Cheap rented GPUs can hide restart costs when jobs stop midway
RunPod and Vast.ai offer and community-cloud for running at lower prices. The risk is that a training job may stop midway because of . The real cost is not only the listed hourly price.
It also includes the time needed to restart, the compute money already spent, and the chance that a becomes unusable. The item raises a cost-planning issue rather than giving a measured answer or fix.
Key points
- RunPod and Vast.ai are being considered for lower-cost .
- can stop midway because of .
- Restart cost can include lost time, wasted compute spend, and damaged .
- Long training or runs need a recovery plan before using cheaper interruptible GPUs.