Builds
Builds by objective
Pick the job first. The parts follow from that, and the page is there to hold a dated price and a measured speed once they exist. Until then every figure is marked placeholder.
01Cheap 8B chatOne person, an 8B-class chat model, bought used if the listing is honest.The smallest build on the site. Enough memory to keep a quantized 8B chat model resident. A GPU is optional, and only after a measurement.Placeholder0230B-class coding assistant on CPUOne developer, a 30B-class code model, no GPU required to start.A CPU machine with enough RAM to hold a 30B-class quant and a prompt long enough to be useful in a repo. Speed is unmeasured on purpose.Placeholder03RAG and a home assistantQuery your own notes and manuals on hardware you control, with a small chat model and an embedder.Chat can be a modest model. The extra cost is memory for a second model left resident, disk for the index, and idle watts if the box stays on.Placeholder04Multi-user family serverA few people sharing one model, queued or overlapping, without buying a GPU per person.Headroom past a single session, a queue you understand, and idle power you have metered. Multi-user here means a household, not a datacenter.Placeholder05CPU-only used XeonLots of inexpensive RAM and big quants, with no GPU in the base build.A used dual-socket Xeon with DDR3, chosen for capacity. Frankenstein is two E5-2697 v2 CPUs, 24 cores, 288 GB DDR3 at 1067 MT/s. The first Ollama speeds are measured. The prices on this page are not.Placeholder06Add a used 24GB GPU laterKeep the CPU box, then add one used 24GB GPU if a measured run says the CPU is the limit.A staged upgrade, not a first purchase. The P40 and the 3090 are the two 24GB classes this page will price when a listing is actually pulled.Placeholder