Brolly Academy | Snowflake

Virtual Warehouse in Snowflake

Sizes, Credits, Multi-Cluster Warehouses and Cost Control (2026)

A plain-English guide to Snowflake virtual warehouses for 2026: what they are, every size and its credits per hour, the warehouse types (Gen2, Snowpark-optimized and Adaptive), auto-suspend and auto-resume, multi-cluster warehouses and scaling policies, Query Acceleration, how to choose a size and a cost control checklist.

Virtual warehouse in Snowflake explained: sizes and credits per hour, Gen2 and Snowpark-optimized warehouses, auto-suspend, multi-cluster scaling, cost.

Sizes and Credits Multi-Cluster Cost Control Updated 9 Nov 2026
Share

By Brolly Academy Team · Brolly Academy · Updated 24 October 2026

Virtual warehouse in Snowflake guide - Brolly Academy

Quick answer

A virtual warehouse in Snowflake is a named cluster of compute resources (CPU, memory and temporary storage) that runs your queries, data loads and other SQL work; you pay for it in credits only while it is running. Storage is separate, so you can start, stop, resize or add warehouses without touching your data. Sizes run from X-Small at 1 credit per hour to 6X-Large at 512 credits per hour, doubling at each step (Snowflake documentation, October 2026). "Clusters in Snowflake" usually means multi-cluster warehouses: one warehouse that adds extra clusters of the same size when many users query at once.

  • Sizes: X-Small, Small, Medium, Large, X-Large, then 2X-Large up to 6X-Large; each size uses twice the credits per hour of the size below
  • Billing: per second, with a 60-second minimum every time the warehouse starts or resumes
  • Auto-suspend and auto-resume: both are on by default, so an idle warehouse stops and wakes up when a query arrives
  • Types: standard (Gen1 or Gen2), Snowpark-optimized for memory-heavy work, and Adaptive Warehouses
  • Multi-cluster warehouses: Enterprise Edition or higher; they fix queuing from many users, not single slow queries

Key takeaways

  1. A warehouse is compute, not storage. Dropping or suspending a warehouse never deletes table data.
  2. Credits per hour double with each size, so a Medium (4 credits per hour) costs four times an X-Small for the same running time.
  3. Scale up (a bigger size) for heavy, complex queries; scale out (more clusters) for many users at the same time.
  4. New standard warehouses default to Gen2 in regions where it is available, and AWS Mumbai and Azure Central India are not on Gen2's exclusion list (Snowflake documentation, October 2026).
  5. Short auto-suspend times, one warehouse per workload and a resource monitor on every warehouse are the three habits that keep most bills under control.

1. Virtual warehouse at a glance

Short answer: Every Snowflake warehouse has a type, a size, suspend and resume settings, a cluster count and optional cost guards; this table maps each setting to what it controls.

If you are new to the platform, read what Snowflake does first. This post focuses only on compute.

Setting What it controls Default (Snowflake docs, Oct 2026) WAREHOUSE_TYPE Standard, Snowpark-optimized or Adaptive STANDARD WAREHOUSE_SIZE Compute per cluster and credits per hour X-Small in SQL; X-Large when created in Snowsight GENERATION Gen1 or Gen2 hardware for standard warehouses Gen2 where available AUTO_SUSPEND Seconds of idle time before the warehouse stops 600 seconds (10 minutes) AUTO_RESUME Wake up automatically when a query arrives TRUE MIN and MAX_CLUSTER_COUNT Single-cluster or multi-cluster 1 and 1 SCALING_POLICY How fast extra clusters start STANDARD ENABLE_QUERY_ACCELERATION Offload parts of big queries to serverless compute Off for single-cluster Gen1; on for Gen2 and multi-cluster RESOURCE_MONITOR Credit limit and alerts None

How a query uses a warehouse, step by step:

  1. You run a query from Snowsight, Snowflake CLI, a driver or a BI tool, with a warehouse set for your session.
  2. The cloud services layer checks your role and builds a query plan.
  3. If the warehouse is suspended and auto-resume is on, it starts, typically within a second or two.
  4. The warehouse reads the data it needs from storage, processes it and keeps recent data in its local cache.
  5. Results come back to you, and credits are counted per second.
  6. After the auto-suspend period with no queries, the warehouse stops and credit use stops.

2. What is a virtual warehouse in Snowflake?

Short answer: A virtual warehouse is the compute layer of Snowflake: an independent cluster that does the actual work of running SQL, while your data sits separately in the storage layer.

Snowflake has three layers: database storage, compute and cloud services. The Snowflake warehouses overview describes a warehouse as a cluster of compute resources that provides CPU, memory and temporary storage for queries, DML and data loading (Snowflake documentation, October 2026). For the full three-layer picture, see our Snowflake architecture diagram guide.

The word "warehouse" confuses many beginners, especially DBAs and ETL developers from Teradata, Oracle or SQL Server. In those systems, the warehouse is the whole database. In Snowflake, a virtual warehouse holds no data at all. Think of it as the engine, not the fuel tank. Warehouses are independent (a heavy ETL job does not slow a BI dashboard on another warehouse), they start and stop on demand, and they resize by changing one parameter.

Is a Snowflake warehouse the same as a database?

No. A database is a container for schemas, tables and views, and it lives in storage. A warehouse is compute. Any warehouse can query any database, as long as your role has the right privileges on both. You can have five databases and one warehouse, or one database and ten warehouses.

Is a virtual warehouse a virtual machine like EC2?

Not in the way you manage it. Snowflake runs warehouses on cloud infrastructure in AWS, Azure or Google Cloud, but you never see or log in to the machines. You only pick a type and a size. Snowflake handles provisioning, patching and the operating system. That is why it is called "virtual".

3. Types of warehouses in Snowflake

Short answer: Snowflake offers standard warehouses (Gen1 and the newer Gen2), Snowpark-optimized warehouses for memory-heavy Snowpark and ML work, and Adaptive Warehouses, where Snowflake manages size and scaling for you.

Type Best for Key facts (Snowflake docs, Oct 2026) Standard Gen1 General SQL, loading, BI The original standard warehouse; sizes X-Small to 6X-Large Standard Gen2 Same work on faster hardware Default for new standard warehouses where available; not for 5X-Large and 6X-Large Snowpark-optimized Memory-heavy Snowpark, Python and ML jobs 16 times the memory per node of a standard warehouse by default Adaptive Teams that do not want to tune size and scaling Enterprise Edition or higher; GA on AWS from 16 June 2026

Gen2 standard warehouses

The Gen2 standard warehouses page describes Gen2 as an updated standard warehouse running on faster hardware with software optimisations. New standard warehouses default to Gen2 in regions where it is available. Gen2 is offered on AWS, Azure and Google Cloud in all regions except a short exclusion list (AWS Zurich, AWS Cape Town, GCP Dammam and two Azure US Government regions), so Snowflake's two India regions, AWS Mumbai (ap-south-1) and Azure Central India (Pune), are not excluded (Snowflake documentation, October 2026). Gen2 is not available for 5X-Large, 6X-Large or Snowpark-optimized warehouses.

You choose the generation in SQL with GENERATION = '1' or '2'. Snowsight does not offer this setting. Gen2 credit rates are listed in Snowflake's Service Consumption Table, not on the docs page, so check that table before you assume Gen2 costs the same per hour as Gen1.

Adaptive Warehouses

Adaptive Warehouses (also called Adaptive Compute) remove most tuning. You no longer set size, multi-cluster counts, Query Acceleration or suspend and resume. Instead you set two main controls: MAX_QUERY_PERFORMANCE_LEVEL (default XLARGE) and QUERY_THROUGHPUT_MULTIPLIER (default 2). You create one with CREATE ADAPTIVE WAREHOUSE or convert an existing warehouse with ALTER WAREHOUSE ... SET WAREHOUSE_TYPE = ADAPTIVE. A Snowflake release note dated 16 June 2026 marks Adaptive Compute as generally available on AWS; Snowflake had not confirmed GA for Azure or Google Cloud regions in the pages we read (Snowflake documentation, October 2026). It needs Enterprise Edition or higher. As a learner, master standard warehouses first, then read about Adaptive Warehouses as the newer option.

4. Snowflake warehouse sizes and credits per hour

Short answer: Snowflake warehouse sizes go from X-Small (1 credit per hour) to 6X-Large (512 credits per hour), and the credit rate doubles at every step.

A credit is Snowflake's unit of compute. Your compute bill is credits used multiplied by the price per credit, which depends on your edition, cloud and region. Snowflake now calls compute credits "Platform Credits" and bills AI features separately in "AI Credits" (Snowflake Service Consumption Table, effective 15 October 2026).

Warehouse size Credits per hour (standard Gen1) Compared with X-Small X-Small 1 1x Small 2 2x Medium 4 4x Large 8 8x X-Large 16 16x 2X-Large 32 32x 3X-Large 64 64x 4X-Large 128 128x 5X-Large 256 256x 6X-Large 512 512x

Source: the Snowflake warehouses overview (Snowflake documentation, October 2026). The 5X-Large and 6X-Large sizes are GA in AWS and Azure regions. X-Small is the default size for CREATE WAREHOUSE in SQL, while Snowsight defaults to X-Large, which is an easy way to burn credits on a trial account if you click through without checking.

How per-second billing works

Warehouses bill per second, with a 60-second minimum each time they start or resume. The warehouse considerations page gives the rule clearly: a run of 30 to 60 seconds is billed as 60 seconds, and a run of 61 seconds is billed as 61 seconds (Snowflake documentation, October 2026). A suspended warehouse uses no credits.

Credits are also counted per cluster. A multi-cluster warehouse running three clusters of Medium size uses 3 x 4 = 12 credits per hour while all three run.

How much does a Snowflake warehouse cost?

Snowflake's Service Consumption Table (effective 15 October 2026) lists on-demand Platform Credit prices in AWS US East (N. Virginia) of $2.00 for Standard, $3.00 for Enterprise, $4.00 for Business Critical and $6.00 for Virtual Private Snowflake. Prices vary by cloud, region and edition, and capacity contracts are priced differently, so treat these as an example, not your rate.

Here is a simple worked example using those US East on-demand list prices:

Scenario Credits Example cost (US East, Enterprise, $3 per credit) X-Small running 1 hour 1 $3 Medium running 1 hour 4 $12 Medium running 3 hours a day for 22 working days 4 x 3 x 22 = 264 $792 Same Medium left running 24 hours for 30 days 4 x 24 x 30 = 2,880 $8,640

The last row is the real lesson. The same warehouse costs about eleven times more when nobody remembers to suspend it. Auto-suspend exists to prevent exactly that. Convert dollar figures to rupees at the current exchange rate when you present them to your manager, and say the figures are list prices for one region.

5. How to create and manage a virtual warehouse

Short answer: You create a warehouse with CREATE WAREHOUSE, choose a size and suspend settings, then use ALTER WAREHOUSE to resize, suspend, resume or change it while it runs.

You need a role with the right privilege; in most accounts SYSADMIN creates warehouses. If you are setting up your first trial account, our Snowflake account setup guide walks through the screens.

Simplified example, based on the CREATE WAREHOUSE and ALTER WAREHOUSE reference pages:

-- Create a small warehouse that stops after 60 idle seconds
CREATE WAREHOUSE IF NOT EXISTS etl_wh
  WAREHOUSE_SIZE = 'SMALL'
  AUTO_SUSPEND = 60
  AUTO_RESUME = TRUE
  INITIALLY_SUSPENDED = TRUE;

USE WAREHOUSE etl_wh;

-- Resize for a heavy month-end job, then go back
ALTER WAREHOUSE etl_wh SET WAREHOUSE_SIZE = 'LARGE';
ALTER WAREHOUSE etl_wh SET WAREHOUSE_SIZE = 'SMALL';

-- Stop or start it by hand
ALTER WAREHOUSE etl_wh SUSPEND;
ALTER WAREHOUSE etl_wh RESUME IF SUSPENDED;

A few behaviours worth remembering, from the ALTER WAREHOUSE reference (Snowflake documentation, October 2026):

  • Resizing does not touch running queries. Queries already running finish on the old size; new and queued queries use the new size once it is provisioned.
  • Suspending is graceful. The warehouse finishes in-flight queries first, then stops. To cancel everything, use ALTER WAREHOUSE ... ABORT ALL QUERIES.
  • You can change properties while it runs. A warehouse does not need to be suspended to change its size, type or other settings.
  • RESUME without IF SUSPENDED fails if the warehouse is already running.

You can also manage warehouses in Snowsight, but learn the SQL route, because the Gen2 setting and Snowpark-optimized memory options can only be set in SQL.

6. Auto-suspend and auto-resume

Short answer: Auto-suspend stops a warehouse after a set number of idle seconds, and auto-resume starts it again when a query arrives; both are on by default, and a short auto-suspend is the single biggest cost saver.

The default AUTO_SUSPEND in CREATE WAREHOUSE is 600 seconds, or 10 minutes. Snowflake's warehouse considerations page suggests a low value such as 5 or 10 minutes or less, because billing is per second (Snowflake documentation, October 2026). Setting AUTO_SUSPEND to 0 or NULL means the warehouse never suspends, which Snowflake does not recommend.

Two details matter when you pick the number:

  • The suspend check runs about every 30 seconds, so values below 30, or values that are not multiples of 30, may not behave exactly as you expect.
  • Match the timeout to the gaps in your workload. If queries arrive every 2 to 3 minutes, a 1-minute timeout makes the warehouse stop and restart repeatedly, and each restart carries the 60-second minimum charge.

What happens to the cache when a warehouse suspends?

Each running warehouse keeps a local cache of the table data it has recently read. When the warehouse suspends, that cache is dropped, so the first queries after a resume can run slower until the cache warms up again (Snowflake documentation, October 2026). The query result cache is a separate cache; our query optimization guide covers both.

That gives you a trade-off:

Workload Suggested auto-suspend thinking Why Nightly ETL or batch loads Short (about 60 seconds) Jobs run in bursts and then stop Ad hoc analyst queries Short to moderate (a few minutes) Gaps between queries are common Busy BI dashboards all day Longer (around 10 minutes) Keeping the cache warm helps dashboard speed Development and practice Short (about 60 seconds) Easy to forget an idle warehouse

These are starting points to test, not Snowflake rules. Watch your actual query gaps for a week and adjust.

7. Scale up vs scale out in Snowflake

Short answer: Scale up means a bigger warehouse size and helps large, complex queries; scale out means more clusters in a multi-cluster warehouse and helps when many users or jobs run at the same time.

This is one of the most asked interview questions on warehouses. The warehouse considerations page says resizing generally improves performance for larger, more complex queries, while adding clusters is meant for concurrency, and that resizing is not the tool for concurrency problems (Snowflake documentation, October 2026).

Question Scale up (resize) Scale out (multi-cluster) What changes Compute per cluster doubles with each size Number of same-size clusters grows Fixes One slow, heavy query; spilling to disk Queries waiting in a queue at busy times Edition All editions Enterprise Edition or higher How ALTER WAREHOUSE ... SET WAREHOUSE_SIZE Set MAX_CLUSTER_COUNT above MIN_CLUSTER_COUNT Cost pattern Higher rate per hour, often shorter run Extra credits only while extra clusters run

Remember it this way: if one query is slow, make the warehouse bigger; if it is slow only at 10 a.m. when the whole team logs in, add clusters. A third option is to give each workload its own warehouse (for example LOAD_WH, TRANSFORM_WH and BI_WH), which also makes cost tracking per team easier.

8. Multi-cluster warehouses: what "clusters in Snowflake" means

Short answer: A multi-cluster warehouse is one warehouse that can run several clusters of the same size, started and stopped automatically as concurrency rises and falls; it needs Enterprise Edition or higher.

When people search "clusters in Snowflake", most want this feature. Looking for clustering keys and micro-partition pruning instead? That is a different topic, covered in section 11 and in our Snowflake query optimization guide.

The multi-cluster warehouses page explains that you set MIN_CLUSTER_COUNT and MAX_CLUSTER_COUNT, and that MIN must be less than or equal to MAX (Snowflake documentation, October 2026). Every cluster has the warehouse's size, so a Large multi-cluster warehouse runs Large clusters only.

Maximized mode vs auto-scale mode

  • Maximized mode: MIN equals MAX, and both are above 1. All clusters start whenever the warehouse runs. Use it when concurrency is high and steady, and you accept paying for every cluster.
  • Auto-scale mode: MAX is greater than MIN. Snowflake starts extra clusters when queries queue and shuts them down when load drops. This is the common choice.

Scaling policy: Standard vs Economy

Scaling policy Behaviour Choose it when STANDARD (default) Starts extra clusters quickly to keep queuing low Dashboards and users who expect fast answers ECONOMY Keeps running clusters fully loaded; starts a new cluster only if there is at least 6 minutes of work for it Batch or background work where some waiting is fine

Scaling policy applies only in auto-scale mode (Snowflake documentation, October 2026).

How many clusters can you add?

The maximum cluster count depends on size. In SQL, the limit is 300 for X-Small, Small and Medium, 160 for Large, 80 for X-Large, 40 for 2X-Large, 20 for 3X-Large and 10 for 4X-Large to 6X-Large. Snowsight caps the setting at 10, so you need SQL for more (Snowflake documentation, October 2026).

Remember the bill grows with clusters. Snowflake's own example: an X-Large multi-cluster warehouse with a maximum of 10 clusters uses 160 credits per hour if all 10 run continuously.

Simplified example (Enterprise Edition or higher):

CREATE WAREHOUSE IF NOT EXISTS bi_wh
  WAREHOUSE_SIZE = 'MEDIUM'
  MIN_CLUSTER_COUNT = 1
  MAX_CLUSTER_COUNT = 4
  SCALING_POLICY = 'STANDARD'
  AUTO_SUSPEND = 300
  AUTO_RESUME = TRUE;

When should you use a multi-cluster warehouse?

  • A shared BI warehouse that slows down at the same hours every day because many users query at once
  • A customer-facing data app where waiting in a queue is not acceptable
  • Many parallel tasks or dbt models hitting one warehouse at the same moment

Do not use it to fix one slow query. Extra clusters run more queries side by side; they do not make a single query faster.

9. Snowpark-optimized warehouses

Short answer: Snowpark-optimized warehouses give far more memory per node than standard warehouses, for Snowpark workloads such as ML training inside stored procedures; ordinary SQL usually does not benefit.

Snowpark is Snowflake's set of libraries for Python, Java and Scala that push DataFrame work down to Snowflake compute. Most Snowpark code runs fine on a standard warehouse. But some jobs, such as training a machine learning model in a Python stored procedure, need lots of memory on a single node. That is what this type is for. Read our Snowpark in Snowflake guide for the language side.

The Snowpark-optimized warehouses page says the default configuration gives 16 times the memory per node of a standard warehouse (Snowflake documentation, October 2026). You set memory with RESOURCE_CONSTRAINT:

RESOURCE_CONSTRAINT Memory (up to) Minimum size Status (Oct 2026) MEMORY_1X or MEMORY_1X_x86 16 GB X-Small GA on all clouds MEMORY_16X or MEMORY_16X_x86 (default) 256 GB Medium GA on all clouds MEMORY_64X or MEMORY_64X_x86 1 TB Large Preview on AWS only

The _x86 options pin the CPU architecture, which helps when a Python package depends on x86. A Snowpark-optimized warehouse defaults to Medium size. Creating or resuming one can take longer than a standard warehouse, and resource constraints can only be set in SQL. Credit rates are in the Service Consumption Table, not on the docs page.

Simplified example:

CREATE WAREHOUSE IF NOT EXISTS ml_wh
  WAREHOUSE_TYPE = 'SNOWPARK-OPTIMIZED'
  WAREHOUSE_SIZE = 'MEDIUM'
  RESOURCE_CONSTRAINT = 'MEMORY_16X'
  AUTO_SUSPEND = 60;

10. Query Acceleration Service

Short answer: The Query Acceleration Service (QAS) offloads parts of unusually large queries to shared serverless compute, so one outlier query does not force you to resize the whole warehouse; it needs Enterprise Edition or higher.

QAS helps with outlier queries: big scans with selective filters, and large INSERT, COPY, UPDATE or DELETE operations. You switch it on with ENABLE_QUERY_ACCELERATION = TRUE and cap its extra compute with QUERY_ACCELERATION_MAX_SCALE_FACTOR (Snowflake documentation, October 2026).

Key facts from Snowflake's Query Acceleration Service documentation (October 2026):

  • Default scale factor: 8 when you enable it yourself; 2 when Snowflake turns it on automatically. Setting 0 removes the cap.
  • On by default for new Gen2 warehouses and multi-cluster warehouses; off by default for single-cluster Gen1 warehouses.
  • Check before you enable it: run SYSTEM$ESTIMATE_QUERY_ACCELERATION for a query ID, or check the QUERY_ACCELERATION_ELIGIBLE view in ACCOUNT_USAGE.

Keep the scale factor modest until you have seen what QAS adds to your bill.

11. A short note on micro-partitions and table keys

Short answer: Micro-partitions and clustering keys are about how table data is stored and pruned, not about warehouses; they are covered in depth in our query optimization guide.

Snowflake stores every table in micro-partitions, automatic storage units of about 50 MB to 500 MB of uncompressed data, stored column by column. A clustering key, set with CLUSTER BY, tells Snowflake to keep rows with similar values together so queries can skip more micro-partitions; Snowflake suggests it mainly for very large, multi-TB tables that are queried often and updated rarely (Snowflake documentation, October 2026). A clustering key changes how data is organised, while a multi-cluster warehouse changes how much compute runs, so do not mix them up in interviews. For clustering depth, pruning and Query Profile, read our Snowflake query optimization guide.

12. How to choose the right warehouse size

Short answer: Start small, run your real workload, read the Query Profile and the queue time, then resize one step at a time; separate warehouses by workload so each can have its own size.

Bigger is not always better. A larger warehouse finishes heavy queries faster, but small queries do not speed up much, and you pay double per hour for each step. Follow this six-step flow:

  1. Split workloads. Create one warehouse per job type: loading, transformation, BI and ad hoc or development.
  2. Start small. Begin at X-Small or Small for development and most loads.
  3. Run real queries. Snowflake suggests test queries that typically finish within 5 to 10 minutes, so you can compare sizes fairly (Snowflake documentation, October 2026).
  4. Read the signals. Check Query Profile for spilling to local or remote storage (a sign that memory is short) and check query history for time spent queued (a sign of concurrency pressure).
  5. Change one thing at a time. Spilling or long single queries: go up one size. Queuing at busy hours: add clusters or split the warehouse.
  6. Compare cost per job, not cost per hour. If a Medium finishes a job in half the time a Small takes, both cost about the same credits, and you get the answer sooner.

Typical starting points (test and adjust for your own data):

Workload Starting size Scaling idea Learning, trial account, practice X-Small None needed COPY INTO loads of moderate files X-Small to Small Size to the number and size of files dbt or SQL transformations Small to Medium Resize for month-end peaks Team BI dashboards Small to Medium Multi-cluster auto-scale at busy hours Heavy one-off backfills Large or above, for the job only Resize back down afterwards Python ML training in procedures Snowpark-optimized Medium Raise memory option only if needed

For loading, the considerations page says the warehouse size should match the number of files and the amount of data in each file (Snowflake documentation, October 2026). Our COPY INTO guide shows a loading lab.

13. Snowflake warehouse cost control checklist

Short answer: Keep auto-suspend short, attach resource monitors, separate workloads, watch WAREHOUSE_METERING_HISTORY every week and right-size based on evidence, not habit.

Use this checklist on any account you manage:

  • Auto-suspend set on every warehouse. No warehouse should have AUTO_SUSPEND = 0 or NULL unless there is a written reason.
  • Default sizes checked. Snowsight creates X-Large by default; resize anything created by clicking through.
  • One warehouse per workload. Loading, transforms, BI and development each get their own warehouse, so costs are traceable.
  • Resource monitors attached. Only ACCOUNTADMIN can create a resource monitor, and each warehouse can belong to only one warehouse-level monitor (Snowflake documentation, October 2026).
  • Max clusters kept realistic. Set MAX_CLUSTER_COUNT to what your budget can carry, not the highest number allowed.
  • Statement timeouts set. STATEMENT_TIMEOUT_IN_SECONDS on a warehouse stops runaway queries.
  • Idle cost reviewed weekly. Look for warehouses that run with nothing to do.
  • Serverless spend tracked separately. Resource monitors cover warehouses and their cloud services, not serverless or AI features; use budgets for those.

Simplified example of a resource monitor, based on the docs example:

CREATE OR REPLACE RESOURCE MONITOR limit1
  WITH CREDIT_QUOTA = 1000
  TRIGGERS ON 100 PERCENT DO SUSPEND;

ALTER WAREHOUSE wh1 SET RESOURCE_MONITOR = limit1;

Monitor actions are NOTIFY, SUSPEND (finish running queries, then stop) and SUSPEND_IMMEDIATE (cancel running queries and stop).

Simplified example to find idle cost per warehouse over the last 10 days, adapted from the WAREHOUSE_METERING_HISTORY reference:

SELECT warehouse_name,
       SUM(credits_used_compute)
         - SUM(credits_attributed_compute_queries) AS idle_credits
FROM SNOWFLAKE.ACCOUNT_USAGE.WAREHOUSE_METERING_HISTORY
WHERE start_time >= DATEADD('days', -10, CURRENT_DATE())
  AND end_time < CURRENT_DATE()
GROUP BY warehouse_name
ORDER BY idle_credits DESC;

ACCOUNT_USAGE views lag by 45 minutes to 3 hours and keep a year of history, so use them for trends, not for real-time alerts (Snowflake documentation, October 2026). Admins own most of this work; our Snowflake admin guide covers roles, monitors and ACCOUNT_USAGE in more depth.

14. Common mistakes with Snowflake warehouses

Short answer: Most warehouse problems come from leaving warehouses running, picking sizes by guesswork, using one warehouse for everything, or confusing multi-cluster warehouses with clustering keys.

  1. Leaving the Snowsight default. A warehouse created in Snowsight starts at X-Large (16 credits per hour), which can drain trial credits quickly.
  2. Turning auto-suspend off. AUTO_SUSPEND = 0 or NULL means the warehouse never stops.
  3. Auto-suspend too low for steady traffic. Constant restarts mean a cold cache and repeated 60-second minimum charges.
  4. Resizing to fix queuing, or adding clusters to fix one slow query. Size helps heavy queries; clusters help concurrency.
  5. One warehouse for everything. ETL, BI and developers slow each other down and costs cannot be traced.
  6. Forgetting the edition. Multi-cluster warehouses, Query Acceleration and Adaptive Warehouses need Enterprise Edition or higher.
  7. Using Snowpark-optimized warehouses for plain SQL. Non-Snowpark workloads may not benefit.

15. Why warehouses matter for Snowflake jobs in India

Short answer: Warehouse sizing, cost control and resource monitors appear in real Snowflake job descriptions, especially admin and senior data engineer roles, so this topic is worth learning well.

Of the 14 Snowflake job descriptions we read on 25 October 2026, several named warehouse and cost skills directly. Capgemini's Snowflake Data Engineer role in Hyderabad, Pune and Bengaluru asked for cost and performance tuning through warehouse sizing, clustering, query profiling and resource monitors. Infosys's Snowflake admin role (2 to 5 years) listed virtual warehouses and resource monitors, capacity planning and cost management. An Insight International Snowflake Administrator role in Hyderabad asked for proven Snowflake cost reduction and knowledge of credit and billing models.

On 25 October 2026, a Naukri search for "snowflake administrator" returned about 1,985 results across India. These are keyword matches, not a count of open jobs. AmbitionBox showed a typical range of ₹9.8 to 14.2 lakh a year for Snowflake Administrators in India (48 salaries, a small sample, 3 to 12 years' experience, October 2026).

If you are a SQL developer, ETL developer or DBA in a Hyderabad, Bengaluru, Pune or Chennai IT services firm or GCC, warehouse tuning is an easy place to show value, because you already understand workload patterns. For B.Tech, BCA, B.Sc and MCA freshers, explaining scale up vs scale out and auto-suspend clearly helps in a first technical round.

16. How Brolly Academy helps

Short answer: Brolly Academy's 45-day Snowflake course covers virtual warehouses, query caching and performance, editions and credit usage, and resource monitoring, with hands-on projects and interview preparation.

  • Our Snowflake Training in Hyderabad runs for 45 days across 15 modules. Virtual warehouses are taught in week 3 of the roadmap, alongside Time Travel, cloning, RBAC and security basics.
  • Module 14 covers query caching (result, metadata and data caches), clustering and performance, and cost optimisation. Module 15 covers the pricing model, editions and credit usage, resource monitoring and best practices.
  • You work on six projects, including an ETL pipeline, a financial data warehouse and e-commerce analytics, where warehouse choices affect speed and cost.
  • The trainer is Mr. Bala Krishna, with 13+ years in data warehousing, SQL and Snowflake.
  • You can choose classroom, live online or a self-paced video plan. The course page listed three fees in October 2026: Video Recording ₹8,999, Online Course ₹17,999 and Classroom Training ₹22,999, with EMI available.
  • You get a Brolly Academy course completion certificate, resume and interview support, mock interviews and placement assistance. Placement assistance is support with your search, not a job guarantee.

SnowPro exams are issued and examined by Snowflake, not Brolly Academy, and the exam fees are separate.

Want to size and tune Snowflake warehouses with confidence?

Join Brolly Academy's 45-day Snowflake course to practise creating warehouses, setting auto-suspend, reading query performance and controlling credits with resource monitors, then apply it in hands-on projects and mock interviews. Book a free demo class online or at our Hyderabad centre and see if it fits your goals.

Frequently asked questions

What is a virtual warehouse in Snowflake?

A virtual warehouse in Snowflake is a cluster of compute resources that runs your queries, data loads and other SQL work. It provides CPU, memory and temporary storage, but it holds no table data, because storage is a separate layer. You pay for a warehouse in credits only while it runs, and you can start, stop or resize it without affecting your data.

Is a Snowflake warehouse the same as a database?

No. A database is a container for schemas, tables and views, and it lives in Snowflake's storage layer. A warehouse is compute that runs queries against databases. Any warehouse can query any database if your role has the right privileges, so you can have many databases and one warehouse, or one database and several warehouses for different teams.

What are the Snowflake warehouse sizes?

Snowflake warehouse sizes are X-Small, Small, Medium, Large, X-Large, 2X-Large, 3X-Large, 4X-Large, 5X-Large and 6X-Large. X-Small is the default when you create a warehouse in SQL, while Snowsight defaults to X-Large. Each step up doubles the compute and the credits per hour, so choosing the right size has a direct effect on your bill.

How many credits does each Snowflake warehouse size use per hour?

A standard warehouse uses 1 credit per hour at X-Small, 2 at Small, 4 at Medium, 8 at Large, 16 at X-Large, 32 at 2X-Large, 64 at 3X-Large, 128 at 4X-Large, 256 at 5X-Large and 512 at 6X-Large. Credits are counted per cluster and billed per second, with a 60-second minimum each time the warehouse starts or resumes.

How much does a Snowflake warehouse cost?

It depends on size, running time, edition, cloud and region. Snowflake's Service Consumption Table effective 15 October 2026 lists on-demand AWS US East credit prices of $2 for Standard, $3 for Enterprise, $4 for Business Critical and $6 for Virtual Private Snowflake. At $3 per credit, a Medium warehouse running one hour costs about $12 in that region.

What is auto-suspend in Snowflake?

Auto-suspend stops a warehouse after it has been idle for a set number of seconds, so it stops using credits. The default in CREATE WAREHOUSE is 600 seconds, or 10 minutes, and Snowflake suggests a low value such as 5 or 10 minutes or less. Setting it to 0 or NULL means the warehouse never suspends, which Snowflake does not recommend.

What is auto-resume in Snowflake?

Auto-resume starts a suspended warehouse automatically when a query is sent to it, so users and jobs never need to start it by hand. It is on by default. Together with auto-suspend, it means a warehouse runs only when there is work to do, which keeps credit use close to your real workload.

What happens to the cache when a Snowflake warehouse is suspended?

The warehouse's local data cache is dropped when it suspends. The first queries after it resumes may run slower until the cache warms up again. The query result cache is a separate cache and is not the same thing. This trade-off is why busy dashboard warehouses sometimes use a longer auto-suspend, trading some credits for faster queries.

What is a multi-cluster warehouse in Snowflake?

A multi-cluster warehouse is a single warehouse that can run several clusters of the same size. In auto-scale mode, Snowflake starts extra clusters when queries start to queue and shuts them down when load drops. It is designed for concurrency, meaning many users or jobs at once, not for making one slow query faster.

Which Snowflake edition supports multi-cluster warehouses?

Multi-cluster warehouses need Enterprise Edition or higher, so they are available on Enterprise, Business Critical and Virtual Private Snowflake. On Standard Edition, MAX_CLUSTER_COUNT must stay at 1. Query Acceleration and Adaptive Warehouses also need Enterprise Edition or higher, which is worth checking before you design a solution for a client.

What does clusters in Snowflake mean?

In most searches, clusters in Snowflake means multi-cluster warehouses, where one warehouse runs several same-size compute clusters to handle many users at once. People sometimes mean clustering keys instead, which control how table data is arranged in micro-partitions to improve pruning. The two are unrelated: one is about compute, the other about storage layout.

What is the difference between scale up and scale out in Snowflake?

Scale up means resizing a warehouse to a bigger size, which helps large and complex queries. Scale out means adding clusters to a multi-cluster warehouse, which helps when many queries queue at the same time. Snowflake says resizing is not the tool for concurrency problems. Scale out needs Enterprise Edition or higher, while resizing works on every edition.

What is the difference between Standard and Economy scaling policy?

Standard is the default policy and starts extra clusters quickly to keep queuing low, which suits dashboards and users waiting for answers. Economy keeps running clusters fully loaded and starts a new cluster only if there is at least 6 minutes of work for it, which saves credits but allows more queuing. Scaling policy applies only in auto-scale mode.

What is maximized mode in a multi-cluster warehouse?

Maximized mode is when MIN_CLUSTER_COUNT equals MAX_CLUSTER_COUNT and both are above 1. All clusters start whenever the warehouse runs, so capacity is always at its maximum. It suits steady, high concurrency where you accept paying for every cluster. Auto-scale mode, where the maximum is higher than the minimum, is the more common and cheaper choice.

How many clusters can a multi-cluster warehouse have?

It depends on the warehouse size. In SQL, the limit is 300 clusters for X-Small, Small and Medium, 160 for Large, 80 for X-Large, 40 for 2X-Large, 20 for 3X-Large and 10 for 4X-Large to 6X-Large. Snowsight caps the setting at 10, so you must use SQL to go higher. Every running cluster adds credits.

What is a Snowpark-optimized warehouse?

A Snowpark-optimized warehouse is a warehouse type with much more memory per node, 16 times a standard warehouse by default. It suits memory-heavy Snowpark work such as training machine learning models in Python stored procedures. Memory options go from 16 GB to 256 GB, with a 1 TB option in preview on AWS only. Plain SQL workloads usually do not benefit.

What are Gen2 warehouses in Snowflake?

Gen2 standard warehouses are an updated version of the standard warehouse that runs on faster hardware with software optimisations. New standard warehouses default to Gen2 in regions where it is available, which includes AWS Mumbai and Azure Central India. Gen2 is not available for 5X-Large, 6X-Large or Snowpark-optimized warehouses, and its credit rates are listed in the Service Consumption Table.

What are Adaptive Warehouses in Snowflake?

Adaptive Warehouses are a newer warehouse type where you no longer manage size, multi-cluster settings, Query Acceleration or suspend and resume. You set a maximum performance level and a throughput multiplier instead. A Snowflake release note dated 16 June 2026 marks Adaptive Compute as generally available on AWS. It needs Enterprise Edition or higher.

What is the Query Acceleration Service in Snowflake?

The Query Acceleration Service offloads parts of unusually large queries, such as big scans with selective filters or large inserts and updates, to shared serverless compute. It needs Enterprise Edition or higher. The default scale factor is 8 when you enable it yourself and 2 when Snowflake enables it automatically for new Gen2 and multi-cluster warehouses.

How do I create a virtual warehouse in Snowflake?

Run CREATE WAREHOUSE with a name, a size and your suspend settings, for example a Small warehouse with AUTO_SUSPEND set to 60 and AUTO_RESUME set to TRUE. Then run USE WAREHOUSE to make it active for your session. You can also create one in Snowsight, but check the size, because Snowsight defaults to X-Large.

Can I resize a Snowflake warehouse while queries are running?

Yes. A warehouse does not need to be suspended to change its size or other properties. Queries already running finish on the old size, and new or queued queries use the new size once the extra compute is provisioned. Downsizing drops the cache tied to the removed compute, so some queries may slow down for a while.

How do I choose the right Snowflake warehouse size?

Start small, run real queries and read the evidence. Check Query Profile for spilling, which suggests a bigger size, and check query history for queuing, which suggests more clusters or a separate warehouse. Change one step at a time and compare cost per job, not cost per hour, because a bigger size often finishes the same work faster.

How do I control Snowflake warehouse costs?

Keep auto-suspend short, leave auto-resume on, give each workload its own warehouse and attach a resource monitor to every warehouse. Review idle credits weekly using WAREHOUSE_METERING_HISTORY in ACCOUNT_USAGE. Remember that resource monitors cover warehouses only, so serverless and AI features need budgets. Check sizes created in Snowsight, which start at X-Large.

Do Snowflake jobs ask for warehouse skills?

Yes, especially admin and senior data engineer roles. In job descriptions we read in October 2026, Capgemini asked for warehouse sizing and resource monitors, Infosys's admin role listed virtual warehouses and resource monitors, and an Insight International administrator role asked for proven cost reduction. AmbitionBox showed ₹9.8 to 14.2 lakh a year for Snowflake Administrators, from a small sample.

Does Brolly Academy teach Snowflake virtual warehouses?

Yes. Brolly Academy's 45-day Snowflake course teaches virtual warehouses in week 3 of its roadmap, and later modules cover query caching and performance, the pricing model, editions and credit usage, and resource monitoring. You practise in six projects and get mock interviews and placement assistance. SnowPro exams are separate and are run by Snowflake.

Learn with Brolly Academy

Review the syllabus, practical projects and batch timings, and book a free demo before you enrol.

Related Articles

Snowflake Admin

Roles, security and cost control.

Read Article

Snowflake Tutorial

Hands-on guide for beginners.

Read Article

Continue Your Learning

Snowflake Training in Hyderabad

45-day course covering Snowflake architecture, warehouses, data loading, security, performance and real-time projects, in classroom, live online and recorded plans.

View Course

dbt Course

Learn dbt, the transformation tool most Snowflake job descriptions ask for, to build tested, documented data models.

View Course

Databricks Course

Add the lakehouse side of modern data engineering with Databricks, Spark and Delta Lake, often used alongside Snowflake.

View Course