Which Snowflake warehouse size? Credits an hour, and when bigger is cheaper
Each Snowflake warehouse size uses twice the credits an hour of the one below, billed by the second after a 60-second minimum. So a query that runs twice as fast on the next size costs the same, and one that spills to disk on a small warehouse costs more there. How to pick the size from your own queries.
Tecyfy · Published · 7 min read

A Snowflake warehouse uses twice the credits an hour of the size below it, from 1 credit at X-Small to 512 at 6X-Large, and is billed by the second after a 60-second minimum. That makes the size you pick less about price per hour and more about whether your query gets faster. If it runs in half the time on the next size up, it costs exactly the same. If it doesn't speed up, the bigger warehouse doubles the bill.
Snowflake's own advice starts from the same place: "Don't focus on warehouse size." This guide shows why, with the numbers, and how to find the right size from the queries you already run.
Credits an hour for each warehouse size
These are the credits a Gen1 standard warehouse uses for each hour it runs, from Snowflake's warehouse overview. The dollar figures use Enterprise edition in AWS US East, where a credit costs $3 on demand.
| Size | Credits an hour | Enterprise, AWS US East |
|---|---|---|
| X-Small | 1 | $3 |
| Small | 2 | $6 |
| Medium | 4 | $12 |
| Large | 8 | $24 |
| X-Large | 16 | $48 |
| 2X-Large | 32 | $96 |
| 3X-Large | 64 | $192 |
| 4X-Large | 128 | $384 |
| 5X-Large | 256 | $768 |
| 6X-Large | 512 | $1,536 |
The credit price depends on your edition and region: in US East it is $2 on Standard and $4 on Business Critical. The Snowflake cost calculator has every region, and the Medium warehouse page works out one size across all of them.
Gen2 warehouses, Snowflake's newer standard warehouses on faster hardware, use more credits an hour for the same size: 1.35 credits for an X-Small on AWS and Google Cloud, 1.25 on Azure, with each size doubling from there. A Medium Gen2 on AWS uses 5.4 credits an hour against the Gen1 Medium's 4. Snowflake says the majority of queries finish faster on Gen2, so the per-query cost can still come out lower; the only way to know is to run your own.
How the per-second billing works
Snowflake bills a running warehouse by the second, with a minimum of 60 seconds each time it starts or resumes. Its documentation gives 3 cases:
- Runs for 30 to 60 seconds: billed for 60.
- Runs for 61 seconds: billed for 61.
- Runs for 61 seconds, suspends, then resumes and runs for under 60: billed for 121 (60 + 1 + 60).
Stopping a warehouse inside its first minute saves nothing, because that minute is already billed. In credits, a minute on an X-Small is 0.017 and a minute on an X-Large is 0.267.
Why a bigger warehouse can cost the same
Take a query that runs for 16 minutes on an X-Small. It uses 16/60 of 1 credit, about 0.27 credits, or $0.80 at $3.
Snowflake says that for the most part queries scale linearly with warehouse size, especially larger, more complex ones. If this query does, each size up halves the time and doubles the rate:
| Size | Run time | Credits | Cost at $3 |
|---|---|---|---|
| X-Small | 16 min | 0.27 | $0.80 |
| Small | 8 min | 0.27 | $0.80 |
| Medium | 4 min | 0.27 | $0.80 |
| Large | 2 min | 0.27 | $0.80 |
| X-Large | 1 min | 0.27 | $0.80 |
| 2X-Large | 30 s, billed 60 s | 0.53 | $1.60 |
Up to X-Large the answer arrives 16 times sooner for the same money. The 2X-Large is the first size that costs more, and only because a 30-second run is billed as a full minute.
Real queries are rarely that tidy. Press the controls below to see what happens when a query only partly speeds up, doesn't speed up at all, or runs out of memory on the small sizes.
| Size | Credits an hour | Runs | Billed | Cost |
|---|---|---|---|---|
| X-Small | 1 | 16 min | 16 min | $0.80Cheapest |
| Small | 2 | 8.0 min | 8.0 min | $0.80Cheapest |
| Medium | 4 | 4.0 min | 4.0 min | $0.80Cheapest |
| Large | 8 | 2.0 min | 2.0 min | $0.80Cheapest |
| X-Large | 16 | 1.0 min | 1.0 min | $0.80Cheapest |
| 2X-Large | 32 | 30 s | 1.0 min | $1.60Most |
Every size from X-Small to X-Large costs $0.80, until runs drop under the 60-second minimum. Pick the fastest of them.
Gen1 standard warehouse credits, $3 a credit (Enterprise, AWS US East), billed per second with 60 seconds minimum, for 1 run from a suspended warehouse. "Speeds up partly" cuts the time by 1.4 times a size; "Spills when small" is a made-up query that runs out of memory on X-Small and Small.
When bigger costs more
A bigger warehouse wastes money when the extra compute has nothing to do. Snowflake's considerations page says it plainly: larger is not necessarily faster for smaller, more basic queries, and small, simple queries typically don't need an X-Large or larger, however many run at once.
The 60-second minimum adds to it. A dashboard query that takes 5 seconds on a Medium is billed 60 seconds of a Medium. On an X-Large it is billed 60 seconds of an X-Large, 4 times as much, even if it finishes in 2 seconds.
When bigger is cheaper
A warehouse that is too small for a query runs out of memory. Snowflake then writes the overflow to the warehouse's local disk, and if that fills, to remote cloud storage. Its documentation says performance degrades drastically when that happens, and worse on remote storage. If moving up one size stops the spill and the query finishes in less than half the time, the bigger size costs fewer credits as well as being faster. Spilling is the same problem Spark has; our Spark SpillSpark SpillWhen a task's sort, join or aggregation no longer fits in execution memory, Spark writes part of it to local disk and reads it back later, so the task finishes, only slower. concept shows it step by step.
To find the queries that spill, Snowflake gives this query on the account usage views:
SELECT query_id, SUBSTR(query_text, 1, 50) partial_query_text, user_name, warehouse_name,
bytes_spilled_to_local_storage, bytes_spilled_to_remote_storage
FROM snowflake.account_usage.query_history
WHERE (bytes_spilled_to_local_storage > 0
OR bytes_spilled_to_remote_storage > 0 )
AND start_time::date > dateadd('days', -45, current_date)
ORDER BY bytes_spilled_to_remote_storage, bytes_spilled_to_local_storage DESC
LIMIT 10;
Its 2 recommended fixes are a larger warehouse or processing the data in smaller batches. One catch: with the query acceleration service on, Snowflake writes a small amount to remote storage for every eligible query, so a small nonzero bytes_spilled_to_remote_storage on its own is not a sign of trouble. Snowflake turns that service on by default for new Gen2 warehouses.
How to pick the size for your workload
Snowflake's guidance comes down to testing, not guessing:
- Pick queries that represent the workload. Snowflake suggests ones that complete within 5 to 10 minutes or less, and keeping queries of similar complexity on the same warehouse so its load is easier to read.
- Run the same queries on several sizes, for example Medium, Large and X-Large, and note the time of each.
- Multiply time by credits an hour. The cheapest size is the one where the query stopped halving its time; the fastest size at that cost is usually the one to keep.
- Check for spilling on the sizes below it, with the query above.
- Size down when you can. Per-second billing and auto-suspend let you start larger and step down, and you can resize at any time. A resize doesn't touch queries already running; new queries use the new size.
As a starting point, Snowflake says X-Small to Medium may be enough for queries in small-scale testing, while Large and up may be more cost effective for large production queries. For data loading, size depends on the number of files and the data in each, not on query complexity.
Size is not the fix for queues
If queries wait because many people run them at once, a bigger warehouse is the wrong lever. Snowflake says resizing is not intended for handling concurrency. It recommends more warehouses, or a multi-cluster warehouse that starts extra clusters of the same size as queries queue and stops them when they clear. Each running cluster is billed at its size's rate, and multi-cluster warehouses need Enterprise edition or higher.
Auto-suspend and the cache
A warehouse only bills while it runs, so auto-suspend decides most of the bill for a warehouse that sits idle between queries. Snowflake recommends a low value, 5 or 10 minutes or less, but matched to the gaps between your queries: with queries every 2 to 3 minutes, a 1-minute suspend makes the warehouse stop and start all day, and each start is billed its 60-second minimum.
Suspending also drops the warehouse's cache of table data, so the first queries after a resume can be slower. A bigger warehouse has a bigger cache. That is the trade Snowflake names: credits saved by suspending against speed kept by the cache.
If you are weighing Snowflake against Databricks for the same SQL work, our Snowflake vs Databricks cost comparison prices the same hours on both in each region they share. For a Snowflake table that queries keep scanning in full, clustering in dbt can cut the work before you pay for a bigger warehouse.
Questions
- How many credits does each Snowflake warehouse size use?
- A Gen1 standard warehouse uses 1 credit an 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. Each size doubles the one below. Gen2 warehouses use more credits an hour for the same size, 1.35 times as many on AWS and Google Cloud and 1.25 times on Azure.
- How is a Snowflake warehouse billed?
- By the second, with a minimum of 60 seconds each time the warehouse starts or resumes. A warehouse that runs for 30 seconds is billed for 60; one that runs for 61 seconds is billed for 61. A warehouse that runs 61 seconds, suspends and then runs for under 60 seconds again is billed for 121.
- Is a bigger Snowflake warehouse more expensive?
- Not per query, if the query speeds up in step with the size. A Medium uses twice the credits an hour of a Small, so a query that finishes in half the time on the Medium costs the same credits. A bigger warehouse costs more when the query does not get faster, which Snowflake says is common for small, simple queries, or when a short run is billed the 60-second minimum.
- What Snowflake warehouse size should I start with?
- Snowflake's guidance is to experiment: run the same queries, ones that take 5 to 10 minutes or less, on several sizes such as Medium, Large and X-Large, and compare. Smaller sizes (X-Small to Medium) may be enough for small-scale testing, while larger sizes may be more cost effective for large production queries. You can resize at any time.
- What is the difference between X-Small and Small in Snowflake?
- A Small warehouse has twice the compute of an X-Small and uses twice the credits an hour, 2 instead of 1. A query that runs twice as fast on the Small costs the same on both. A query that does not speed up costs twice as much on the Small, and one that runs out of memory on the X-Small can be cheaper on the Small.
- How do I know if my Snowflake warehouse is too small?
- Look for queries that spill. When a warehouse runs out of memory, Snowflake writes to local disk and then to remote storage, which it says degrades performance drastically. The QUERY_HISTORY view in ACCOUNT_USAGE has bytes_spilled_to_local_storage and bytes_spilled_to_remote_storage; Snowflake's fixes are a larger warehouse or smaller batches.
Sources