估算ChatGPT的年度能耗

Would love if there was a public API that showed energy use stats for AI models.

Anyone know if OpenAI is moving towards more energy-efficient hardware or cloud providers?

What about user devices? The energy to send requests and display results might be small but could add up with millions of users.

Would be great if we had some community projects measuring AI energy footprints more accurately.

I've heard some rough estimates but nothing super accurate. It probably depends a lot on usage and server efficiency though.

I tried to find a number but it’s so variable and depends on how often the model is called. Plus, OpenAI doesn't publish exact figures.

I’m worried about the environmental impact if AI usage keeps growing like this.

I bet OpenAI has teams working on reducing the carbon footprint but it’s gotta be tough balancing performance and sustainability.

I guess the best guess would be comparing it to similar large AI models. Those can use megawatt-hours annually just for training, not even counting usage.

我刚开始考虑人工智能的能源方面，这有点令人担忧，但也很吸引人。

有人在跟踪人工智能的能源使用与其他行业的比较吗？

有没有办法计算每个聊天会话的碳足迹？

专用人工智能芯片能显著降低能耗吗？

我见过一些估计，大型模型每年可能消耗的能量相当于一个小城市，但这只是非常粗略的估计。

碳抵消怎么样？你认为OpenAI会购买抵消来平衡能源使用吗？

我在某处读到，随着这些模型的发展，人工智能的能源使用正成为全球电力需求中越来越大的一部分。

如果OpenAI开始发布一些关于能源使用的透明度报告，那将非常棒。

别忘了冷却！服务器需要大量能量来防止过热。

我很好奇ChatGPT每次查询平均使用多少能量。有人见过相关数据吗？

我想知道是否减少使用ChatGPT的频率或更高效地使用它可以总体上减少能源消耗。

从能源角度来看，在本地设备还是云服务器上运行人工智能模型更好？

我认为人们很容易忘记幕后发生了多少工作。我们只看到文本，但实际上是成千上万的GPU在进行计算。