5 ms·
GPT inputs outgrow world chip and electricity capacity
- fulafel 3y agoBadly editorialized title. Regarding the substance of the article, the curve from the 3 data points (1: 100x 2: 25x 3: 25x) could fit lots of different ways besides "growing 30x per generation".
- marsissippi 3y agoInteresting Pull quote: GPT-4 needed about 50 gigawatt-hours of energy to train. Using our scaling factor of 30x, we expect GPT-5 to need 1,500, GPT-6 to need 45,000, and GPT-7 to need 1.3 million.
- toomuchtodo 3y agoAssuming OpenAI is training exclusively using Microsoft compute, it should be low carbon energy. https://news.microsoft.com/europe/features/as-the-world-goes-digital-datacenters-that-make-the-cloud-work-look-to-renewable-energy-sources/ https://news.microsoft.com/europe/features/as-the-world-goes...
- fulafel 3y agoDo the dynamics of PPAs and RECs really translate to lowering electricity production CO2 footprint? It seems that it would work only when there is significantly more demand for that kind of electricity that would otherwise we produced, but since wind and solar are competitive on their own, and in many places there's no carbon cap-and-trade system to prevent the increased power demand leaking over to fossil energy, it's not obvious.
- toomuchtodo 3y ago> Do the dynamics of PPAs and RECs really translate to lowering electricity production CO2 footprint? With new PPA models where purchases are made when generation occurs, yes. https://www.datacenterdynamics.com/en/news/google-to-expand-247-clean-energy-matching-with-flexidao/ https://www.datacenterdynamics.com/en/news/google-to-expand-... https://finance.yahoo.com/news/engie-supply-microsoft-data-centers-100000051.html https://finance.yahoo.com/news/engie-supply-microsoft-data-c... https://www.mckinsey.com/industries/electric-power-and-natural-gas/our-insights/decarbonizing-the-grid-with-24-7-clean-power-purchase-agreements https://www.mckinsey.com/industries/electric-power-and-natur...
- fulafel 3y agoI'm not sure the first couple links support your claim (didn't have time for the last). in any case would be better to see analysis from some independent researchers than trade pubs.
- Nullabillity 3y agoUntil we're in a world of 100% renewables, anything you do that uses renewables just pushes someone else to use more fossil fuel instead. Access to renewables isn't a blanket excuse to waste it. (Plus "renewable" is kind of a misnomer, there are still limited resources going into the production and maintenance of those power plants too.)
- az09mugen 3y agoMay people listen to your wisdom, I totally agree with you.
- faeriechangling 3y agoMicrosoft is hiring nuclear scientists presumably to power AI. If this becomes the case, I don't think it will be the case that Microsoft's green energy use is just pushing people to use dirty energy elsewhere. If anything they would be helping to accelerate a green transition. I don't see what's happening now as "Green" but I don't think you need 100% renewables worldwide for a datacenter to be green. You could also build a datacenter by a Dam which is in a location where people wouldn't normally live. This is not all zero sum.
- krunck 3y agoThere is finite low carbon energy available. Would they choose to slow the rate of training the next generations of LLMs and use only greener every sources or would they go full speed regardless of the source of the energy? I think they would do the later in order to be first in the market with a next generation LLM.
- toomuchtodo 3y agoProbably because they're the one making the commitments (and renewables are cheaper anyway). https://www.datacenterknowledge.com/energy/microsoft-pledges-emit-zero-carbon-2030 https://www.datacenterknowledge.com/energy/microsoft-pledges... https://www.lazard.com/media/2ozoovyg/lazards-lcoeplus-april-2023.pdf https://www.lazard.com/media/2ozoovyg/lazards-lcoeplus-april...
- dekhn 3y agoThe are all the same arguments that wrongly predicted that DNA sequencing would overtake hard drive storage capacity. People aren't going to do truly uneconomic things just to scale language models exponentially.
- danielscrubs 3y agoReally wish AMD could hurry up and implement a transparent module for gpu computing into llvm. It’s not good that GPUs are this opaque.
- RecycledEle 3y agoNew technologies experience very rapid exponential growth for a while. This should not be a surprise.