Thanks for your wisdom Bob. It's good to get an expert comment. I think we have quite a few decades yet before Universal free green power and in the meantime i think it is a scarce resource with lots of competition: EV's, home heating, green hydrogen, direct air capture and many more. So i'm delighted to hear you think data centre power efficiency will improve steeply.
Like many, I'm struggling to navigate and weigh up what I think about AI - seeing both enormous potential benefits and huge problems and risks - so thanks for this useful information to help with this.
Hi Tom. I'm a retired Computer Scientist from the (pre-Turnip) US Government's National Institute of Standards and Technology, and think your analysis of Dr Schien's comments was spot on. Data centers are noisy and hot, which is a result of friction generated by motors driving colossal numbers of fans and disk drives inside the thousands of computer servers inside each data center, plus the intense heat generated by the microprocessors served by those fans and hard drives. Heat output is the real correlation with power input.
However, I am optimistic about the future. As you mentioned, we are in the midst of an AI gold rush, and what the companies want right now is what they term "compute" (computing power) and the hell with the cost. As the field becomes established, energy costs will move up the list of priorities to cut, etc.
Also, hardware design and efficiency will dramatically reduce power consumption, as will the eventual refinement of the software involved. Not to mention the eventual transition away form spinning metal disks for data storage!
At the moment, AI processing happens almost entirely on highly specialized nVidia modules, but if for example, we take a look at ARM-based processors (the ones in almost every phone and in all Apple products) the power consumption is much lower, while the "compute" is relatively high. I use an Apple M4 Mac Mini for most of my AI tinkering, and I'm not even sure it has a fan!
It is important to bear in mind that we are moving towards universal, free, green power, so the future looks good. I do think data centers should be required to build their own green power sources (this excludes nuclear!) into their design, and perhaps submit a rating card similar to the new EU "phone facts" labels.
I'm generally optimistic about data processing and AI - just don't count on the US to provide any solutions going forward.
Thanks for your wisdom Bob. It's good to get an expert comment. I think we have quite a few decades yet before Universal free green power and in the meantime i think it is a scarce resource with lots of competition: EV's, home heating, green hydrogen, direct air capture and many more. So i'm delighted to hear you think data centre power efficiency will improve steeply.
Like many, I'm struggling to navigate and weigh up what I think about AI - seeing both enormous potential benefits and huge problems and risks - so thanks for this useful information to help with this.
JFYI: French company Mistral, quite a prominent player in the AI field, have released an environmental impact report:
https://mistral.ai/news/our-contribution-to-a-global-environmental-standard-for-ai
Hi Tom. I'm a retired Computer Scientist from the (pre-Turnip) US Government's National Institute of Standards and Technology, and think your analysis of Dr Schien's comments was spot on. Data centers are noisy and hot, which is a result of friction generated by motors driving colossal numbers of fans and disk drives inside the thousands of computer servers inside each data center, plus the intense heat generated by the microprocessors served by those fans and hard drives. Heat output is the real correlation with power input.
However, I am optimistic about the future. As you mentioned, we are in the midst of an AI gold rush, and what the companies want right now is what they term "compute" (computing power) and the hell with the cost. As the field becomes established, energy costs will move up the list of priorities to cut, etc.
Also, hardware design and efficiency will dramatically reduce power consumption, as will the eventual refinement of the software involved. Not to mention the eventual transition away form spinning metal disks for data storage!
At the moment, AI processing happens almost entirely on highly specialized nVidia modules, but if for example, we take a look at ARM-based processors (the ones in almost every phone and in all Apple products) the power consumption is much lower, while the "compute" is relatively high. I use an Apple M4 Mac Mini for most of my AI tinkering, and I'm not even sure it has a fan!
It is important to bear in mind that we are moving towards universal, free, green power, so the future looks good. I do think data centers should be required to build their own green power sources (this excludes nuclear!) into their design, and perhaps submit a rating card similar to the new EU "phone facts" labels.
I'm generally optimistic about data processing and AI - just don't count on the US to provide any solutions going forward.