Chip war: Chinese start-up aims to break Nvidia’s grip on AI with new model framework
A new AI framework called Chitu, developed by teams associated with China’s Tsinghua University, aims to reduce reliance on Nvidia chips for AI model inference. This development marks a significant step towards enhancing China's technological self-sufficiency in the face of US export controls on advanced GPUs.
Chitu, designed for large language models (LLMs), can operate on domestically produced Chinese chips, challenging Nvidia's dominance. Tests using Nvidia's A800 GPUs showed that Chitu achieved a 315% increase in model inference speed and a 50% reduction in GPU usage compared to foreign open-source frameworks when running the DeepSeek-R1 model. The open-source framework supports mainstream models, including those from DeepSeek and Meta's Llama series, providing a viable alternative amid restrictions on Nvidia's high-performance GPUs like the H100 and H800 in China.
via SCMP Full Text Feed
A new AI framework called Chitu, developed by teams associated with China’s Tsinghua University, aims to reduce reliance on Nvidia chips for AI model inference. This development marks a significant step towards enhancing China's technological self-sufficiency in the face of US export controls on advanced GPUs.
Chitu, designed for large language models (LLMs), can operate on domestically produced Chinese chips, challenging Nvidia's dominance. Tests using Nvidia's A800 GPUs showed that Chitu achieved a 315% increase in model inference speed and a 50% reduction in GPU usage compared to foreign open-source frameworks when running the DeepSeek-R1 model. The open-source framework supports mainstream models, including those from DeepSeek and Meta's Llama series, providing a viable alternative amid restrictions on Nvidia's high-performance GPUs like the H100 and H800 in China.
via SCMP Full Text Feed
