Hey everyone! I've been hearing a lot about DeepSeek lately and really want to get it running locally for my coding projects. Privacy is a big deal for me, so I’d rather not rely on their cloud API if I can avoid it.
I’m currently rocking an old RTX 3060 with 16GB of RAM, but it's clearly not going to cut it for the larger 67B models or the new V3 weights. I'm trying to figure out the most cost-effective way to upgrade without breaking the bank.
I'm feeling a bit overwhelmed by the requirements. What is the best hardware setup for a smooth DeepSeek experience right now?
Grab two ASUS TUF Gaming GeForce RTX 3090 24GB cards. Its the most cost-effective way to get 48GB VRAM for those 67B models, though V3 really needs a massive unified memory setup.
Agreed, but honestly, for reliability a single NVIDIA GeForce RTX 4090 24GB is simpler than dual cards. Or try the Apple Mac Studio M2 Max 96GB if you're worried about heat.
If you are looking at the V3 weights specifically, you should probably look at a Apple Mac Studio M2 Ultra 192GB RAM. Unified memory is a game changer because it lets the GPU access that massive RAM pool. Its way more power-efficient than a multi-GPU setup and much quieter for a home office, tho the entry price is higher. It just handles huge models better than consumer cards.