Biren prioritizes AI breakthrough; software stack key to Chinese GPU industry

Jack Wu, DIGITIMES Asia, Taipei 0

Credit: Biren Technology

Chinese chip startup Biren Technology has selected AI to be its development focus, meaning the company's focus will be on general-purpose GPUs (GPGPUs) used in AI and its corresponding software stack. The presence of the US chip restrictions is also well-noted by China's GPU sector, with each company taking its own approach to circumvent the restrictions.

The US chip restrictions toward China have resulted in many Chinese GPU startups springing up to take advantage of the market void left by US companies. Generally, these companies have to choose between two routes: GPGPU or GPU for image rendering. Biren Technology has opted for the former route, alongside the likes of Denglin Technology, MetaX, and Iluvatar CoreX.

Biren co-CEO Xinrong Li stated to JW Insights that while GPGPUs and image-rendering GPUs have different technical challenges and market traits, they both require a significant investment of talent and capital resources. For a startup in its early stages, taking on both routes at the same time is quite challenging. Biren has chosen to focus on GPGPU and prioritizing breakthroughs in AI, an application with greater market potential.

Data center products have already surpassed traditional GPU products on the market. GPU leader Nvidia's latest financial report showed that in the first three quarters of 2022, the sales revenue for games and data centers is US$7.2 billion and US$11 billion respectively. Nvidia has also been increasingly referring to itself as a computing company on many occasions.

Jianwen Li, founder and chairman of Denglin Technology, pointed out that focusing on GPGPU allows more control over its product cycle and the ability to change clients. AI computing, for instance, is an ever-changing market. This kind of change allows startups to break through. In contrast, the image-rendering GPU market is very stable and is already dominated by major international brands, making it difficult for startups to survive.

With the hardware direction selected, the next step for GPU makers is to address application and sales issues like software adaptations and product launches. Biren president Lingjie Xu told JW Insights that "the key to determine a qualified GPGPU product is its launch performance, and the software stack is the core element of a product launch."

Industry sources pointed out that software adaptation is a problem facing the entire industry and that almost 90% of AI applications in the Chinese software ecosystem are based on solutions from mainstream foreign suppliers. This has become a major annoyance for China's GPGPU development.

According to an industry insider, as things stand now, Chinese GPU products can only survive in non-commercial fields with small market potential because "the products of international companies are cheaper with better performance and reliability."

Supply chain issues are also something every Chinese GPU maker needs to consider due to the ongoing US-China conflict. Li stated that GPU performance will be directly affected in the short term. Companies need to figure out a way to improve "product competitiveness" through architectural innovation while also having high compatibility.

He also believes that current US restrictions are formulated according to Nvidia's product specifications, meaning that one needs to meet the relevant I/O and theoretical computing power to be impacted. This also shows that copying Nvidia and other commercially available IPs will not only make you unable to surpass existing products in cost-effectiveness but also make the impact of the restrictions worse.

In a growing GPU market, Chinese companies need to find a new incremental market and perform differentiated innovation to grasp long-term opportunities.