NVIDIA (NVDA.US) announced the latest data on Wednesday, indicating that its latest artificial intelligence servers can enhance the computational performance of several new models (including two popular models from China) by up to 10 times.
This data primarily focuses on the "expert mixture" AI model. This technology improves efficiency by breaking down problems and assigning them to different experts within the model for processing.
On the other hand, NVIDIA emphasized that even with a reduction in the chips required for training related models, its products remain competitive in model inference deployment. Its latest AI server is configured with 72 top-tier chips, interconnected at high speed, allowing the performance of Moonshot's Kimi K2 Thinking model to improve by 10 times compared to the previous generation NVIDIA server; related models from DeepSeek also saw similar performance increases