> ## Documentation Index
> Fetch the complete documentation index at: https://help.nops.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Use Standard Performance Mode for Serverless Databricks Jobs

> Run serverless Databricks jobs that are not time-sensitive in standard performance mode, which uses fewer DBUs.

Serverless jobs and pipelines can run in performance-optimized or standard performance mode. Standard mode bills the same rate per DBU but uses fewer DBUs for the same work, and can take longer to start. nOps finds serverless jobs running in performance-optimized mode and, where the same job has also run in standard mode, measures the difference to estimate the saving.

| | |
| - | - |
| **Cloud** | Databricks |
| **Category** | Compute |
| **Service** | Serverless jobs |
| **Action** | Modify |
| **Effort** | Low |
| **Restart needed** | No |
| **Can be rolled back** | Yes |
| **Analysis window** | Last 14 days |
| **Source** | nOps |

## Why it matters

Performance-optimized mode is often left on for scheduled jobs that nobody is waiting on, where a few extra minutes to start do not matter.

## How nOps finds it

nOps reads the last 14 complete days of serverless jobs and pipelines DBUs tagged as performance-optimized. A job must be billed on at least 3 of those days, in less than 90% of the hours (so streaming and continuous work is left out), and must not already run mostly in standard mode. The DBU reduction is measured from the job's own runs over the last 30 days: at least 3 runs in each mode are needed, and the reduction is capped at 70%. The saving is the last 7 days of performance-optimized DBU cost, scaled to a month, times that reduction, and must be at least \$25 a month. Jobs without enough runs in both modes are not flagged, and workspaces with 90% or more of Databricks spend covered by credits are left out.

## Before you start

Know which jobs have a deadline tight enough that a slower start matters.

## How to implement

1. Confirm the job is not time-sensitive, for example a nightly or hourly batch with slack before its deadline.
2. In **Workflows**, open the job, and in the serverless compute settings turn off **Performance optimized** (or set `performance_target` to `STANDARD` in the job definition or Databricks Asset Bundle).
3. For Lakeflow pipelines, change the same setting on the pipeline.
4. Check run times and DBU use over the next few runs.

## Risks and considerations

* Standard mode can take several minutes longer to start, so runs finish later.
* To roll back, turn performance-optimized mode back on.

## FAQs

> [accordion] Why does nOps need runs in both modes?
>
> The saving is measured from the job's own runs. Without at least 3 runs in each mode there is nothing to measure, so the job is not flagged.

## Related

* Schedule Always-On Databricks Streaming Jobs
* [Supported recommendations](https://help.nops.io/inform/supported-recommendations.md)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.