China is making great efforts to compete with US computing power in artificial intelligence. Huawei Ascend’s artificial intelligence chips are already widely used by Chinese companies, but users complain of performance problems and difficulties switching from Nvidia’s products.
Last October, Washington further tightened export controls on high-performance silicon, which is why a big competition started in China to develop a domestic alternative to industry leader Nvidia. In the fight, Huawei has become the leader, the Ascend series has become an increasingly popular option for Chinese AI groups to run models. However, according to several industry insiders, including an AI engineer at a partner company, the chips are still far behind Nvidia’s chips in initial model training. Stability issues, slower chip-to-chip connectivity and inferior software called Cann, developed by Huawei, were blamed.
Nvidia’s Cuda software platform is known as the company’s “secret ingredient” for its ease of use for developers and its ability to accelerate data processing. Huawei is one of many companies trying to break Nvidia’s stranglehold on AI chips by creating alternative software. However, it is not easy, Huawei’s own employees are also among those who complain about Cann. One researcher, who did not want to be named, said it made it “heavy and unstable” and hindered work related to testing. “In the case of random errors, it is very difficult to find out their origin due to poor documentation. Talented developers are needed, because you can only find errors by reading the source code, which slows everything down. The coding is weak,” they said.
Another Chinese engineer said the chips often froze, making AI development work difficult. “It’s easy to get bad results because people don’t know much about the hardware itself.” To solve the problem, Huawei sent engineers to the customers to help on the spot in transferring the training code previously written for Cuda to Cann. Baidu, iFlytek and Tencent are among the tech companies that have received engineering teams. A former Baidu employee confirmed that “Huawei’s customer service is outstanding, so of course they have engineers at their big customers to help them use their chips.”
Huawei can speed up the switch by deploying a lot of people. The company says more than 50 percent of its 207,000 employees work in research and development, including engineers sent to install the technology at customers. “Huawei’s advantage over Nvidia is that it can work closely with customers,” said Tilly Zhang, a technology analyst at consulting firm Gavekal. “Unlike Nvidia, it has a large engineering team to help solve customers’ problems and get them to switch. to their hardware.” Huawei has also created an online portal for developers to provide feedback on how to improve its software.
Huawei is seeing strong demand for its AI chips: last week it reported a 34 percent rise in first-half revenue, without disclosing sales broken down by business. He is taking advantage of his position, because after the United States tightened export controls in October, he raised the price of the Ascend 910B chip used for AI training by 20-30 percent. Huawei’s customers have also expressed concern about supply constraints for the Ascend chip, likely stemming from production difficulties as Chinese companies are unable to buy advanced chip-making machinery from Dutch company ASML. Huawei CEO Zhang Ping’an said at the World Artificial Intelligence Conference in Shanghai in July that more than 50 basic models have been “trained and iterated” on the Ascend chip.
Source: sg.hu
