运营热点解读
Setting a World Record for MoE Pre-Training on NVIDIA GB300 NVL72
雷达摘要
AI-Generated Summary NVIDIA GB300 NVL72 achieved a world record of 1,648 TFLOPs per GPU in pre-training DeepSeek-V3 671B, demonstrating a rack-scale, fully co-designed AI infrastructure with fifth-generation NVLink, 1.8 TB/s per-GPU bandwidth, and 130 TB/s non-blocking all-to-all bandwidth. Efficient scaling of Mixture of Experts (MoE) architectures is enab…
原始信息
本页是基于公开来源生成的摘要与运营提示,不替代原文。请以原始发布方内容为准。
电商热点雷达