Standard vs Dedicated Compute: When to Use Each
Databricks offers two access modes for clusters — Standard (formerly Shared) and Dedicated (formerly Single-User). Standard clusters allow multiple users to share compute with Unity Catalog-enforced isolation, while Dedicated clusters give one user exclusive access with full language support and no restrictions. Choose Standard for cost-efficient team use and Dedicated when you need unrestricted library access, custom Spark extensions, or GPU workloads.
- Understand the differences between Standard and Dedicated access modes
- Learn which workloads require Dedicated and which benefit from Standard
- Configure access mode settings on clusters and policies
Who this is for: Platform administrators and data engineers deciding how to provision clusters for teams with varying isolation and access requirements.
Part of the Databricks Compute section of the Databricks tutorial series.
Architecture / Concept Overview: Standard vs Dedicated Compute: When to Use Each
Access mode controls how a cluster handles multi-tenancy and security boundaries. Standard mode uses process-level isolation to let multiple users share a cluster safely, with Unity Catalog enforcing per-user data access. Dedicated mode assigns the entire cluster to one user with no isolation overhead, enabling unrestricted access to all four languages, arbitrary JVM libraries, and FUSE mounts.
*Standard clusters isolate users at the process level while sharing the underlying JVM, governed by Unity Catalog.*
*Dedicated clusters give one user unrestricted access to the full cluster, including GPUs, FUSE mounts, and arbitrary JARs.*
*The workload type drives the access mode decision: Standard for shared work, Dedicated for unrestricted access.*
Key Terms
- Standard Access Mode
- A cluster mode that allows multiple users with Unity Catalog-enforced isolation and some language/library restrictions.
- Dedicated Access Mode
- A cluster mode that assigns the entire cluster to one user with no restrictions on languages, libraries, or system access.
- Process Isolation
- The mechanism Standard clusters use to separate user sessions so one user cannot access another's data or variables.
- Lakeguard
- The security technology that enforces user isolation on Standard clusters.
Prerequisites and Setup
- A Databricks workspace with Unity Catalog enabled
- Understanding of your team's workload requirements (languages, libraries, GPU needs)
- Cluster-create permission or an appropriate cluster policy
- Knowledge of which users need unrestricted access
Step-by-Step Implementation
Configuration Reference
| Capability | Standard (Shared) | Dedicated (Single-User) |
|---|---|---|
| Multi-user access | Yes | No (one user) |
| Unity Catalog enforcement | Yes, per-user | Yes, single identity |
| Python | Yes | Yes |
| SQL | Yes | Yes |
| Scala | Limited (no arbitrary JARs) | Full |
| R | Yes | Yes |
| Custom JVM libraries | No | Yes |
| FUSE mounts | No | Yes |
| GPU support | No | Yes |
| Credential passthrough | No (uses UC) | Yes |
| Lakeguard isolation | Yes | Not applicable |
Monitoring, Cost, and Security Considerations
Monitoring
Track cluster creation events filtered by data_security_mode to see how many Standard vs Dedicated clusters are running. Monitor per-user DBU consumption to identify whether Dedicated clusters are being fully utilised or sitting idle.
Cost Optimisation
- Standard clusters let multiple users share infrastructure, reducing total cluster count and cost.
- Dedicated clusters charge the same DBU rate but serve only one user, so utilisation matters more.
- Set shorter auto-termination on Dedicated clusters since only one person uses them.
- Use cluster policies to default to Standard and require justification for Dedicated.
Security and Governance
- Standard mode enforces Unity Catalog per-user access controls, preventing users from seeing each other's data.
- Dedicated mode runs as a single identity, so the user has full access to everything their credentials permit.
- On Standard clusters, Lakeguard prevents users from accessing the filesystem, environment variables, or other users' sessions.
- Use Dedicated mode only when the workload genuinely requires it (custom JARs, GPU, FUSE).
Common Pitfalls and Recommended Patterns
- Defaulting to Dedicated for all users: this is expensive and unnecessary when most users only need Python and SQL.
- Running GPU workloads on Standard clusters: GPU support requires Dedicated mode.
- Assuming Standard limits all Spark features: Standard supports Python, SQL, R, and most PySpark operations.
- Forgetting to set
single_user_nameon Dedicated clusters: without it, anyone with attach permission can use the cluster. - Over-provisioning Dedicated clusters: since only one user connects, right-size aggressively.
- Not using cluster policies to enforce mode: without policies, users default to the most permissive option.
Frequently Asked Questions
Can I change the access mode on a running cluster?
No. Access mode is set at cluster creation time and cannot be changed. You must create a new cluster with the desired mode.
Does Standard mode affect performance?
There is a small overhead from process isolation, but for most workloads the difference is negligible. Photon and adaptive query execution work the same in both modes.
Why can I not install a custom JAR on a Standard cluster?
Custom JVM code could bypass process isolation and access other users' data. Databricks restricts this to maintain security boundaries. Use a Dedicated cluster for custom JAR workloads.
Can job clusters use Standard mode?
Job clusters typically use Dedicated mode since each job run has a single identity. For multi-task jobs with different user contexts, each task gets its own cluster.