US EditionDeepSeek hints latest model will be compatible with China’s ‘next generation’ homegrown AI chips
DeepSeek's V3.1 AI model uses a new precision format to enhance training efficiency by reducing memory use up to 75%, aligning with China's push for self-sufficient AI technology.
- DeepSeek, a Hangzhou-based AI startup, released its V3.1 model on Thursday, designed for upcoming Chinese home-grown AI chips.
- This release follows U.S. export restrictions in April that banned Nvidia’s advanced H20 chips, prompting China to urge domestic chip development.
- V3.1 integrates a hybrid reasoning architecture supporting reasoning and non-reasoning modes, trained with the UE8M0 FP8 8-bit data format to enhance efficiency and compatibility.
- DeepSeek said the UE8M0 FP8 format could cut memory use by up to 75%, increase training speed, and is tailored for soon-to-launch Chinese chips.
- This development signals Chinese progress toward AI chip-stack self-sufficiency, potentially easing reliance on restricted U.S. imports amid ongoing geopolitical tensions.
6 Articles
6 Articles
Chinese AI startup DeepSeek releases upgraded model with domestic chip support
Chinese artificial intelligence startup DeepSeek released on Thursday an upgrade to its flagship V3 model that the company says has a feature that can optimize it for Chinese-made chips, along with faster processing speeds.
Deep Thinking: DeepSeek V3.1 Adds Chip Support, Hybrid AI Modes
V3.1 version offers faster performance, hybrid inference modes, and strategic alignment with China’s push for tech self-sufficiency DeepSeek… The post Deep Thinking: DeepSeek V3.1 Adds Chip Support, Hybrid AI Modes appeared first on IMP.NEWS.
Deepseek Hints The newest model will be compatible with the 'next generation' of China in their own soil AI chips
Anthony Kwan | Getty Images News | Getty images Chinese artificial intelligence Startup Deepseek has hinted that China will soon have “next generation” of home -grown chips to support his AI models, while announcing an update of one of the great language models. In a comment under a message about his official WeChat account, Deepseek […]
Coverage Details
Bias Distribution
- 100% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
