon="download" href="/tutorials/install-ai-app">
    Step-by-step from download to first run.
  </Card>
</CardGroup>