Shulou(Shulou.com)11/24 Report--

CTOnews.com February 19 news, after the popularity of ChatGPT around the world, technology giants have entered the bureau, generative AI behind the large model-based artificial intelligence has become the direction of investment in the industry.

CTOnews.com learned that the so-called "big model" is usually trained by self-supervised learning on the unlabeled big data set. Later, in the application of other scenarios, developers only need to fine-tune the model, or use a small amount of data for secondary training to meet the needs of new application scenarios.

However, the training of general-purpose models is very expensive. According to Guosheng Securities report "how much arithmetic does ChatGPT need", the cost of a GPT-3 training is about $1.4 million, and for some of the larger LLM (large language models), the training cost is between $2 million and $12 million. Based on the average of 13 million unique visitors of ChatGPT in January, the corresponding chip demand is more than 30,000 Nvidia A100 GPU, with an initial input cost of about $800 million and a daily electricity bill of about $50,000.

If you deploy the current ChatGPT to each search conducted by Google, you need 512820.51 A100 HGX servers and a total of 4102568 A100 GPU, the total cost of these servers and networks is more than $100 billion in capital expenditure alone.

Guosheng Securities believes that in the public cloud, for global technology giants such as Google, million-to-tens-million-dollar training is not cheap, but it is still acceptable and not expensive.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.