Stability issues hold back Huawei AI chips to outpace Nvidia: Report

Huawei AI chips are reportedly facing stability issues and this could be a piece of good news for Nvidia. China is seeking ways to reduce dependence on U.S. suppliers. But these efforts are being ruined by flawed software in Huawei’s chip core.

Financial Times notes that many Huawei customers complained about the buggy performance of Ascend AI chips, resulting in stability issues. An AI engineer stated that the processor frequently crashes leading to difficult and unstable work.

A major reason behind the stability issues of Huawei AI chips is its buggy software called “Cann”. The Huawei-developed software launched as an alternative to Nvidia’s Cuda. It was expected to help in enhancing the chip performance for developers.

But it seems the software has run into a strange bug that caused the AI chip to perform poorly in various training models. It also showed a slower inter-chip connection. Several Huawei employees also poured out their opinions and said:

“When random errors occur, it is very difficult to find out where it comes from due to poor documentation. You need talented developers to read the source code to see what the issue is, which slows everything down. The coding is imperfect.”

A Huawei researcher mentioned that bad results are a part of the journey and mostly occur when developers have the least knowledge of how the hardware works.

Stability issues hold back Huawei AI chips to outpace Nvidia: Report (Image Credits: Huawei)

Even after several attempts, the AI chip performance remained the same for developers. An AI engineer said that Huawei AI chips still lagged far from Nvidia due to stability problems. They aren’t contributing well to the initial training of models.

Solution:

Good news is that Huawei has started sending engineers that can help its customers to tackle the problem. The report reveals that experienced developers transfer the training code written on Cuda into Cann to ease the overall operation.

Baidu, Tencent, and iFlytek are the major companies that now have a new team of engineers to solve the AI chip-related problems. On this matter, a tech analyst said:

“Huawei’s advantage over Nvidia is it can work closely with its customers. Unlike Nvidia, it has a large team of engineers to help solve clients’ problems and get them to transition to their hardware.”

Apart from sending teams, the company also opened an online portal that can help customers with constant services and feedback to improve its buggy software.

The Chinese tech giant has a huge workforce that can help it resolve the AI chip problem to some extent. However, the overall scenario suggests that Huawei AI chips still have room for improvement. To hold back Nvidia, China needs more strength and technical support.

Huawei AI chipsets (Image Credits: Huawei)

[Source – Financial Times]