The Gaudi2 deep learning processor has double the performance of Nvidia’s A100 GPUs, which are banned in China.
American semiconductor giant Intel Corp has introduced its new artificial intelligence (AI) deep learning processor to the Chinese market. This processor is not subject to US export restrictions, unlike the products of Intel’s competitor – the company Nvidia Corp, whose graphics processing units (GPUs) have been smuggled into China due to the huge demand for them.
At a press conference in Beijing on Tuesday, Intel showed off its Gaudi2 processor, which they say is an alternative to Nvidia’s premium A100 GPU, widely used for AI training.
The latest move by Intel – a company that generated 27% of its total 2022 revenue in China, according to its latest annual report – highlights the continued importance of the huge Chinese market for U.S. semiconductor technology suppliers despite Washington’s export restrictions.
It follows Nvidia’s attempts earlier this year to promote modified versions of its flagship A100 and H100 GPUs in China to comply with U.S. restrictions and support shipments to its customers in mainland China, where new AI development projects are flourishing, building services similar to OpenAI’s ChatGPT.
Last August, the US Department of Commerce banned Nvidia and Advanced Micro Devices (AMD) from selling some advanced chips to China. Nvidia has been instructed to stop selling its A100 and H100 GPUs and AMD to export its MI250 chips to the mainland.
This has left the Chinese AI sector scrambling to acquire limited stocks of cutting-edge chips from Nvidia, which has a near monopoly on GPUs used to train AI systems.
Strong demand has even created a booming market for bootleg GPUs like Nvidia’s A100 and H100 devices.
The A100 Tensor Core GPU powers the world’s leading data centers used for AI applications, data analytics, and high performance computing, according to Nvidia’s website.
For Intel, demand in China reinforces its commitment to providing customers with “a wide choice of hardware solutions,” Sandra Rivera, executive vice president of the company, who is also general manager of Intel’s data center and AI group, said at a press conference in Beijing.
She said that Gaudi2 was designed to lower the barrier to entry and enhance the ability of her Chinese customers to “implement AI with cloud and intelligence at the edge of the network, helping to create the future of AI in China.”
Intel announced that it is partnering with Inspur Group – the world’s second largest AI server maker based in east China’s Shandong province – to build new Gaudi2-based machines for the mainland market.
Habana Labs, Intel’s data center team specializing in AI deep learning processor technologies, originally launched Gaudi2 last May in the US, where it claimed that the processor’s learning performance was twice that of the A100 GPU. from Nvidia for 80 gigabytes for the ResNet-50 computer vision model and the BERT natural language processing model.
Intel’s latest proposal comes at a time when China could face new challenges in its AI ambitions.
The US government is reportedly considering restricting Chinese companies’ access to US cloud computing services, which would end Amazon Web Services and Microsoft Corp’s use of advanced AI chip power in favor of their mainland Chinese customers, according to a report by The Wall Street Journal earlier this month. .
The US has also considered stepping up its action and putting Nvidia’s modified A800 GPU on its export ban list, according to a report by The Wall Street Journal last month.