Comment by HenryMulligan

One RTX 3090 would allow you to run either Qwen 3.6 model with ~100k context at Q4_K_M quantization. Two would definitely allow full context and then some, or go for Q8 quantization. You can do quote a lot of LLM stuff with one or more 3090s. Or give them to me and I will happily put them to use!