The China Internet Investment Fund (CIIF), a project jointly started by the State Cyberspace Administration of China and the Ministry of Finance, has recently provided cash to Shanghai-based startup Denglin. With this investment, NVIDIA hopes to assist Denglin in creating domestically marketed processors that are compatible with the CUDA and OpenCL programming models. Denglin intends to position itself as a competitor to NVIDIA in the GPU industry by obtaining this compatibility.
Established in 2017, Denglin plans to leverage its GPU+ technology, which offers a software-defined on-chip heterogeneous computing architecture. This technology will enable their chips to support both CUDA and OpenCL, providing flexibility for various applications. According to reports from JonPeddie, Denglin’s chips will primarily target the GPGPU (General-Purpose Graphics Processing Unit) market but can also find applications in more premium sectors such as high-performance computing (HPC) and artificial intelligence (AI).
Denglin has already unveiled a lineup of four products, catering to gaming and AI training needs. Among them, the Goldwasser chip stands out as a solution designed for AI-based compute acceleration, suitable for edge and cloud computing platforms. Additionally, the UL32 and UL64 chips offer power-efficient solutions with compute capabilities of up to 64 TOPs INT8 and 16 TFLOPs FP16, supporting up to 16 GB of memory and 64-channel video encoding at 1080P@30FPS.
The company announced that their second-generation GPUs completed tape-out in 2022 and are expected to enter mass production this year. These GPUs promise a 3-5 times performance improvement in transformer-based models, as well as significant reductions in hardware costs for workloads like ChatGPT and Generative AI.
The availability of a chipmaker offering CUDA programming capabilities and high-end GPUs for AI processing holds great significance for China’s domestic market. While NVIDIA and Intel strive to increase chip supplies within the country, the current accelerators are limited due to US-based restrictions. Additionally, the rising demand and higher costs have made it increasingly difficult to obtain these chips, with an H800 GPU priced above 250,000 RMB or over $35,000 US.
China currently boasts 13 GPU developers, and it remains to be seen whether Denglin can successfully bring its chips to market or face a fate similar to Biren Tech. Biren Tech had aspirations to become a major GPU manufacturer and compete against NVIDIA in the HPC segment but encountered obstacles due to US regulations, resulting in no substantial progress.
In conclusion, Denglin’s receipt of investment from the China Internet Investment Fund marks a significant step forward in their pursuit of developing CUDA and OpenCL compatible chips to challenge NVIDIA’s dominance. If successful, Denglin’s GPUs could play a crucial role in fulfilling the growing demand for high-performance GPUs in China’s domestic market, especially considering the restrictions on acquiring US-based accelerators.