System Requirements
Supported systems, recommended hardware, and which GPUs XianScan can use.
Supported Systems
- Windows 10 or 11, 64-bit (x86_64).
- Linux x86_64 with a recent C library. The release is built on Ubuntu 24.04, so older distributions such as Ubuntu 20.04 cannot run it. On an older system, use the Docker image instead.
- macOS on Apple Silicon (M1 or newer). There is no Intel Mac build.
- Docker on any Linux x86_64 host.
Hardware
| Minimum | Recommended | |
|---|---|---|
| CPU | 4 cores | 6 cores or more |
| Memory | 8 GB | 16 GB |
| GPU | Not required | A dedicated GPU with 6 GB or more of memory (see below) |
| Disk | About 2 GB for the app | Plus room for your library; every page is stored as original, cleaned and translated images |
The translation model is separate from this:
- Cloud API (DeepSeek, Google AI Studio and others): no extra hardware.
- Local model with Ollama or LM Studio: needs GPU memory of its own. As a rough guide,
qwen3.5:4bfits 4 to 6 GB,qwen3.5:9b8 to 12 GB, andqwen3.5:27b16 GB or more. The XianScan models and the local LLM share the GPU.
Which GPUs Are Used
| System | Used for | Notes |
|---|---|---|
| Windows | Detection and cleaning, through DirectML | Only on a dedicated NVIDIA, AMD or Intel Arc GPU. Text recognition (OCR) always runs on the CPU on Windows. No CUDA install is needed. |
| Linux | Detection, OCR and cleaning, through CUDA | NVIDIA only. Needs the NVIDIA driver, CUDA 13 and cuDNN 9, see GPU Acceleration. |
| macOS | Detection, OCR and cleaning, through CoreML | Apple Silicon. |
| Docker | CPU only | For a GPU on Linux, run the native binary instead. |
Integrated graphics (Intel UHD / Iris, AMD Radeon Graphics / Vega) are skipped on purpose and run on the CPU, because they are usually slower than the CPU for these models.
If a GPU fails to start a model (for example because of a missing library), XianScan switches to the CPU so pages keep working, and stays on the CPU until you choose a device again in Settings -> Hardware & Compute or restart XianScan.