Pure Domestic GPU Performance Comparison: Who Is the King of Domestic Computing Power? - Zhihu

The latest analysis on Zhihu has sparked a heated debate about which domestic GPU truly reigns supreme in computing power. Chinese companies like Moore Threads and Biren Technology are pushing their hardware to compete with NVIDIA’s older architectures, but the gap remains significant in raw performance.

Benchmark results show that Moore Threads’ MTT S80 delivers impressive performance for AI inference tasks, rivaling NVIDIA’s RTX 3060 in specific workloads. However, when it comes to complex training scenarios and double-precision floating-point calculations, domestic chips still lag behind their Western counterparts by at least two generations.

The real story here is not just about hardware specs but the growing software ecosystem around these Chinese GPUs. Developers are increasingly porting popular AI frameworks like PyTorch and TensorFlow to support domestic chips, making them more practical for everyday use.

China’s push for self-sufficiency in computing power is paying off in niche applications like edge computing and government cloud services. Yet, for high-end scientific computing and advanced AI research, the dominance of NVIDIA’s CUDA ecosystem remains a formidable barrier.

What makes this competition fascinating is the unique approach of Chinese companies, who are optimizing their GPUs for specific domestic needs like video transcoding and industrial automation. This targeted strategy could help them carve out profitable markets even without matching the top-tier performance of Western giants.

Ultimately, the king of domestic computing power is still being crowned, with no clear winner yet. The race is not just about speed but about building a complete ecosystem that can support China’s ambitious technological goals.