百度把千卡级超节点的上市时间从 2028 年提前到了明年,单套就能训 5 万亿参数模型,多轮对话场景首 token 延迟降了 90%。做大规模训练或长上下文推理的团队,可以把它和 Kunlun 芯片路线一起看看。
#compute clusters
Moore Threads的2112卡S5000集群商用数据出来了,跑DeepSeek V4 Flash能到H800的92%,利用率92.9%,看看国产GPU实际表现。
华为昇腾 950DT 千卡超节点落地武汉,FP8 算力 1EFLOPS,1 个月就建成了,湖北搞大模型的可以关注本地算力。
Optical co-packaging via Hi-ONE NPO points to where large-scale AI cluster interconnects are heading, trading discrete pluggable modules for integrated optics with measurable savings in parts count and power. Hardware engineers planning training clusters and procurement teams evaluating alternatives to NVIDIA-based superpods will find the scale and efficiency figures directly comparable.
The mobile computing factory reduces costs by 25% and enables flexible deployment for renewable energy integration, benefiting ASEAN regions with similar energy infrastructure challenges.
曙光发了新一代AI集群Dawn 8000,密度提升20倍还能连10万张卡,搞大规模训练的有福了。
55亿、11520张昇腾芯片、9000P总算力,这个国产集群全自主可控,90天建成,算力自主的新标杆。
烽火通信这款 13824 芯光缆解决了超大规模算力集群的楼间互联痛点,做数据中心基建或光通信的团队值得关注,能大幅节省管道资源并提升布线效率。