Im so hyped to finally build a dedicated rig for running local LLMs and messing around with stable diffusion! Ive been researching for weeks. I read that the RTX 4090 is obviously the king because of 24gb vram but then other threads say two 3090s used is better value for tensor parallelism and now my brain is fried trying to figure out what actually matters for day to day model training.

Here is what I am working with:

• Budget is around 2k max
• Mainly fine-tuning 7b to 13b models and image generation
• Need it up and running by next month for a side project

Should I just drop the cash on a single 4090 or hunt down a used enterprise card with tons of vram instead...