Nvidia releases a new generation of powerful GPUs supporting 10 trillion parameter AI large model training and inference
阿豆学长长ov
发表于 2024-3-19 09:51:18
251
0
0
At 4am Beijing time on March 19th, the NVIDIA GTC 2024 Developer Conference was held, and NVIDIA founder and CEO Huang Renxun announced the new generation AI graphics processor architecture Blackwell. The Blackwell architecture based GPU has 208 billion transistors, while the previous generation H100 GPU had 80 billion transistors.
It is reported that the Blackwell architecture adopts six revolutionary technologies, including powerful chips, second-generation Transformer engine, and fifth generation NVLink, to support AI large model training and inference with up to 10 trillion parameters.
Nvidia currently upgrades its GPU architecture every two years, significantly improving its performance. The H100 GPU based on Hopper architecture was released in 2022.
CandyLake.com 系信息发布平台,仅提供信息存储空间服务。
声明:该文观点仅代表作者本人,本文不代表CandyLake.com立场,且不构成建议,请谨慎对待。
声明:该文观点仅代表作者本人,本文不代表CandyLake.com立场,且不构成建议,请谨慎对待。
猜你喜欢
- The most expensive and cheapest models will be launched on the same day! Li Bin: NIO will double its efforts to become one of the top ten global car companies
- Fali Store Building Wal Mart's New Generation Store Strengthening Scene Content
- Attraction crushing wide base index! Retail investors net purchase $29.8 billion worth of Nvidia stocks in 2024
- Nvidia New Product Countdown: New 'Nuclear Bomb' RTX 5090 Coming Soon, B300 Coming Soon
- Over 210 billion yuan in explosive purchases! Retail investors' fierce pursuit 'of Nvidia investment bank, optimistic about next year's performance
- NVIDIA's new 'nuclear bomb' leaked!
- NVIDIA's latest statement! Robot 'ChatGPT Moment' is Coming, Bet on the Next Growth Driver
- How will Google respond under stricter regulation in a more competitive track? CEO: Focus on Gemini model next year
- Nvidia may launch robot 'brain' in the first half of next year, with the company's stock price increasing by over 176% since the beginning of this year
- Nvidia plans to release a new generation of humanoid robot computing platform in the first half of next year, supporting multimodal AI models