Samsung Electronics and Nvidia said an AI-assisted workflow reduced a chip-scale thermal and mechanical simulation from about 18 hours on a 32-core CPU to 41 seconds on a GPU, a roughly 1,600-times speedup, Nvidia wrote in a blog post. Samsung described the same result after a joint session at Nvidia AI Day Seoul 2026.
Why it matters: Both companies said the work uses Nvidia PhysicsNeMo and an AI homogenization approach, in which AI-predicted effective material properties, coarser grids and GPU acceleration replace a fine-grid CPU solve. Nvidia said the example combined those elements for a roughly 1,600x speedup. Samsung, in more detail, said a structure represented by roughly 1 billion fine voxels could be represented at a coarser scale with about 1 million cells, and that the 41-second run was completed on a single Nvidia Hopper GPU. Samsung also said the workflow maintained 99.17% accuracy in effective vertical thermal conductivity compared with the fine-grid CPU baseline.
Samsung’s Computational Science & Engineering team from the Semiconductor R&D Center joined Nvidia for a session titled “Building Autonomous AI Engineers With NVIDIA PhysicsNeMo,” Samsung said. The companies also described related work on GPU acceleration for process, device and optical simulation and on Samsung’s sMOL materials platform. Nvidia said selected quantum chemical calculations on sMOL ran up to five times faster end to end with Nvidia cuEST. Those are vendor-reported results from a conference demonstration, not an independent lab benchmark of every workload.