NVIDIA-Certified Professional AI Rack and Interconnect

A data center architect is designing infrastructure for training very large language models across many GPUs and needs a high-speed interconnect that enables direct multi-GPU communication within the scale-up domain. Which NVIDIA technology is specifically described as providing this capability?

A team is standardizing on a networking platform to accelerate AI training and inference traffic and wants performance improvements over conventional Ethernet. According to NVIDIA, what performance advantage does the Spectrum-X networking platform provide over traditional Ethernet?