07/30/2024, 08:06 PM UTC
苹果放弃使用英伟达GPU进行AI模型训练，转而采用数千个谷歌TPUApple skips Nvidia's GPUs for its AI models, uses thousands of Google TPUs instead
❶ 苹果选择使用谷歌的TPU而非英伟达的GPU来开发其AI模型。❷ 苹果基础模型（AFMs）采用谷歌TPUv4和TPUv5芯片进行训练，其中AFM-server使用了8,192个TPUv4芯片，而AFM-on-device则使用了2,048个TPUv5p芯片。❸ 苹果的详细研究论文揭示了公司在AI透明度和性能优化方面的努力。❶ Apple has opted to use Google's TPUs instead of Nvidia's GPUs for developing its AI models. ❷ The Apple Foundation Models (AFMs) were trained using Google TPUv4 and TPUv5 chips, with AFM-server utilizing 8,192 TPUv4 chips and AFM-on-device using 2,048 TPUv5p chips. ❸ Apple's detailed research paper reveals the company's efforts in AI transparency and performance optimization.
---
本文由大语言模型（LLM）生成，旨在为读者提供半导体新闻内容的知识扩展（Beta）。
