Nvidia has unveiled details of a new networking platform designed for generative AI workloads that promises lightning-fast and lossless networking.
The company’s new Spectrum-X technology, built using Spectrum-4 Ethernet switches and BlueField-3 DPUs, promises a 1.7x increase in performance and power efficiency.
in a Press release“The delivery of end-to-end capabilities reduces the runtime of Transformer-based generative artificial intelligence models at scale,” which in turn enables companies to operate and make decisions faster, unlocking cost-saving potential, the company said.
Spectrum-X Ethernet
“Spectrum-X is a new type of Ethernet networking that removes barriers to next-generation artificial intelligence workloads that have the potential to transform entire industries,” said Gilad Shainer, Nvidia’s senior vice president of networking.
Its key component, the Spectrum-4 Ethernet switch, is capable of speeds of 51 Tb/s. What this means for consumers is that the combination of the BlueField-3 DPU and Nvidia LinkX optics can enable 400 Gigabit Ethernet networking.
The company wanted to gain insight into the powerful networking capabilities that AI supercomputers can now leverage in real-world use cases:
“Nvidia Spectrum-X enables unprecedented scale of 256 200Gb/s ports connected by a single switch, or 16,000 ports in a two-tier leaf-spine topology, to support AI cloud growth and scaling while maintaining high levels of performance And minimize network latency.”
Elsewhere, Nvidia CEO Jensen Huang showed off the Spectrum-4 switch chip in his opening keynote at Computex, sharing more insights into its design. The 100 billion transistors on a 90x90mm chip consume about 500 watts, he said.
Nvidia is already testing its Spectrum-X, a hyperscale generative AI supercomputer running Dell PowerEdge XE9680 servers based on the Nvidia HGX H100 eight-GPU platform, at its Israel data center in Israel 1.
