System Requirements & Hardware Specs
Hardware recommendations and performance benchmarks across CPU, GPU, and Apple Silicon architectures.
Preview Content
Please note that the details on this page are pre-content drafts. More detailed technical specifications, architectural breakdowns, and updated guides will be added soon.
Hardware Specification Matrix
XianScan is built in pure Rust with lock-free mimalloc memory allocation and SIMD optimizations:
| Tier | Minimum (CPU Only) | Recommended (Local GPU) | High-End / Server |
|---|---|---|---|
| Processor | 4-Core x86_64 with AVX2 or Apple M1+ | 6 to 8 Core CPU (AVX2 / AVX-512) | 8+ Core CPU / Modern Xeon / EPYC |
| System RAM | 8 GB (Engine RSS ~1.2 GB + image buffers) | 16 GB | 32 GB+ |
| Graphics (VRAM) | None (Integrated / CPU inference) | NVIDIA RTX 3060 / 4060 / 5060 (6 GB - 8 GB+) | NVIDIA RTX 4070+ / RTX 5080 / Tesla T4 / L4 / A10G (16 GB+) |
| Inference Time (per page) | ~3.5 - 6.0 seconds | ~0.4 - 0.9 seconds | ~0.15 - 0.35 seconds |
| Translation Engine | Cloud API (DeepSeek V4 Flash / Gemini 3.7) | Local 7B - 14B LLM (Qwen2.5 / DeepSeek-R1) | Local 32B - 70B LLM (Qwen2.5:32B / Llama 3.3:70B) |
Supported Execution Providers (EP)
- DirectML (Windows): Uses Win32 DXGI adapter enumeration to automatically route ONNX model inference to NVIDIA GeForce, AMD Radeon RX, or Intel Arc GPUs without external CUDA setup.
- CUDA 12 + cuDNN 9 (Linux / Windows): Dedicated GPU acceleration with automated driver persistence mode (
nvidia-smi -pm 1). - CoreML (macOS): Hardware acceleration on Apple Silicon (M1/M2/M3/M4) leveraging the Apple Neural Engine (ANE) and Metal compute.
- CPU Fallback: Multi-threaded SIMD inference for standard laptops and systems without dedicated graphics.