NVIDIA’s Blackwell platform has achieved impressive results in the MLPerf Training 4.1 industry benchmarks, showcasing exceptional performance in AI training capabilities. The platform highlights significant per-GPU improvements and has set new standards in generative AI training performance. This success is a result of new kernels that optimize Tensor Cores, enabling higher compute throughput per GPU while utilizing larger and faster high-bandwidth memory.
NVIDIA’s ongoing software development also contributes to improved performance and features for various frameworks and applications. Partners such as ASUSTek, Azure, Cisco, Dell, and Fujitsu have also submitted impressive results to MLPerf, further emphasizing the importance of industry-standard benchmarks in AI computing.
