GPU Setup

Local GPU Auto-Detect

copyblast simulate --engine openmm --detect-gpu

Supported backends:

• NVIDIA CUDA — nvidia-smi detection, cuDNN auto-check
• Apple Metal / MPS — torch.backends.mps.is_available()
• AMD ROCm — experimental, via HIP_VISIBLE_DEVICES

If not detected, the engine prints exact install commands:

copy# conda install -c conda-forge openmm
# pip install openmm

Engine Categories

Category	Engines	Typical Install
Molecular Dynamics	GROMACS, LAMMPS, OpenMM, MDAnalysis	conda install -c conda-forge gromacs
Quantum Chemistry	PySCF, Psi4, Quantum ESPRESSO	pip install pyscf
CFD	OpenFOAM, FEniCSx	conda install -c conda-forge fenics
Neuroscience	NEURON, Brian2, Jaxley	pip install neuron
Systems Biology	COPASI, Tellurium	pip install tellurium
Multi-physics	MuJoCo, PyBullet, Taichi	pip install mujoco
Astrophysics	Rebound, AMUSE	pip install rebound
ML/MD	JAX MD, JAX-LaB	pip install jax-md

Cloud GPU (vast.ai)

For engines too large for local hardware:

copyblast simulate --engine gromacs --estimate-cost

Example output:

Local: 8h on RTX 4090 ($0)
vast.ai: 1h on A100 ($0.80)
vast.ai: 2h on RTX 3090 ($0.40)

Workflow:

1. blast simulate --engine <name> --dry-run — shows exact conda/pip commands
2. Install locally OR rent on vast.ai
3. Re-run without --dry-run

Honest Fallback

If an engine is not installed, c4reqber automatically falls back to:

• Toy analytical model (same physics equations, coarse grid)
• Clear labeling: [FALLBACK: toy model — install <engine> for full simulation]
• No fake data — equations are real, just lower resolution