Chinese GPU Maker Receives Investment To Develop OpenCL & CUDA Compatible Chips To Tackle NVIDIA

A Chinese GPU startup has received investment from the country to develop chips that feature compatibility with NVIDIA's CUDA & OpenCL programming models.

China Startup On Its Way To Develop NVIDIA Rivaling & CUDA-Compatible GPUs After Securing Funding From China Internet Investment Fund

The Shanghai-based startup, Denglin, was founded in 2017 and has now received funding from the CIIF (China Internet Investment Fund) which was also initiated jointly by the State Cyberspace Administration of China and the Ministry of Finance to develop chips for the domestic market that are both CUDA and OpenCL compatible. The chips will primarily be used for GPGPU usage but can be further used in more premium markets such as HPC and AI after their development succeeds.

Denglin will be developing their chips on the GPU+ technology which is said to offer a software-defined on-chip heterogeneous computing architecture and will support both CUDA and OpenCL, reports JonPeddie.

As for the product lineup, the China-based GPU maker initially unveiled four products which ranged from gaming to AI training. The top solution is the Goldwasser chip which is designed for AI-based compute acceleration and will be getting edge and cloud computing platforms. The UL32 and UL64 are 10-15W solutions, offering up to 64 TOPs INT8 and 16 TFLOPs FP16 compute with support of up to 16 GB memory and 64-channels video encoding at 1080P@30FPS.

The company states that its 2nd generation GPUs were taped out in 2022 and mass production is expected this year. The new GPUs will deliver a 3-5 time performance gain in transformer-based models and greatly reduce the hardware cost in workloads such as ChatGPT and Generative AI.

Specification	Goldwasser UL32	Goldwasser UL64
Power consumption	10W	15W
Interface	MXM3.1	MXM3.1
INT8	32 TOPS	64 TOPS
FP16	8 TFLOPS	16 TFLOPS
Video decoding (H.264/H.265)	32 channels, 1080p@30 FPS	64 channels, 1080p@30 FPS
Memory (configurable)	4G/8G/16G	4G/8G/16G

A chipmaker that offers CUDA programming capabilities and high-end GPUs for AI processing can be a big deal for China's domestic market. While NVIDIA and Intel are both pushing to supply more chips within the country, the accelerators themselves are cut-down variants of what is available in other regions due to US-based restrictions. Furthermore, it is becoming harder to acquire these chips due to increasing demand and higher costs with an H800 GPU priced above 250,000 RMB or over $35,000 US.

Currently, there are 13 GPU developers in China and we will have to wait and see whether Denglin can actually get these chips out in the market or if they would end up with the same fate as Biren Tech who were poised to become a major GPU manufacturer and compete against NVIDIA within the HPC segment but ultimately saw no progress due to US regulations.

Written by Hassan Mujtaba