15 ms·
I got news for you. The 1M GPUs that meta purchased running at full (or less probably) load 24/7 is more than the energy of a single household. The energy cos
by dartos 2y ago
I got news for you.
The 1M GPUs that meta purchased running at full (or less probably) load 24/7 is more than the energy of a single household.
The energy cost is in training, not inference.
- kaba0 2y agoWhich is the same order of magnitude? Also, how often do they train from scratch?
- dartos 2y agoI don’t understand your first question, sorry. But for the second: Llama 1, 2, and 3 all have different architectures and needed to be trained from scratch. Llama 1 was released February 2023. Same training story for openAI’s Sora, dalle, and 4o. All of mistral’s models Mamba, Kan, and Each version of rwkv (they’re on 6 now) Not that this list is a result of survivor bias. It’s only looking at their published models too. Not the probably 1000s of individual training experiments that go into producing each model.
- kaba0 2y agoWhich is still absolutely nothing compared to something like youtube’s servers, which is absolutely nothing compared to something like the food industry. Like, if a couple of millions of people can use chatgpt in the manner they do today, would it matter if a house’s yearly energy budget was used up for that? Or 10?
- dartos 2y agoI guess because training AI doesn’t use as much energy as the largest energy consumers on the planet, we just shouldn’t even worry about it.