The latest report released by the market research agency TrendForce TrendForce pointed out that if the processing power of the Nvidia A100 graphics card is calculated, running ChatGPT will need to use 30,000 Nvidia GPUs.
The survey agency TrendForce pointed out in the report that it is estimated that ChatGPT needs 20,000 graphics cards to process the training data. As OpenAI continues to commercially deploy ChatGPT and the company’s generative GPT models, more than 30,000 graphics cards may be required.
The current price of the Nvidia A100 graphics card is between 10,000-15,000 US dollars, depending on how many graphics cards OpenAI needs and whether Nvidia will provide order discounts for AI companies. TrendForce estimates that Nvidia can earn more than US$300 million (currently about 2.082 billion RMB) in revenue.
Nvidia also sells the A100 as part of the DGX A100 system, which has eight accelerators and costs a whopping $199,000. Given the scale of OpenAI’s operations, the company may buy the A100s individually and stack them into clusters.
