Documentation

Building from Source

Build NIKOLA Chess Engine from source. The build system automatically downloads the MIND compiler and runtime.

Source files go through the MIND compiler and come out as one binary per GPU backend.

System Requirements

Linux

  • Ubuntu 20.04+
  • x64 with AVX2
  • 4 GB RAM minimum
  • 500 MB storage

macOS

  • macOS 12+
  • Intel or Apple Silicon
  • 4 GB RAM minimum
  • 500 MB storage

Windows

  • Windows 10+
  • x64 with AVX2
  • 4 GB RAM minimum
  • 500 MB storage

Quick Build

The build system automatically downloads the MIND compiler and runtime libraries:

git clone https://github.com/star-ga/NikolaChess.git
cd NikolaChess
make setup
make release

Build Targets

make releaseDefault

Build optimized release binary with auto-detected GPU backend.

make cuda

Build with NVIDIA CUDA backend for RTX 5090, A100, H100, GB200, GB300, Vera Rubin.

make metal

Build with Metal backend for Apple Silicon (M1/M2/M3/M4 series).

make rocm

Build with ROCm backend for AMD GPUs (RX 7000/8000, MI200/MI300X).

make oneapi

Build with Intel oneAPI backend for Arc/Xe GPUs and Ponte Vecchio.

make cpu

Build CPU-only version with AVX-512/AMX (Intel) or NEON/SVE (ARM).

GPU Requirements

For GPU-accelerated builds:

  • NVIDIA (CUDA) - RTX 30/40/50 series, A100, H100, GB200, GB300, Vera Rubin with CUDA 12.x, 8GB+ VRAM recommended
  • AMD (ROCm) - RX 7000/8000 series, MI200/MI300X with ROCm 6.0+
  • Intel (oneAPI) - Arc A770/A780, Xe HPG, Ponte Vecchio with oneAPI 2024+
  • Apple (Metal) - M1/M2/M3/M4 series Apple Silicon

Build Output

After a successful build, the executable will be in the project root:

  • nikola-cuda - NVIDIA GPU version
  • nikola-metal - Apple Silicon version
  • nikola-rocm - AMD GPU version
  • nikola-cpu - CPU-only version

Verify the Build

./nikola-cuda --version

Manual Compiler Installation

If you prefer to install the MIND compiler manually:

Linux / macOS
curl -fsSL https://mindlang.dev/install.sh | bash
Windows
irm https://mindlang.dev/install.ps1 | iex

Troubleshooting

  • make: command not found - Install build-essential (Linux) or Xcode Command Line Tools (macOS)
  • CUDA errors - Verify CUDA toolkit version matches your driver with nvidia-smi
  • Out of memory - Reduce parallel jobs with make release -j4
  • AVX2 not supported - Your CPU is too old; use a newer system or try WSL2
UCI ConfigurationNeural Network