I've been looking at the Blackwell systems. About $3,500. But it can only be used for local AI models. I couldn't do anything else with it. And that is fine I guess. But if I bought a 5090 and spend a...
The challenge is when you wanna do / have toolchains of LLMs … 32GB gives you space for a solid LLM & the ability to cache … but 48GB you could do a pipeline that’s STT->LLM->TTS