Hi there Openskies
Thanks for posting on QandA platform
Firstly, you'll need to select the appropriate GPU instances based on your workload requirements, such as the NV-series or NC-series. Next, estimate the compute costs by considering factors like the duration of video processing tasks (e.g., 13-hour videos) and the number of concurrent videos to be processed
For a detailed info follow this below documentation
https://azure.microsoft.com/en-in/pricing/details/machine-learning/#pricing
If this helps kindly accept the answer thanks much