Huawei Struggles to Replace Nvidia in Microchips

by time news

Huawei’s customers in China are complaining that its chip software performs poorly, falling short of Nvidia’s: can Beijing win the AI ​​race without computing power?

China’s efforts to catch up with the United States in AI computing power are running into problems, and the main culprit is Huawei, whose customers are complaining about performance and bugs within its microchip software.

Huawei is one of China’s most important technology companies and is central to the government’s plans to acquire the capacity to design and manufacture advanced microchips, such as those that power artificial intelligence systems: the dominant company in this market is the American Nvidia, whose products are however difficult to replace and also difficult for Beijing to access, given the restrictions introduced by Washington on the trade of semiconductors and machinery. chipmaking.

HUAWEI TECHNOLOGIES ARE NOT CONVINCING

Unable to access US technologies, Chinese groups investing in artificial intelligence have fallen back on Huawei’s Ascend microchips to train their linguistic models. The problem is that Huawei’s chips are not as high-performance as Nvidia’s: they have problems stability, are slower to connect with other electronic circuits and are managed by inferior and difficult to use software (it’s called Cann, developed by Huawei). In contrast, Nvidia’s software platform – Cuda – is highly regarded for its ease of use and for being able to accelerate data processing.

According to the testimonies collected by the Financial TimesHuawei’s software contains programming errors, crashes frequently, and slows down developers due to the complexity of the source code. To try to mitigate these problems, Huawei is reportedly sending skilled engineers to its customers — such as Baidu, Tencent, and iFlyTek — to assist them in porting previously written Cann code to Cann.

By being able to work more closely with Chinese customers, Huawei can be said to have an advantage over Nvidia in this specific case. The company has also set up a web portal where developers can submit comments or suggestions for improving the software.

PRICE AND AVAILABILITY

After the United States tightened export controls on semiconductors and related technologies last October, Huawei raised the price of its Ascend 910B microchip, the one used to train artificial intelligence, by 20-30 percent.

Unable to purchase the machinery chipmaking more modern, to manufacture the Ascend 910B Huawei found itself forced to re-adapt old devices, designed however for previous generations of chips, often causing them to break and raising concerns among customers about the continuity of supplies.

Nonetheless, demand for Huawei’s AI chips is strong, with the company reporting 34 percent revenue growth in the first half of the year.

“Taiwan Did Not Steal US Chip Industry”: Finance Minister Responds to Accusations of …

According to Politico, former ECB president and former Italian prime minister Mario …

Google, together with the Indian startup Salcit Technologies, is creating a new artificial intelligence technology …

A negative day at Piazza Affari at the beginning of the week for Leonardo, who leaves on the ground …

You may also like

Leave a Comment