Articles by tag "GPU cluster networking"

1. Liquid Cooling Ready: Mellanox Switches for 100kW+ Rack Density
   Selene Gong

   18
   As AI workloads surge in 2026, data center architects (DCAs) and GPUaaS providers in Singapore and APAC face a harsh reality: next-generation GPUs like Blackwell GB200 NVL72 push rack power densities beyond 140kW. In such extreme environments, traditional air-cooled switches cannot maintain ...
2. Why NVIDIA Spectrum-X is Overtaking InfiniBand in AI Data Center Architecture
   Selene Gong

   18
   For over a decade, InfiniBand has been the default networking fabric for high-performance computing (HPC) and early-stage AI clusters. Known for its ultra-low latency and lossless transport, it enabled tightly coupled GPU training workloads at scale. However, AI infrastructure is no longer ...