Virtual Warehouses: Types, Sizing, and Scaling

Snowflake Architecture

Emily Melhuish

Technical Curriculum Developer, Snowflake

One Warehouse, Many Workloads

Depicting different workloads

One Warehouse, Too Many Workloads

Too many workloads.png

Standard vs Snowpark Optimized

Standard:

Use cases:
- SQL queries
- Data loading
- BI reporting
Gen2 option available: Improved performance
Test Gen1 vs Gen2 for your workload if cost matters

Snowpark Optimized

Use cases:
- ML model training
- Snowpark DataFrames
- Memory-intensive
Available from Medium upward

Warehouse Sizing

Warehouse size use cases:

XS = Light queries
XL = Large and Complex work

Query

CREATE WAREHOUSE analytics_wh
  WAREHOUSE_SIZE = 'LARGE'
  AUTO_SUSPEND = 300
  AUTO_RESUME = TRUE;

The sweet spot: cost vs warehouse size

bills per second with a 60-second minimum

The Sweet Spot

The sweet spot where

Cost Control: Auto-Suspend and Auto-Resume

Cost Control Setting Method 1

Auto-Suspend

Defines seconds of inactivity before warehouse suspends
Automatic suspension
300 seconds (5m) = DEFAULT

Cost Control Setting Method 2

Auto-Resume

= TRUE means it wakes up when a query arrives
No manual intervention required

Multi-cluster Warehouse

Use Case:

Answer to high-concurrency workloads
Clusters spin up/suspend automatically

Query

CREATE WAREHOUSE reporting_wh
  WAREHOUSE_SIZE = 'LARGE'
  MIN_CLUSTER_COUNT = 1
  MAX_CLUSTER_COUNT = 3
  SCALING_POLICY = 'STANDARD';

Enterprise edition feature

Screenshot 2026-05-13 at 4.33.22 pm.png

Scaling Up vs Scaling Out

Scaling up vs scaling out: larger warehouse vs more clusters

Economy vs Standard Scaling Policy

Standard

Spins up when anything is queued
Throughput over savings
Unpredictable Workloads

Economy

Conservative
Load will keep it busy >6 minutes
Steady and predictable workloads

Let's practice!

Snowflake Architecture

Preparing Video For Download...