Software

The image on every machine

One stack, identical on both systems, so code developed on the tower moves to the rack unchanged.

Base OS

  • Ubuntu 20.04 LTS
  • Cinnamon desktop

Build & numerics

  • NVIDIA CUDA 11.4.48
  • NVIDIA driver 470.256.02
  • GNU C/C++ 9.4.0
  • CMake 3.31
  • BLAS / LAPACK

Analysis & viz

  • Python
  • NumPy / SciPy
  • VTK visualization

Examples

Worked examples, source included

Three complete CUDA programs ship on the machine. They compile out of the box and are meant to be taken apart.

Mandelbrot set viewer

Interactive escape-time renderer. A clean introduction to grid/block decomposition and per-pixel parallelism.

Laplace solver

Relaxation method on a 2D grid: stencil operations, halo exchange, convergence testing on device.

Monte Carlo photon scattering

Embarrassingly parallel transport simulation: per-thread RNG state and reduction across millions of paths.