据市场调查机构Factorial Funds最新报告显示，OpenAI在推进其先进的文本转视频模型Sora的部署过程中，可能面临巨大的硬件需求。该机构估计，在模型运行的峰值阶段，OpenAI将需要多达72万张英伟达的顶级H100 AI加速卡，这一数字对应的价值高达216亿美元。这无疑是对计算资源的一次巨大挑战，也反映出高精度人工智能模型开发背后的资本密集型特性。
Factorial Funds的报告指出，为了训练Sora模型，OpenAI在一个月内的消耗可能在4200到10500片H100加速卡之间。这意味着，即使使用单片H100加速卡，生成短短1分钟的视频也可能需要长达12分钟的渲染时间。这种计算效率对于大规模的内容生成应用而言，既是技术上的考验，也是效率上的挑战。
考虑到英伟达H100加速卡的高性能和高价位，OpenAI在实现Sora的商业化应用过程中，不仅要解决技术瓶颈，还需要在成本控制和计算效率提升上找到平衡点。这一预测引发了业界对AI技术发展成本与效益的深度思考，同时也预示着未来AI领域可能将迎来一轮新的硬件升级竞赛。
英语如下：
News Title: “OpenAI Develops Text-to-Video Model Sora, Estimated to Consume 720,000 NVIDIA H100 GPUs Worth $216 Billion”
Keywords: OpenAI, Sora model, NVIDIA H100
News Content: According to the latest report from market research firm Factorial Funds, OpenAI may face substantial hardware requirements in deploying its advanced text-to-video model, Sora. The agency estimates that at peak operation, OpenAI could require as many as 720,000 top-of-the-line NVIDIA H100 AI accelerator cards, a figure equivalent to a staggering $216 billion. This poses a significant challenge to computational resources and underscores the capital-intensive nature of developing high-precision AI models.
Factorial Funds’ report suggests that for the training of the Sora model, OpenAI’s consumption could range between 42,000 and 105,000 H100 accelerator cards within a month. This implies that even with a single H100 accelerator, generating a one-minute video might necessitate a rendering time of up to 12 minutes. Such computational efficiency poses both a technological test and an efficiency challenge for large-scale content generation applications.
Considering the high performance and premium pricing of the NVIDIA H100 GPUs, OpenAI will not only have to overcome technological hurdles in commercializing Sora but also strike a balance in cost control and boosting computational efficiency. This projection has sparked industry-wide contemplation on the cost-benefit dynamics of AI technology development and foreshadows a potential new round of hardware upgrading race in the AI sector.
【来源】https://www.ithome.com/0/758/413.htm
Views: 1