A company runs a large nightly rendering workload. Tasks are independent, checkpoint progress, and can tolerate interruption and retry. The completion window has several hours of flexibility. Which TWO choices best reduce compute cost while making interruptions manageable? (Choose TWO.)

1. Disable checkpointing so interrupted work always restarts from the beginning.
2. Use Spot capacity with a diversified set of compatible instance types and suitable allocation behavior.
3. Persist checkpoints in durable storage and automatically retry unfinished tasks when capacity is interrupted.
4. Use only On-Demand instances even though the job can tolerate interruption.
5. Purchase a Capacity Reservation and assume it provides the same discount as Spot by itself.