Intel, a leading semiconductor company, has announced the introduction of its Habana Gaudi2 accelerators for AI and HPC (High-Performance Computing) in the Chinese market.
Intel, a leading semiconductor company, has announced the introduction of its Habana
Gaudi2 accelerators for AI and HPC (High-Performance Computing) in the Chinese
market. With the demand for AI technologies surging in China, the Habana Gaudi2
accelerator offers an alternative to NVIDIA’s A800 chips.
Intel claims that the Gaudi2 chip provides faster training performance than NVIDIA’s
Ampere A100 GPU. Its benchmark tests showing a significant 69% performance gain.
This move comes in the context of the US seeking to tighten export controls on AI chips
to China.
During a press conference held in Beijing, Intel unveiled its deep learning processor, the
Habana Gaudi2. The chip is designed to enhance AI training greatly and inference tasks.
Several prominent Chinese AI server manufacturers are expected to incorporate the
Gaudi2 processor into their server products, including: Inspur, New H3C, and xFusion.
Rising to Meet Demands
Recognizing the increasing popularity of Generative AI, Intel has collaborated with BCG
(Boston Consulting Group) to utilize the Gaudi2 accelerators in Generative AI workloads.
As NVIDIA dominates the AI market, the demand for GPUs capable of handling AI tasks
has outpaced supply, leading to soaring prices.
Additionally, a potential “Extended” chip ban on China could exacerbate the supply
issues further. In response, Intel aims to capitalise on this opportunity and challenge
NVIDIA’s dominance by bringing the Gaudi2 accelerators to China.
To cater to the specific needs of the Chinese market, Intel plans to release a custom
version of the Gaudi2 accelerator optimised for large language models (LLMs), similar
to NVIDIA’s custom solution for China.
By partnering with prominent HPC companies like Inspur and H3C, which collectively
hold a significant market share in China, Intel aims to make significant headway in the
AI GPU market.
However, reports suggest that the Gaudi2 accelerator hasn’t gained significant traction
within the year since its launch, with only one major deal with BCG. Nonetheless, Intel
remains committed to expanding its offerings and intends to introduce the next-
generation Gaudi3 accelerators and GPU-based AI solutions by 2025.
Measuring up to Its Competitor
The move by Intel follows a similar initiative by Nvidia, which introduced modified
versions of its A100 and H100 GPUs in China to comply with US restrictions while
maintaining a steady supply for its customers.
The Chinese market holds great significance for US semiconductor technology
providers, as evidenced by Intel’s commitment to serving its needs. China accounts for
27% of the company’s total revenue in 2022, according to its annual report.
Last year, the US Department of Commerce imposed a ban on the sale of certain
advanced chips to China, affecting companies like Nvidia and AMD. Nvidia was directed
to halt the sale of its A100 and H100 GPUs.
Meanwhile, AMD faced restrictions on exporting its MI250 chips to mainland China.
These measures were implemented due to trade regulations on advanced technologies.
In light of the intense competition in the tech war between Intel and Nvidia, both
companies are striving to offer purpose-built chips for training AI systems in China.
Intel’s Gaudi2 processor, unaffected by US restrictions, is positioned as a direct rival to
Nvidia’s A100 GPU. The importance of the Chinese market to US semiconductor
technology providers remains evident. As China’s AI sector continues to flourish with the
development of AI-based services, the demand for advanced chips has grown
exponentially, leading to a thriving market for smuggled GPUs in China.
Intel’s commitment to providing a diverse range of hardware options to clients further
solidifies its position in the Chinese market. Sandra Rivera, Executive Vice President of
Intel’s Data Center and AI Group, emphasised this commitment during the press
conference in Beijing.