Units

CPU, GPU, and NPU inherit from Unit. They share the same shape so a tool can treat every compute device uniformly, without forcing the machine through a CUDA-only model.

from mainboard import Machine

for unit in Machine().units:
    print(unit.kind, unit.vendor, unit.label)
    print(unit.snapshot())

Every unit exposes:

attribute	meaning
kind	class identity: UnitKind.CPU, UnitKind.GPU, or UnitKind.NPU
vendor	provider identity such as Vendor.APPLE or Vendor.NVIDIA
name	human-readable model name
architecture	architecture string when known
memory	a Memory reading: total, used, and free bytes for the memory the unit sees
snapshot()	typed telemetry snapshot

Every unit reports memory the same way, so unit.memory.total_bytes and unit.memory.used_gb work uniformly across CPU, GPU, and NPU.

Snapshots

Unit.snapshot() returns UnitSnapshot. GPU.snapshot() returns GPUSnapshot, which extends UnitSnapshot with GPU-specific telemetry.

machine = Machine()
cpu_snapshot = machine.cpu.snapshot("setup")
gpu_snapshot = machine.gpus[0].snapshot("kernel")

Snapshot models are Pydantic models and can be serialized with model_dump().