# Download Rugnux

`rugnux` is one self-contained executable: no CUDA toolkit, no Qt, no service to run. Take the
archive for your machine, unpack it into a directory of its own, and run it. The links point at the
current release; the full list of what each release ships is on its release page.

```{eval-rst}
.. jfjoch-downloads:: rugnux
```

On a machine with an NVIDIA GPU take the **GPU** archive: it is the only one that uses the card,
and it also runs on a machine without one (it reports zero CUDA devices and takes the CPU path).
The **CPU only** archive is smaller and for machines that will never have a GPU. A V100 needs the
CUDA 12 build; the CUDA 13 builds start at Turing (T4). The driver versions and the GPU generations
each build covers are in [Release contents](RELEASE_CONTENTS.md#gpu-generations-and-the-nvidia-driver).

## After downloading

The archive has **no top-level directory**: it unpacks straight into `bin/` and `share/`, so always
give it a directory of its own.

```
mkdir -p ~/rugnux
tar xzf rugnux-<version>-linux-x86_64-cuda12.tgz -C ~/rugnux      # Linux, macOS
~/rugnux/bin/rugnux                                               # prints the usage
```

On Windows, extract the `.zip` into a folder and run `bin\rugnux.exe` from a terminal. On macOS the
binary is signed and notarized; the first run of a freshly downloaded copy checks the notarization
online.

Nothing is written outside that directory, nothing needs root, and several versions can sit side by
side; to remove one, delete its directory. Put `bin/` on your `PATH` to type `rugnux` rather than
the full path, then:

```
rugnux dataset_master.h5
```

[Rugnux](RUGNUX.md) takes it from there: what the run writes, what the terminal shows and the few
options worth knowing.

## Linux packages

On RHEL / Rocky 8 and 9 and Ubuntu 22.04 / 24.04, `rugnux` is also a package of its own in the
[package repositories](REPOSITORIES.md), which keeps it current with the system's package manager:

```
sudo dnf install rugnux          # RHEL / Rocky
sudo apt install rugnux          # Ubuntu
```

The package depends on nothing from the acquisition side and installs `/usr/bin/rugnux`. The
repository's CUDA variant keeps cuFFT as a system dependency, where the archive above links it in;
either works on the same machine.

## From source

`cmake -S . -B build -DJFJOCH_RUGNUX_ONLY=ON` builds `rugnux` alone, on Linux, macOS and Windows,
with every dependency fetched during the first configure. The flags that match the released
binaries, and what a build needs on a Mac, are in
[Installing Rugnux ▸ Building from source](RUGNUX_INSTALL.md#building-from-source).

## Memory and hardware

A run wants host and GPU memory in proportion to the number of reflections it integrates, not to
the detector: 16 GB of RAM and a 6 GB card for an ordinary sweep, 32 GB and 8 GB for the largest
measured, and far more for a very large unit cell. The sizing table and the flags that lower it are
in [Installing Rugnux ▸ Memory](RUGNUX_INSTALL.md#memory).