Skip to main content

Which Workload Type Do I Need?

Lakemeter supports 14 workload types covering the full Databricks platform. This page helps you choose the right one for your use case.

All workloads in a Lakemeter estimate A Lakemeter estimate showing multiple workload types with their individual costs and configuration summaries.

Adding a workload — select type, configure parameters, and save Animated: adding a new workload to an estimate — choose the workload type, configure compute and usage, then save.

Quick decision guide

I need to...Use this workloadGuide
Run scheduled ETL/data pipelinesJobsJobs Compute
Work interactively in notebooksAll-Purpose ComputeAll-Purpose Compute
Build declarative data pipelines with quality checksDLTDelta Live Tables
Run SQL queries, dashboards, BI toolsDBSQLDBSQL Warehouses
Deploy a custom ML model as an APIModel ServingModel Serving
Store and query vector embeddings (RAG)Vector SearchVector Search
Call open-source LLMs (Llama, DBRX, Gemma)FMAPI DatabricksFMAPI — Databricks
Call commercial LLMs (Claude, GPT, Gemini)FMAPI ProprietaryFMAPI — Proprietary
Use a managed PostgreSQL databaseLakebaseLakebase
Host a web app on DatabricksDatabricks Apps
Parse documents with AIAI Parse (Document AI)
Generate images with AIShutterstock ImageAI

Workload categories

Compute Workloads

These workloads run Spark clusters or SQL warehouses. Costs are based on instance types, cluster size, and hours of usage.

WorkloadPricing modelSupports ClassicSupports ServerlessVM costs
JobsDBU/hr x hoursYesYesClassic only
All-PurposeDBU/hr x hoursYesYesClassic only
DLTDBU/hr x hoursYesYesClassic only
DBSQLDBU/hr x hoursYes (Classic/Pro)YesClassic/Pro only

AI/ML & Data Services

These workloads are always serverless with no VM costs. Pricing is based on GPU type, token volume, or provisioned capacity.

WorkloadPricing modelUnitVM costs
Model ServingDBU/hr by GPU typeHoursNever
Vector SearchDBU/hr by endpoint unitsHours + storageNever
FMAPI DatabricksDBU per million tokens or DBU/hr (provisioned)Tokens or hoursNever
FMAPI ProprietaryDBU per million tokensTokensNever
LakebaseDBU/hr (compute) + DSU (storage)Hours + GBNever
Databricks AppsDBU/hr by app sizeHoursNever
AI ParseDBU per 1000 pagesPagesNever
Shutterstock ImageAIFixed per imageImagesNever

Common fields

All workloads share these base fields:

FieldDescription
Workload NameA descriptive label (e.g., "ETL Pipeline", "Analytics Warehouse")
Workload TypeThe Databricks service type from the dropdown
Display OrderSort position in the estimate (drag to reorder)

Usage input modes

Compute workloads (Jobs, All-Purpose, DLT, DBSQL) support two ways to specify usage:

Run-Based (default): Specify runs per day, average runtime, and days per month. The formula is:

Hours/Month = (Runs Per Day x Avg Runtime Minutes / 60) x Days Per Month

Direct Hours: Enter total hours per month directly. Common values:

  • 730 = 24/7 (always-on)
  • 176 = 8 hours x 22 business days
  • 44 = 2 hours x 22 business days

Model Serving, Vector Search, and Lakebase use direct hours only. FMAPI workloads use token volume (millions/month) or provisioned hours instead of either mode.

Pricing tier restrictions

Some workloads require a Premium pricing tier or above:

WorkloadStandard tierPremium tier
Jobs (Classic)YesYes
Jobs (Serverless)NoYes
All-Purpose (Classic)YesYes
All-Purpose (Serverless)NoYes
DLTYesYes
DBSQL (Classic/Pro)YesYes
DBSQL (Serverless)NoYes
Model ServingNoYes
Vector SearchNoYes
FMAPI DatabricksNoYes
FMAPI ProprietaryNoYes
LakebaseNoYes
Databricks AppsNoYes
AI ParseNoYes
Shutterstock ImageAINoYes

If you select Standard tier for your estimate, workloads that require Premium will not be available.

Cost components

Depending on the workload type, your total cost may include one or more of:

  • DBU Cost -- Databricks Units consumed, multiplied by the $/DBU rate for your SKU
  • VM Cost -- Cloud infrastructure costs for driver and worker VMs (Classic compute only)
  • Storage Cost -- Provisioned storage (Lakebase DSU, Vector Search overflow)
  • Token Cost -- For FMAPI workloads, based on millions of tokens processed

The total monthly cost for each workload is displayed in the estimate summary. All workloads sum to the estimate total.