GPU & AI Accelerator Diagnostics

Knows What’s Wrong Before You Do · Shipped · workstation

Dedicated hardware inspection for GPUs (NVIDIA, AMD, Intel, Apple Metal) and AI accelerators (Coral TPU, Hailo, MemryX, NPUs).

Specialized deep inspection for compute hardware that goes beyond basic discovery. Detects GPU architecture, driver versions, VRAM allocation, thermal state, and active workloads. For AI accelerators, identifies attached edge TPUs and neural processing units with their firmware and runtime status.

GPU tools probe nvidia-smi, rocm-smi, intel_gpu_top, and Metal system profiler. Accelerator tools detect Coral Edge TPUs, Hailo-8/8L/10H, MemryX MX3, Intel NPU, AMD XDNA NPU, and Apple Neural Engine through platform-specific APIs and device enumeration.

Limits and invariants

GPU analysis is cached to YAML with a 7-day TTL to avoid redundant expensive probes.

Where this lives

• halbert_core/halbert_core/tools/gpu_tools.py
• halbert_core/halbert_core/tools/accelerator_tools.py
• halbert_core/halbert_core/discovery/scanners/gpu.py
• halbert_core/halbert_core/discovery/scanners/ai_accelerator.py
• halbert_core/halbert_core/dashboard/routes/gpu.py