NVIDIA GB10 Edge AI Hardware Lacks Energy Attribution for Agentic Workloads

An audit of the ASUS Ascent GX10, utilizing NVIDIA's GB10 SoC, uncovers a significant gap in energy observability. Notably, the platform lacks a CPU energy counter, INA power-rail monitor, IPMI/BMC, or SCMI powercap protocol accessible through any supported software interface. The sole energy telemetry available on the device is the instantaneous GPU power via NVML. This oversight is alarming, particularly as agentic AI tasks—where a single user objective initiates a series of orchestration steps, tool calls, retries, and recovery efforts—are being prioritized for edge deployment. In 2026, companies such as NVIDIA, Dell, HP, ASUS, MSI, Acer, and Gigabyte will release GB10-based desktop AI systems. Previous studies indicate that orchestration structure significantly influences energy costs, with workflows using 4.33 times more energy per successful goal compared to linear baselines, and up to 7.63 times for multi-step reasoning tasks. Furthermore, CPU processing can contribute to 90.6% of total latency and 44% of dynamic energy in these workloads. The absence of process-level energy tracking on GB10 systems hinders developers' ability to enhance energy efficiency, which is increasingly vital as edge AI implementations expand.