I recently put together a cheap-and-cheerful machine-learning-capable PC.
- Pentium G4560 CPU
- NVIDIA 1060 6GB GPU
- 16 GB RAM
- 512 GB SSD
The main difference from a budget gaming rig is that I chose a relatively overpowered GPU. This is because for machine learning I think the main bottleneck will be GPU RAM size. This particular GPU model has a lot of RAM (6GB) for a relatively low price.
The other extravagance is the 512 GB SSD. It would have been cheaper to use a smaller SSD coupled with a traditional hard drive. I went with a single SSD because I don’t want to deal with the hard drive.
So far my son’s done far more gaming than I’ve done machine learning. He likes the system a lot. It’s >10x the graphics performance of his previous machine, a 2012 Mac Mini running Windows 10 in Bootcamp.