Breadcrumb

VSCLAB introduced open-sourced thermal map dataset from commercial CPU/GPU/TPU multi-core and many-core processors

VSCLAB introduced the first publicly available thermal map dataset from commercial CPU/GPU/TPU multi-core and many-core processors! These datasets were gathered using our advanced thermal imaging system at VSCLAB, UC Riverside and include both raw thermal maps and workload-dependent utilization metrics (e.g., Intel PCM, AMD uProf 4.0). This unique combination enables real-time full-chip thermal modeling using deep learning for runtime thermal, power, and reliability management (see our recent works on this).

Access the dataset here:https://github.com/sheldonucr/commercial_thermal_map_dataset

We hope this resource will foster innovative research in thermal map estimation, modeling, and management. Below is an example of measured thermal maps and hotspots from an AMD Ryzen 7 4800U (8-core) CPU under real workloads.

See the linkedin post here

AMD hotspot