Huawei's Chip Struggles and the "Clustering" Solution
Huawei's creator discusses the robustness of their in-house processor technology.
In the midst of the ongoing trade spat between China and the U.S., Huawei's founder, Ren Zhengfei, lays bare the company's chip tech deficiencies, acknowledging that Huawei's individual chips are still a skip behind their American counterparts. Yet, Ren is quick to remind us that his company's not up against tough competition alone - many Chinese firms are developing chips and excelling at it, Huawei being one of them [source].
The U.S. has been using Huawei as its primary target, imposing sanctions, and casting the company as a significant threat. However, the game has taken a new twist, and Huawei, backed into a corner, has turned to focusing more on developing proprietary chips [source]. In a surprising twist of events, Huawei's efforts have made it serious competition for Nvidia, due to the U.S. regulations, hampering Nvidia's deliveries to China, and limiting their access to advanced AI chips [source].
Ren shares that Huawei is attempting to bridge the technology gap through a process called "clustering." By interlinking several chips, Huawei aims to bundle the computing power, thereby countering the weaknesses of individual chips [source]. The whole process goes down like this:
Clustering: Huawei's Secret Sauce
Huawei amasses multiple Ascend chips into systems, creating high-performance conglomerations akin to the CloudMatrix 384 [source][5]. This system synchronizes 384 Ascend 910C chips, allowing it to outpace or match some U.S. counterparts, such as Nvidia's GB200 NVL72 system [source][5].
Another layer includes advanced packaging techniques, where Huawei combines multiple chiplets into a single unit by stacking chips [source][5]. This approach helps achieve higher compute power without the necessity for the latest 3nm or 2nm fabrication nodes, largely controlled by U.S. allies like TSMC [source].
Lastly, the software landscape is fortified by Huawei's strong software ecosystems. The MindSpore framework, tweaked for Da Vinci architecture, supports major AI frameworks like TensorFlow and PyTorch, inviting more users and accelerating ecosystem growth [source].
Advantages and Up miserables
While Huawei grapples with lower chip yields and slower manufacturing processes compared to frontrunners such as TSMC, their clustering strategy presents opportunities for growth and innovation [source]. The approach could pave the way for fresh markets and draw investors who are keen to back China's AI infrastructure supply chain [source].
Shoutout to Huawei, playing the very best hand in a harsh global scene!
Note: While Huawei's "clustering" strategy enables them to skirt around U.S. restrictions on access to advanced chip fabrication tools, it does not fully bypass these limitations [source]. Additionally, the approach is not a magic wand, solving every single issue. But it sure is a heck of a smart workaround, George Bush style!
In the face of Huawei's chip deficiencies and US sanctions, the company is leveraging technology, particularly a process called "clustering," to compete with industry leaders like Nvidia. This strategy involves combining multiple Ascend chips to create high-performance systems, such as the CloudMatrix 384, which can match or outpace some US counterparts.
While Huawei faces challenges with lower chip yields compared to frontrunners like TSMC, the clustering strategy could open up new markets and attract investors interested in China's AI infrastructure supply chain. The approach is not a complete bypass of US restrictions but serves as a smart workaround, offering opportunities for growth and innovation in the face of a complex global situation.
