Skip to main content

Quick Reference

A concise reference for all 14 workload types in Lakemeter. Use this page to quickly identify which workload type to use and what configuration options are available.

Workload types at a glance

WorkloadWhat it isWhen to use itMin. Tier
JobsBatch compute for ETL, ML training, data processingScheduled or triggered pipelines that run and terminateStandard
All-PurposeInteractive compute for notebooks and developmentAd-hoc analysis, prototyping, development clustersStandard
DLTDelta Live Tables declarative data pipelinesManaged ETL with built-in data quality and monitoringStandard
DBSQLSQL analytics warehousesBI dashboards, SQL queries, analyst workloadsStandard (Classic/Pro), Premium (Serverless)
Model ServingReal-time ML model inference endpointsDeploying custom ML models for online predictionsPremium
Vector SearchManaged vector databaseSimilarity search, RAG applications, embeddingsPremium
FMAPI (Databricks)Foundation Model API for open-source modelsLlama, DBRX, Mixtral via Databricks-hosted endpointsPremium
FMAPI (Proprietary)Foundation Model API for third-party modelsGPT-4, Claude, Gemini via Databricks gatewayPremium
LakebaseManaged PostgreSQL-compatible databaseTransactional workloads, application backendsPremium
Databricks AppsManaged app hosting on DatabricksHosting web applications, dashboards, internal toolsPremium
AI Parse (Document AI)AI-powered document parsing and extractionProcessing PDFs, invoices, contracts with AIPremium
Shutterstock ImageAIAI image generation via ShutterstockGenerating images for marketing, content, prototypingPremium

Key terms

TermMeaning
DBUDatabricks Unit -- the billing unit for Databricks services. Different workload types have different $/DBU rates.
SKUStock Keeping Unit -- identifies the specific product being priced (e.g., JOBS_COMPUTE, SERVERLESS_SQL_COMPUTE).
PhotonHardware-accelerated query engine. Doubles the DBU rate but often halves runtime for compatible workloads.
ServerlessDatabricks-managed infrastructure. No VM configuration needed; infrastructure cost is included in the DBU price.
ClassicCustomer-managed infrastructure. You choose instance types and pay DBU + VM costs separately.

Configuration by workload type

Compute workloads (Jobs, All-Purpose, DLT)

These workloads share a common configuration pattern:

FieldDescriptionDefault
ServerlessToggle on for Databricks-managed compute, off for classicOff
Serverless ModeStandard or Performance (Performance uses 2x DBU rate)Standard
PhotonHardware-accelerated engine (classic mode only)Off
Driver Instance TypeVM instance for the driver (classic only)--
Worker Instance TypeVM instance for workers (classic only)--
Worker CountNumber of worker nodes (classic only)2
Driver Pricing TierOn-Demand, 1-Year Reserved, 3-Year ReservedOn-Demand
Worker Pricing TierSpot Instances, On-Demand, 1-Year Reserved, 3-Year ReservedSpot Instances
Payment OptionNo upfront, partial upfront, all upfront (reserved only)No upfront

DLT-specific (classic mode only):

FieldDescriptionOptions
SDP EditionFeature tier for the pipeline (Core, Pro, Advanced). Hidden when Serverless is enabled.Core, Pro, Advanced

Usage fields (Jobs):

FieldDescriptionDefault
Runs Per DayNumber of job executions daily1
Avg Runtime (minutes)Average duration of each run30
Days Per MonthActive days per month22

Usage fields (All-Purpose, DLT):

FieldDescriptionDefault
Hours Per MonthTotal compute hours--

DBSQL

FieldDescriptionOptions
ServerlessCheckbox to enable Serverless SQL (Premium+ only). When off, choose Pro or Classic.On (Serverless)
SizeDetermines DBU/hr consumptionSmall (12 DBU/hr)
Number of ClustersConcurrent cluster count for scaling1+
Hours Per MonthWarehouse uptime--

DBSQL warehouse sizes and DBU rates:

SizeDBU/hr
2X-Small4
X-Small6
Small12
Medium24
Large40
X-Large80
2X-Large144
3X-Large272
4X-Large528

Model Serving

FieldDescriptionOptions
Endpoint TypeCompute tier for the endpointCPU, GPU Small (T4), GPU Medium (A10G 1x), GPU Large (A10G 4x), etc.
Number of EndpointsConcurrent endpoint instances1+
Hours Per MonthEndpoint uptimeDefault: 730 (24/7)
FieldDescriptionOptions
Vector Search TypeStandard (4 DBU/hr per 2M vectors) or Storage Optimized (18.29 DBU/hr per 64M vectors)Standard
Capacity (M vectors)Number of vectors in millions1
Storage (GB)Additional storage capacity (20 GB free per endpoint unit, $0.023/GB/mo above)0
Hours Per MonthService uptimeDefault: 730 (24/7)

FMAPI (Databricks models)

FieldDescription
ModelDatabricks-hosted model (e.g., Llama, DBRX, Mixtral) — LLMs and Embedding models
Rate TypeToken-based (input token, output token) or Provisioned (provisioned scaling, provisioned entry)
QuantityFor token-based: millions of tokens/month. For provisioned: hours/month (730 = 24/7)

FMAPI (Proprietary models)

FieldDescription
ProviderOpenAI, Anthropic, or Google
ModelSpecific model (GPT-4, Claude, Gemini, etc.)
Endpoint TypeGlobal or In-Geo
Context LengthAll, Short, or Long (affects pricing for some models)
Rate TypeInput Token, Output Token, Cache Read, Cache Write, Batch Inference, Provisioned Scaling (availability depends on model)
Quantity (millions)Token volume

Lakebase

FieldDescriptionDefault
Capacity Units (CU)Compute units: 1, 2, 4, or 8 CU1
Number of Nodes1 = primary only, 2-3 = primary + read replicas (HA)1
Storage (GB)Database storage capacity (0-8192 GB)0
Hours Per MonthInstance uptimeDefault: 730 (24/7)

Databricks Apps

FieldDescriptionDefault
App SizeMedium or LargeMedium
Hours Per MonthApp uptimeDefault: 730 (24/7)

AI Parse (Document AI)

FieldDescriptionDefault
ModePages-based or direct DBUPages
ComplexityLow (text), Low (images), Medium, High — affects DBU rate per 1000 pagesMedium
Pages (thousands)Number of pages to process in thousands--

Shutterstock ImageAI

FieldDescriptionDefault
Images Per MonthNumber of images to generate--

Cost formula summary

Workload typeCost formula
Classic compute (Jobs, All-Purpose, DLT)(DBU/hr x hours/month x $/DBU) + (VM $/hr x hours/month)
Serverless compute (Jobs, All-Purpose, DLT, DBSQL)DBU/hr x hours/month x $/DBU (VM cost included)
DBSQL Classic/Pro(Warehouse DBU/hr x clusters x hours/month x $/DBU) + VM costs
Model ServingEndpoint DBU/hr x endpoints x hours/month x $/DBU
Vector SearchEndpoint units x DBU/unit x hours/month x $/DBU
FMAPIToken quantity (M) x DBU per M tokens x $/DBU
Lakebase(CU x nodes x hours/month x $/DBU) + storage costs
Databricks AppsApp DBU/hr x hours/month x $/DBU
AI ParsePages (thousands) x DBU per 1000 pages x $/DBU
Shutterstock ImageAIImages x fixed cost per image