Build 01Placeholder
Cheap 8B chat
One 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.
- Price
- Placeholder
- Price pulled on
- Placeholder
- Source
- Placeholdername the sold listing or retailer page on the day it is pulled
- Idle
- Placeholder
- Load
- Placeholder
- Model
- Placeholder
- Quant
- Placeholder
- Context
- Placeholder
- Gen tok/s
- Placeholder
- Prompt tok/s
- Placeholder
Wall watts during the same run as the throughput row. A spec-sheet TDP does not belong here.
No run yet. A blank is an empty cell, not a slow machine.
Don’t buy this if
- You want a 30B-class coder in the same box. That is a different build.
- Several people will use it at once. Look at the family server.
- You are about to buy new parts to chase a speed this site has not measured.
Parts
| Part | Item | Price (USD) | Source | Notes |
|---|---|---|---|---|
| Host | Placeholderused mini PC or office desktop | Placeholder | Placeholder | Something you can open. A second RAM slot matters if the machine ships with one stick. |
| Memory | Placeholderroom for the quant, the cache, and the OS | Placeholder | Placeholder | Write down the stick count. One DIMM means single-channel. |
| Storage | Placeholderan SSD with room for a few model files | Placeholder | Placeholder | Weights live here. The speed row is about memory, not the drive. |
| GPU | Placeholderleave this out until a CPU run says you need one | Placeholder | Placeholder | Optional. The used 24GB GPU page is the later step. |
No affiliate links on this page yet. If one is added, the source cell will say so. See the affiliate disclosure.
Gotchas
- 8B is a size class. The benchmark row has to name the model, the quant, and the context. Those three change the result.
- A model that fits can still feel bad when memory bandwidth is poor. That shows up in tokens per second, once a run exists.
- Used office PCs often ship with a single stick of RAM. The photo is part of the listing.
This page is the shape of a build. The parts list says what to write down when a machine is actually chosen. It does not name a product, because a product without a dated price is how a guess gets onto the site.
An 8B-class chat model is the right target when one person wants a local conversation and can live with a modest pace. The useful questions are whether the model stays resident, and whether it answers at a context length you actually use. Both of those are runs. Until a machine in the lab produces one, the speed cells stay marked placeholder.
When this page becomes a real build, it should carry:
- The exact model and quant, copied into the benchmark table.
- The context length of that run, not the maximum the software advertises.
- Idle watts and load watts at the wall.
- The price of each part, the day it was pulled, and whether the figure was a sold price or an asking price.
If the first CPU run is already pleasant, the GPU line stays empty. A card bought “just in case” is how a cheap chat box stops being cheap. The used 24GB GPU page is the later step, after a measurement says the CPU is the limit.
Models that point here
A judgment from the model card and the size, not a speed ranking. Memory and the measured-or-not speeds are on each model page.
- Qwen3-4B · 4B
- Qwen3-8B · 8.2B
- DeepSeek-R1-Distill-Qwen-7B · 7B
- DeepSeek-R1-Distill-Llama-8B · 8B