hardware — compute resources

Schema: HardwareConfig in src/thesis/core/config/validators.py.

Drives Nipype scheduler hints, the startup GPU check, and downstream node-level resource hints. Copy config/hardware.example.yaml to config/hardware.yaml (gitignored) and adjust for your machine.

Field	Type	Default	Constraints	Description
threads	int	4	1 ≤ n ≤ 128	CPU threads available to workflow execution. Read by Nipype plugin args and per-node thread hints.
memory_gb	int	16	1 ≤ n ≤ 1024	Memory budget in GB used by Nipype memory_gb plugin arg.
gpu_enabled	bool	False	—	Enable GPU-aware node selection (e.g. probtrackx2_gpu). On the first run the CLI validates CUDA availability via thesis.core.gpu.check_gpu() and prompts to fall back to CPU on interactive shells (auto-falls back without prompting in non-TTY contexts).
gpu_device	int | None	None	—	GPU device index for multi-GPU systems. Currently informational; the binary selects its own device unless CUDA_VISIBLE_DEVICES is set.
n_gpu_procs	int	1	≥ 1	Nipype scheduler GPU worker slots — maximum number of GPU-bound nodes that may run concurrently. Prevents CUDA OOM. Only effective when gpu_enabled: true and the active plugin honours n_gpu_procs (MultiProc).
n_gpus	int	1	≥ 1	Physical GPUs visible per worker slot. Passed through as a scheduler hint.

Example

hardware:
  threads: 16
  memory_gb: 64
  gpu_enabled: true
  n_gpu_procs: 2     # allow 2 concurrent GPU nodes (e.g. left + right hemisphere)
  n_gpus: 1          # each slot sees 1 GPU

Interaction with nipype.plugin_args

hardware values are injected as plugin args at CLI build time:

• n_procs ← falls back to cli --max-workers, then nipype.plugin_args.n_procs, then the plugin default.

• memory_gb ← hardware.memory_gb is propagated when not already set in nipype.plugin_args.

• n_gpu_procs / n_gpus ← copied when hardware.gpu_enabled: true.

You can still set the underlying nipype.plugin_args explicitly to override.