Guides
Questions about hardware for local AI and PC builds, answered with today's prices.
- How much VRAM do you need to run a 70B model?What Llama 3.3 70B takes at each quantisation, and the cheapest GPUs that hold it today.
- Can you run big AI models on a CPU instead of GPUs?When server memory beats a stack of GPUs, and the cheapest platforms for the biggest models.
- Does the CPU matter for AI if you're using GPUs?What the CPU does in a GPU rig for local LLMs: PCIe lanes, offloading to system RAM, AVX-512 and AMX, and which CPU to buy.
- Why do memory channels matter for AI, and how many does your CPU have?Bandwidth by platform: desktop dual channel against 8 and 12 channel servers.
- DDR4 or DDR5 for running AI models?What each channel gives, and what a GB of each costs on eBay today.
- Is a used RTX 3090 still worth it for AI?Its price per GB of VRAM against the other 24 GB+ cards on eBay right now.
- What's the cheapest way to run DeepSeek R1 or V3 (671B) at home?About 400 GB at Q4: why a server with lots of memory beats GPUs, and the cheapest platforms now.
- Which CPU gives the most gaming performance per dollar?Every desktop chip's gaming score against its lowest price today.
- Which motherboard for an AMD EPYC 7002 or 7003?What each SP3 board runs (it differs by model and revision), its PCIe and slots, and the price now.
- Does PCIe 3.0, 4.0 or 5.0 matter for GPUs and AI?When the PCIe generation and lane count matter, and what each platform gives.
- RDIMM, UDIMM or LRDIMM: which memory does your system take?Registered and unbuffered memory, which platforms take which, and what each costs per GB now.
- How do you spot a scam or fake GPU listing on eBay?The warning signs, and how this site flags listings that look too good to be true.