AMD calls out Nvidia's math shortcuts for supercomputers
Original headline: AMD attacks Nvidia's FP64 emulation at SC26, touts native double-precision on Instinct MI430X — tomshardware.com
More in Semiconductor and chip news
At the Supercomputing 2026 conference, AMD is highlighting a key technical difference between its data center GPUs and Nvidia's. AMD's latest Instinct MI430X accelerator supports native FP64 (double-precision) math, which is essential for high-fidelity scientific simulations in fields like hydrodynamics, molecular dynamics, and finite element analysis. Nvidia, having shifted its chip designs to prioritize the lower-precision math used for AI, relies on emulation techniques to handle FP64 operations.
For engineers and scientists at national labs and supercomputing centers, this means workloads that aren't simple dense matrix math could suffer performance penalties on Nvidia hardware. AMD argues that for the vast majority of complex, real-world scientific computing tasks, native double-precision support remains non-negotiable for accuracy and speed. This battle underscores a fundamental split in the GPU market: raw AI power versus precision for traditional supercomputing.
Read the original at tomshardware.com
Satirical commentary on real tech news, summarised with AI. Read the original source for the facts. How ZeroDaily works
