As soon as Nvidia CEO Jensen Huang unveiled the world’s highest-spec ‘Blackwell’ artificial intelligence (AI) server on the CES keynote, a photograph of the particular product appeared on social media. This time, Microsoft (MS) is the protagonist.
MS CEO Satya Nadella released a photograph of NVIDIA’s ‘NVlink 72’ cluster through X (Twitter) on the eighth (local time) and announced that it had been applied to the ‘Azure’ service. A number of hours after CEO Hwang revealed this within the keynote, CEO Nadella was visiting India.
He said, “Now we have launched the primary NV Link 72 cluster live to tell the tale Azure in keeping with testing time scaling,” and added, “We are going to construct the following generation AI based on this technique.”
NV Link is an AI supercomputer that increases processing speed by introducing the newest networking technology to the Blackwell-based ‘GB200’ server. By connecting 72 Blackwell GPUs and 36 Grace CPUs, the 72 GPUs act as one giant computing chip. Particularly, it provides a two-way bandwidth of 1.8 terabytes per second, which is twice that of the previous generation.
CEO Hwang introduced within the CES keynote, “It weighs 1.5 tons and is made up of 600,000 parts, which is reminiscent of 20 cars.” It was also revealed that 120 kilowatts (kW) of power was consumed, and three.2 kg of copper cable and 5,000 cables were used. Particularly, he emphasized that “the manufacturing process is unbelievable.”
As is well-known, Microsoft is Nvidia’s best customer. In response to an evaluation by global market research firm Omdia, 485,000 units of Nvidia’s flagship chip ‘Hopper’ were purchased last 12 months. This corresponds to twenty% of all Hopper GPUs sold, which is greater than double that of other corporations. Due to this fact, I used to be the primary to receive the brand new product.
Last October, Microsoft boasted that it was the primary to construct Blackwell servers. At the moment, the self-built NVIDIA ‘GB200’ server was released through the official Azure X account.
This NV Link can be seen as getting used to construct the Open AI model. OpenAI has recently been known to be having difficulty improving the performance of GPT-5, also often called ‘Orion’, and a focus is being paid to what results it can produce using the most effective supercomputer in existence.
Reporter Lim Da-jun ydj@aitimes.com