Odaily Planet Daily News According to official news, the AI World computing power network has completed the second hardware expansion, the main force is NVIDIA A100 and NVIDIA H100 professional cards, and the newly added H100 enables double-precision Tensor Core floating point operations per second (FLOPS) Increased by 3 times, providing HPC with 60 teraFLOPS of FP64 floating-point operations.
High-performance computing applications that incorporate AI can leverage the H100's TF32 precision to achieve 1 petaFLOP of throughput, enabling single-precision matrix multiplication operations without code changes.