LesenForschungRadarAnlageframework
Anmelden / Registrieren
← Zurück zur Lektüre
ThemaDisconfirmAI RedaktionAktualisiert 2026-11-19 07:29

Lambda

3 Schritte1 Beiträge4 Quellen

美国云服务商Lambda于2026年11月16日宣布,在魁北克园区完成全球首个大规模共封装光学(CPO)部署,使用NVIDIA Photonics交换机连接超10,000颗NVIDIA Blackwell Ultra GPU。据Lambda自测数据,该方案使机架级网络延迟降低约3.8%,Pod级最高降低10.1%,同时保持吞吐量不变;光学引擎中位温度从传统可插拔光模块的51°C降至36°C。NVIDIA声称该技术可提供高达5倍的电源效率和15倍的可靠性提升,但此为架构一般性规格,非本次部署实测结果。【存疑】Lambda未披露绝对延迟数值及测试流量配置文件,限制了与其他网络架构的直接比较。下一步需关注NVIDIA在SC26特别演讲中公布的更多技术细节,以及长期运行中的现场故障率与维护成本数据。

Wie es dazu kam

  1. 2026-11-16

    Lambda宣布完成全球首个大规模CPO部署,使用NVIDIA Photonics交换机连接魁北克园区的10,000+ NVIDIA Blackwell Ultra GPUs。

    lambda.com ↗
  2. 2026-11-16

    NVIDIA博客发布文章,确认Lambda部署细节:每个交换机拥有144个800Gb/s端口,液冷设计,并报告了延迟和温度改善的具体数据。

    blogs.nvidia.com ↗
  3. 2026-11-17

    Silicon Report报道指出,相比传统可插拔光模块,CPO在机架规模下延迟降低约4%,在Pod规模下最高降低10%,且吞吐量保持一致。

    siliconreport.com ↗

Was noch ungeklärt ist

  • NVIDIA声称的5倍电源效率提升是否会在长期运行中得到独立第三方验证?
  • Lambda未披露绝对延迟数值和测试流量配置文件,这如何影响与其他网络架构的比较?
  • 现场可更换的外部光源模块在实际运维中的故障率和维护成本表现如何?

Quellen

Anmelden / Registrieren
LesenForschungRadarAnlageframework
blogs.nvidia.com
NVIDIA Blog
↗
  • lambda.comLambda Blog↗
  • siliconreport.comSilicon Report↗
  • convergedigest.comConverge Digest↗
  • Abdeckung in diesem Thema

    • Cloud Provider Lambda Deploys NVIDIA Photonics Switches: Self-Tests Show 4% Latency Drop; NVIDIA Claims 5x Efficiency Gain2026-11-17News

      US cloud provider Lambda announced the deployment of NVIDIA Photonics co-packaged optics (CPO) switches in a cluster connecting over 10,000 NVIDIA GB300 GPUs. According to their tests, the solution reduced rack-scale network latency by approximately 4% and lowered the median optical engine temperature from 51°C to 36°C. NVIDIA claims the technology offers up to 5x better power efficiency and 15x higher reliability, though these are general architectural specifications rather than results from this specific deployment.