Kaggle Notebook GPU Quota Allocation Limits
0 reputation · 14 Jul 2021, 10:05 UTC
Resource Management in Managed Notebooks
Kaggle Notebooks provide GPU acceleration through a weekly quota system, utilizing accelerators such as the T4 or P100. These resources are managed at the account level rather than the individual notebook level.
Concurrency and Quota Constraints
When multiple notebooks are active simultaneously, the platform must balance the shared weekly quota against the immediate hardware requirements of each session. There is uncertainty regarding how the system prioritizes resource allocation when multiple GPU-enabled notebooks are running concurrently under a single account.
Specifically, it is unclear if the platform implements a strict limit on the number of concurrent GPU sessions or if it allows multiple sessions until the total weekly hour quota is exhausted.
- Does the platform restrict the number of simultaneous active GPU notebooks per user?
- How does the system handle resource contention when the weekly quota is nearly depleted across multiple active sessions?
kaggle