[gymbeaux]

I just sold an RTX 3090 (released in 2020) for $1000. It should have been more like $200. I don’t see a point in running local LLMs with it but plenty of other people disagree, or want to “play around”. It’s so old that I can’t imagine a scenario where I’d be glad to have it- even if local LLMs ultimately catch up to the cloud frontier models (Opus, et al.). In a world where I can run Claude Opus 4.8 on an RTX 3090 at reasonable speeds, Anthropic doesn’t need even 1/20th the GPUs it has and it will dump them on the secondary market.