Ask HN: How would you build a budget CPU compute cluster in 2023?
So lately I've been dabbling in a lot of stuff that requires a lot of CPU compute, 0 GPU and can be basically linearly scaled across threads and nodes in a cluster.
Now using only my own box is proving to be a bottleneck so I've been thinking of either using AWS spot instances or building my own mini-cluster (2-3 machines + switch) at home. Does it make sense to go cloud (even spot) when I would aim at high utilization?
As for the potential node spec:
- Ryzen 4500/5500 (seems best perf/$)
- Some mATX AM4 mobo with integrated GPU
- 2x8GB RAM
- mATX case, the smaller the better. ITX seems pricier.
Now using only my own box is proving to be a bottleneck so I've been thinking of either using AWS spot instances or building my own mini-cluster (2-3 machines + switch) at home. Does it make sense to go cloud (even spot) when I would aim at high utilization?
As for the potential node spec:
- Ryzen 4500/5500 (seems best perf/$)
- Some mATX AM4 mobo with integrated GPU
- 2x8GB RAM
- mATX case, the smaller the better. ITX seems pricier.
All the box does is basically run the k8s pod(s).
WDYT?