Download

Ready-to-run program — no Python needed

Unpack the archive and run it. There is one program per system: it computes on the CPU, and if the NVIDIA driver and CUDA Toolkit 12 are installed — on the GPU too, tens of times faster.

Builds are coming soon.

Computing on an NVIDIA GPU

Optional: without a GPU everything runs on the CPU. The driver and CUDA are installed separately:

GPU: NVIDIA GeForce GTX 900 (Maxwell) or newer
NVIDIA driver: at least 527.41 (Windows) or 525.60.13 (Linux); a recent one is better
CUDA Toolkit: 12.x — 12.0–12.9, 12.9 recommended; RTX 50xx needs 12.8+. CUDA 13 does not fit — 12.9 can be installed next to it.

CUDA Toolkit — from the NVIDIA archive; on Ubuntu 24.04 and Debian 13 the nvidia-cuda-toolkit package fits. Step by step — README, «Computing on an NVIDIA GPU».

For plate solving it is worth installing ASTAP with the D50 or D20 star database; without it Drift uses the Gaia catalog over the internet.

All versions

The newest version is on top. The SHA-256 checksum helps make sure the archive downloaded completely.

Builds are coming soon.