SambaNova 在 SC 大会展示 Air-Cooled Token Factory 推理方案
SambaNova 和 Intel 搞了个全风冷的推理方案,prefill 用 Intel GPU D-Series,decode 用 SambaRack SN50。不用上液冷就能跑快推理,老数据中心直接用,运维和采购的人值得看看。
SambaNova 在 Supercomputing 大会现场展示了 Air-Cooled Token Factory 推理系统。该方案采用分工架构:Intel GPU D-Series 负责 prefill 阶段,SambaRack SN50 负责 decode 阶段。
整条推理链路完全采用风冷散热,不需要液冷改造,可以直接部署在现有的数据中心机房里,省去了为高密度推理负载重建机房的成本。prefill 与 decode 分离的设计让两阶段各自用更匹配的硬件,提高 token 生成效率。
对已有数据中心的企业来说,这种"风冷+现有机房"的组合降低了高性能推理的部署门槛,与需要液冷和专用机房的主流 AI 机柜方案形成了差异化路线。
The Air-Cooled Token Factory is on the floor at @Supercomputing!
@intel GPU D-Series does prefill, SambaRack SN50 does decode, & it's all air-cooled. Fast inference in the data centers that already exist.
🔗 Full blog here: https://t.co/iKHN752ZDI
This report is categorized under AITOP's "Products" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include SambaNova, Intel, inference deployment, data centers.
Source attribution: curated from SambaNova (@SambaNovaAI).