feat: Node Total Power Monitoring - Track system-wide power consumption per compute node

[inureyes]

Overview

This is a tracking issue for implementing node-level total power monitoring across all supported hardware platforms in all-smi. The goal is to provide users with visibility into the total power consumption of each compute node, including chassis-level power where accessible.

Motivation

• Power Efficiency Monitoring: Track and optimize power usage across GPU/NPU clusters
• Capacity Planning: Understand power requirements for infrastructure scaling
• Thermal Management: Correlate power consumption with thermal conditions
• Cost Estimation: Enable accurate power cost calculations for workloads
• Environmental Impact: Support sustainability reporting and carbon footprint tracking

Scope

Platforms to Support

Platform	Power Source	Notes
Apple Silicon	powermetrics	Combined Power (CPU + GPU + ANE)
NVIDIA	nvidia-smi	GPU power; IPMI/BMC for chassis
Intel Gaudi	hl-smi	NPU power consumption
Tenstorrent	luwen API	NPU power metrics
Rebellions	rbln-stat	NPU power consumption
Furiosa	furiosa-smi	NPU power metrics
x86 Servers	IPMI/BMC/RAPL	Chassis-level or CPU power

Implementation Approach

1. Platform-Specific Collection

   • Use native tools for each platform (powermetrics, nvidia-smi, hl-smi, etc.)
   • For server-class hardware, explore IPMI/BMC interfaces for chassis power
   • Fall back to component-level aggregation where chassis power is unavailable

2. Data Model

   • Add total_power_watts field to node-level metrics
   • Distinguish between chassis power (if available) and component sum
   • Include power source metadata in API responses

3. Display & Export

   • Show total power in TUI view mode
   • Export as Prometheus metric all_smi_node_total_power_watts
   • Include power efficiency metrics where applicable

Limitations

• Some platforms may only provide component-level power (GPU/NPU only)
• Chassis-level power requires IPMI/BMC access (often requires elevated privileges)
• Power accuracy varies by platform and sensor quality

Sub-issues

Implementation will be tracked in platform-specific sub-issues:

• feat: macOS Apple Silicon - Total Power monitoring via powermetrics #82 - macOS Apple Silicon - Total Power via powermetrics
• feat: Add Chassis/Node-level monitoring with per-node power and BMC metrics #84 - Chassis/Node-level monitoring with per-node power and BMC metrics
• NVIDIA - GPU Power aggregation
• Intel Gaudi - NPU Power monitoring
• Other platforms (as needed)

---

This is a tracking/meta issue. Individual implementation details will be in linked sub-issues.