# Snowflake Auto Tune modes and safe backoff > Altimate Auto Tune right-sizes and suspends Snowflake warehouses automatically, and reverts to the default size if queue time or latency crosses a threshold. Source: https://altimate.ai/product-highlights/snowflake-warehouse-auto-tune-right-sizing [.md](https://altimate.ai/product-highlights/snowflake-warehouse-auto-tune-right-sizing.md) Copy as MD Enterprise Platform ## Auto Tune Right-Sizes Snowflake Warehouses Without Slowing Queries Altimate Auto Tune right-sizes and suspends Snowflake warehouses automatically, and reverts to the default size if queue time or latency crosses a threshold. Pradnesh PatilMay 24, 2026Aug 11, 20266 min read Auto Tune right-sizes Snowflake warehouses without slowing queries TL;DR - **What it is.** Altimate AI's Auto Tune agents manage your Snowflake warehouse size, idle time, and clustering configuration. It cuts costs while staying within the performance bounds you set. Once enabled, your team doesn't need to do anything further. - **Who it is for.** Data platform leads and engineering managers who own the Snowflake bill and the query SLAs those warehouses have to meet. - **What you get.** Altimate manages your warehouses based on their workload patterns, inside your performance bounds. Customers save up to 20% of warehouse costs, and the savings are reported in a form both engineering and finance can use. ## Auto Tune Runs Autonomous Agents with Human Guardrails Auto Tune lowers your Snowflake warehouse cost. It runs autonomous agents that decide the optimum warehouse size and idle times. You turn it on warehouse by warehouse. | Agent | What it does | | --- | --- | | **Warehouse Shutdown and Scaling** | Scales clusters back and suspends the warehouse before Snowflake's own auto-suspend would fire. | | **Warehouse Sizing** | Drops the warehouse one level below its default during low-demand hours, reading the pattern from the same hours in earlier weeks. | Warehouse Sizing runs in one of two modes: 1. **Approval mode.** You get next week's downsizing schedule and approve the blocks you're happy with. Nothing resizes until you say so. Approved blocks run as Snowflake tasks, so Altimate needs task permissions on your account. 2. **Real-Time mode.** The agent decides on its own, every 15 minutes, without asking. It carries a built-in backoff that watches performance after each resize and reverts the change if performance degrades. Both modes take custom schedule blocks, so the hours you care about stay yours. A warehouse that idles from 2am to 4am on Tuesdays can sit one size down for those two hours. A warehouse running month-end close stays at its default size for the whole run, whatever the agent would otherwise decide. You can check the numbers before turning anything on. With the toggle still off, the warehouse page shows what Auto Tune would save you on any warehouse. The COMPUTE\_WH Snowflake warehouse detail page with the Auto Tune slider still off, showing scaling policy, cluster range, Small size and a 60 second auto-suspend. One red circle marks the off slider and a second marks the savings above it, realized savings of $945.13 to $1004.61 and $12,276.13 of estimated annual savings. ## Auto Tune Rolls Back Its Changes If Performance Degrades Real-Time mode measures two things after every resize: queue time and query execution time. Its baseline is the peak from the same day the week before, so a busy Monday is judged against last Monday. | Metric | What triggers a rollback | | --- | --- | | Queue time | Queue time passes 150% of the baseline | | Query latency | Execution time passes 150% of the baseline | When either threshold is crossed, Auto Tune restores the default size and leaves that warehouse alone for the next hour. After the cool-off, it can trigger again. Every agent change goes into an audit trail: the new size, the new config, and the reasoning behind it. You can dig in to understand why a particular change was made. The Auto Tune history timeline for one warehouse, listing resize-to-Medium, resize-to-Large and suspend events by DataPilot with timestamps, one row expanded to explain a 55.82 GB local spill on a Medium warehouse, plus an event filter across scaling, suspension, size changes, backoff and config changes. ## Auto Tune Reports Savings as a Range Snowflake checks for suspendable warehouses roughly every 30 seconds, so after your auto-suspend timer fires, the warehouse actually stops sometime in the next 30 seconds. Nobody can tell in advance when. Auto Tune accounts for this by giving you two numbers: - **The best case.** The warehouse stopped the instant the timer fired. - **The likely case.** It stopped 15 seconds later, halfway through that check interval. A single number would have to assume the best case, which inflates the savings. The second number is the one that holds up in a finance review. The savings chart tracks cost and savings together, by day, week, or month. Yellow is the spend; green stacked on top is what Auto Tune took off it. Hover a bar and it shows how much of that saving came from each agent. The Auto Tune Savings chart for one warehouse across a month, with the granularity dropdown set to Daily. Spend is yellow, Auto Tune savings stack in green on top, and a tooltip splits one day's saving into a range for Warehouse Suspension plus Scaling and a smaller range for Warehouse Sizing. Across the whole account, the same numbers tell you where to start. The Infra page filters warehouses by whether Auto Tune is resizing them. Sort by annual saving and the biggest wins come to the top. The Infra warehouses list with the three Auto Tune quick filters highlighted at the top, Eligible for Resizing, Resizing Real-Time and Resizing Approval, above a table of warehouses carrying cost, size, 90% execution time, opportunities, insights and auto-suspend setting. ## What You Get with Auto Tune Left On - **Warehouse Shutdown and Scaling.** Suspends a warehouse sooner than Snowflake's own auto-suspend, with nobody filing a ticket for it. - **Warehouse Sizing.** Drops a warehouse one size during its quiet hours. You choose whether it asks you first. - **Automatic rollback.** If queue time or query time crosses your threshold, Auto Tune puts the default size back and leaves that warehouse alone for the next hour. - **Custom schedule blocks.** Force a smaller size on a window you know is quiet, or hold the default through a window you can't risk. - **Two savings numbers.** The best case and the case that usually happens. Proof Auto Tune cuts warehouse cost by **up to 20%** across our customers, without giving up the latency budget those warehouses run under. The backoff is why that holds. Cross your queue-time or latency threshold and the default size comes straight back. You do not have to take the 20% on trust. With both agents still off, the warehouse page already shows Possible Savings for that warehouse in dollars. That number is measured on your own workload. Read it and compare it against our 20% before you switch anything on. Auto Tune is part of the Enterprise Platform, and [Altimate for Snowflake](https://www.altimate.ai/use-cases/altimate-for-snowflake) is where a Snowflake team starts. ### Talk to us about your warehouse [Book a 30-min demo →](https://calendly.com/d/ctnv-rbc-yr3/altimate-ai-overview) ## Frequently Asked Questions What are the two Auto Tune agents on a Snowflake warehouse? How much does Auto Tune actually save? What makes the real-time backoff revert a resize? What thresholds suit a latency-sensitive warehouse? Why is Warehouse Sizing missing on some of my warehouses? What is the difference between Approval mode and Real-Time mode? Why are auto-suspend savings shown as a range? Where can my team see Auto Tune on our own warehouses? On this page - Auto Tune Runs Autonomous Agents with Human Guardrails - Auto Tune Rolls Back Its Changes If Performance Degrades - Auto Tune Reports Savings as a Range - What You Get with Auto Tune Left On ## Related Resources Altimate Lite for Snowflake Reduce Snowflake Idle Warehouse Spend with Altimate Lite https://altimate.ai/product-highlights/altimate-lite-snowflake-warehouse-cost Enterprise Platform Snowflake Cost Analysis in Plain English, with AI That Cites Every Number https://altimate.ai/product-highlights/snowflake-cost-analysis-ai-cited-answers Help Doc AI Cost Analysis (DataPilot) https://help.altimate.ai/platform/summary/datapilot-analysis/ Altimate Lite for Snowflake Snowflake Cortex AI Cost Breakdown by Service, User, and Model https://altimate.ai/product-highlights/snowflake-cortex-ai-cost-breakdown