What is a Snowflake Virtual Warehouse? Compute Management Guide

A Virtual Warehouse is an independent compute cluster that makes up the Compute Layer of Snowflake's three-layer architecture. Every compute operation — SQL query execution, data loading, data transformation — runs on a Virtual Warehouse. It is completely decoupled from the Storage Layer, so you can create, size, and manage multiple Warehouses independently for different workloads.

Basic Architecture of a Virtual Warehouse

A Virtual Warehouse is a set of compute resources dynamically provisioned on a cloud provider (AWS/Azure/GCP). Each Warehouse is fully isolated, so load on one Warehouse never affects another.

• Storage isolation: every Warehouse can read the same data in the Storage Layer — no data duplication required
• Elasticity: resizing takes effect immediately, with no downtime
• Independence: no resource contention between Warehouses, making workload isolation easy
• Pay-per-use billing: credits are consumed based on runtime, with zero cost while suspended

Size Comparison Table (XS to 6XL)

Warehouses come in 8 sizes. Each step up roughly doubles the compute resources — and the credit consumption.

Billing starts at a 60-second minimum, after which it shifts to per-second billing. Note that even if you suspend a Warehouse immediately after resuming it, you are still charged for a full 60 seconds of credits.

Creating and Modifying a Warehouse

-- Create a Warehouse
CREATE OR REPLACE WAREHOUSE analytics_wh
  WAREHOUSE_SIZE = 'MEDIUM'
  AUTO_SUSPEND = 120
  AUTO_RESUME = TRUE
  MIN_CLUSTER_COUNT = 1
  MAX_CLUSTER_COUNT = 1
  INITIALLY_SUSPENDED = TRUE
  COMMENT = 'Analytics Warehouse for analysts';

-- Resize (effective immediately, no downtime)
ALTER WAREHOUSE analytics_wh
  SET WAREHOUSE_SIZE = 'LARGE';

-- Change Auto-suspend setting
ALTER WAREHOUSE analytics_wh
  SET AUTO_SUSPEND = 60;

-- Manual suspend / resume
ALTER WAREHOUSE analytics_wh SUSPEND;
ALTER WAREHOUSE analytics_wh RESUME;

Auto-suspend / Auto-resume

Auto-suspend and Auto-resume are the most fundamental — and most effective — settings for Warehouse cost management.

Auto-suspend

When a Warehouse stays idle for the specified number of seconds, it is automatically suspended. While suspended, it consumes zero credits. The default is 600 seconds (10 minutes).

Auto-resume

When a query is submitted to a suspended Warehouse, it automatically resumes. This is enabled by default. Resuming takes a few seconds to tens of seconds, so for latency-sensitive workloads consider keeping the Warehouse always on.

Recommended Settings by Workload

Warehouse Cache (Local Disk Cache)

While running, a Warehouse caches micro-partitions read from the Storage Layer onto local SSD. Subsequent reads of the same data are served quickly from this cache.

• Suspending the Warehouse clears the Warehouse Cache
• After resume, the cache is empty, so the first query has to perform a full scan from the Storage Layer
• Setting Auto-suspend too low can hurt cache hit rates and actually degrade performance

Multi-cluster Warehouse

A Multi-cluster Warehouse runs multiple compute clusters in parallel under the same Warehouse. It is available on Enterprise Edition or higher and prevents queuing when concurrent query volume grows.

Scale-up vs. Scale-out

Creating a Multi-cluster Warehouse

-- Multi-cluster Warehouse (up to 3 clusters, Standard policy)
CREATE OR REPLACE WAREHOUSE dashboard_wh
  WAREHOUSE_SIZE = 'MEDIUM'
  MIN_CLUSTER_COUNT = 1
  MAX_CLUSTER_COUNT = 3
  SCALING_POLICY = 'STANDARD'
  AUTO_SUSPEND = 300
  AUTO_RESUME = TRUE;

-- Switch to Economy policy
ALTER WAREHOUSE dashboard_wh
  SET SCALING_POLICY = 'ECONOMY';

Scaling Policies

Cost Optimization Best Practices

Separate Warehouses by Workload

-- ETL (larger size, short Auto-suspend)
CREATE WAREHOUSE etl_wh
  WAREHOUSE_SIZE = 'LARGE'
  AUTO_SUSPEND = 60
  AUTO_RESUME = TRUE;

-- Analytics (medium size, moderate Auto-suspend)
CREATE WAREHOUSE analytics_wh
  WAREHOUSE_SIZE = 'MEDIUM'
  AUTO_SUSPEND = 300
  AUTO_RESUME = TRUE;

-- Development (smallest, short Auto-suspend)
CREATE WAREHOUSE dev_wh
  WAREHOUSE_SIZE = 'XSMALL'
  AUTO_SUSPEND = 60
  AUTO_RESUME = TRUE;

Budget Control with Resource Monitors

-- Create a resource monitor
CREATE RESOURCE MONITOR monthly_budget
  WITH CREDIT_QUOTA = 500
  FREQUENCY = MONTHLY
  START_TIMESTAMP = IMMEDIATELY
  TRIGGERS
    ON 75 PERCENT DO NOTIFY
    ON 90 PERCENT DO SUSPEND
    ON 100 PERCENT DO SUSPEND_IMMEDIATE;

-- Apply the resource monitor to a Warehouse
ALTER WAREHOUSE analytics_wh
  SET RESOURCE_MONITOR = monthly_budget;

Cost Analysis with WAREHOUSE_METERING_HISTORY

-- Credit consumption per Warehouse over the last 30 days
SELECT
  warehouse_name,
  SUM(credits_used)         AS total_credits,
  SUM(credits_used_compute) AS compute_credits,
  SUM(credits_used_cloud_services) AS cs_credits
FROM SNOWFLAKE.ACCOUNT_USAGE.WAREHOUSE_METERING_HISTORY
WHERE start_time >= DATEADD(DAY, -30, CURRENT_TIMESTAMP())
GROUP BY warehouse_name
ORDER BY total_credits DESC;

Diagnosing Query Performance

• Check for spilling: if Query Profile shows Bytes Spilled to Local/Remote Storage, consider scaling up the Warehouse size
• Check for queuing: if queued_provisioning_time or queued_overload_time in QUERY_HISTORY is high, consider enabling Multi-cluster or raising the cluster count
• Leverage caches: repeated queries on the same Warehouse benefit from the Warehouse Cache and Result Cache

Check Your Understanding

Frequently Asked Questions