NVIDIA's Blackwell Chip Revolutionizes Large AI System Training Efficiency, Slashing Chip Requirements
2025-06-05 / Read about 0 minute
Author:小编   

NVIDIA's latest chip, the Blackwell, has achieved a groundbreaking leap in the training of large AI systems, boasting a speed over twice that of its predecessor, Hopper. According to MLCommons data, a mere 2496 Blackwell chips are sufficient to complete the training test within 27 minutes, a stark contrast to the triple amount required by the previous generation of chips. The current trend in the AI industry favors the use of smaller, interconnected chipsets to form subsystems, thereby accelerating the training of large models.