Recently, OpenAI launched a significantly upgraded image generation feature on its ChatGPT platform, attracting considerable user attention and usage. However, this popular feature has presented OpenAI with new challenges. OpenAI founder Sam Altman revealed that the surging demand has nearly overwhelmed the company's GPU computing power, stating figuratively that the "GPUs are smoking," forcing them to implement rate limits on the image generation feature.
The new feature, "Images in ChatGPT," is built upon the GPT-4o model. Users can generate and edit images using simple natural language instructions and iterate through multiple rounds of optimization. The launch of this feature marks a deeper integration of multimodal capabilities—text, images, and code—in ChatGPT, transforming it into a true all-purpose AI entity. Users can easily input instructions to transform ordinary photos into Ghibli-style artwork, a convenience that has quickly sparked widespread enthusiasm.
However, with the surge in users, OpenAI realized that its initial plan to roll out this feature to all users, especially free users, needed to be delayed. Altman explained that despite possessing substantial GPU resources, handling such a massive volume of requests proved overwhelming.
Image generation requires AI to perform pixel-by-pixel calculations, demanding significant computing resources. To address the GPU bottleneck, OpenAI is exploring two main avenues: improving GPU performance and optimizing algorithms to enhance AI model efficiency. The latter aims to improve algorithms so that the same computing resources can handle more tasks, leading to more efficient resource utilization.
It's noteworthy that OpenAI boasts top-tier GPU reserves. Microsoft, as a major investor, provides strong support to OpenAI, purchasing a large number of Nvidia Hopper chips, giving it a competitive advantage in computing power. Despite this, the demand for computing power from AI applications remains enormous, and the industry continues to call for technological iteration and improved resource utilization efficiency.
OpenAI is facing a balancing act between resources and demand in its development of multimodal technology.
