Intel has quietly launched its Gaudi 2 processor for artificial intelligence (AI) workloads in compliance with the latest export regulations for China. The new version has reduced scale-up interconnects and is already being used in Inspur’s NF5698G7 server. This machine is primarily designed for training and inference using Large Language Models (LLMs). This will help fill some of the shortages in computing GPUs like Nvidia’s A100 and H100, as well as the China-bound cut-outs.
Intel has two Gaudi 2 versions offered in the 600W OCP model form factor.called normal version Gaudi 2 HL-225H has 24 100Gbps RDMA (RoCE v2) interconnects, while a reduced version of the product is Gaudi 2 HL-225B There are 21 100Gbps RDMA (RoCE v2) scale-up interconnects. The reduced number of scale-up interfaces will lead to lower performance of AI servers based on this version of Gaudi 2, but bearing in mind the huge demand for AI capabilities in China, Inspur’s NF5698G7 finally prevailed. There is likely to be.
‘processor is compliant US BIS regulationFor supercomputers and semiconductors, Intel describes a version of its Gaudi 2 accelerator for the Chinese market.
The availability of Intel’s Habana Gaudi 2 in China will ease the shortage of GPU servers, but it should be noted that applications already using Nvidia’s GPUs are unlikely to be ported to Gaudi. I have. Existing customers should therefore continue to purchase from his Nvidia.
Digi Times A shortage of computing GPUs, coupled with the high cost of advanced GPUs, has forced Inspur to raise prices, according to reports citing Chinese media. In addition, Inspur’s LLM application, which focuses on small-scale industrial scenarios and emphasizes AI conversation, naturally stands out as a differentiator compared to the full-fledged AI Generated Content (AIGC) application, which is Inpur’s AI server. Lack of competitive advantage.
Inspur has a strong market position in China’s AI server space, but a partnership with Intel could affect the company’s position in China’s AI chip market and shape competition with Nvidia.