9621 Agnes Crossing, Lake Suzanneview, New Mexico Island 84604-9295.
Chinese companies face challenges in switching from Nvidia chips to Huawei equivalents. Unclear software would lead to instability. Huawei is working on providing an alternative to Nvidia. Specifically,
Chinese companies face challenges in switching from Nvidia chips to Huawei equivalents. Unclear software would lead to instability.
Huawei is working on providing an alternative to Nvidia. Specifically, the company is working on Ascend accelerators that will have to compete with Nvidia’s (slightly older) portfolio. The accelerators are quite powerful on paper and show that Huawei is well on its way to becoming a real challenger to Nvidia’s dominance.
This is also necessary: Chinese companies do not have access to advanced Western AI accelerators such as those from Nvidia due to the American embargo rules. Huawei wants to offer a domestic alternative.
The Financial Times now knows that the transition from Nvidia chips to Ascend accelerators is going much less smoothly than hoped. Large Chinese companies such as Baidu and Tencent are relying on the chips, but anonymous internal sources reveal that the changeover is accompanied by many errors and crashes.
According to the sources, Huawei’s Cuda alternative is complex and unstable. The Cuda software layer makes Nvidia chips relatively easy to access for developers and is an important factor in Nvidia’s success. Huawei equips its Ascend chips with a software layer called Cann, but it is significantly more difficult to use.
The hardware crashes regularly, and engineers would be unclear as to why. Because of the lack of pre-written documentation, all too often engineers with domain knowledge must manually search through the source code to find the cause of a problem. When using Ascend and Cann, errors and crashes can all too easily occur.
Huawei has a lot of talent on its payroll. It is bringing engineering teams together with customers to accelerate the transition from Nvidia to Huawei Ascend. However, the gap between the Chinese software layer and that of Nvidia currently seems to be very large.
This slows down the adoption of AI hardware, which is actually the intention of the US embargo rules. The situation shows that it is complex to offer an alternative to Western chips. Without Western tools like ASML’s machines, not only is the design very difficult, but the mature ecosystem around Nvidia chips cannot be easily copied.
Source: IT Daily
As an experienced journalist and author, Mary has been reporting on the latest news and trends for over 5 years. With a passion for uncovering the stories behind the headlines, Mary has earned a reputation as a trusted voice in the world of journalism. Her writing style is insightful, engaging and thought-provoking, as she takes a deep dive into the most pressing issues of our time.
