TechCompare LogoTechCompare

Hot vs warm vs cold storage: how to tier your data for cost and performance

Tier your storage. Keep hot data on NVMe, warm data on HDD or cheap object storage, and cold data in archival. The cost difference between NVMe and archival is roughly 50-100x per GB. For a 100 TB dataset, correct tiering can save thousands per month.

Not all data needs the same storage performance. Tiering puts frequently accessed data on fast, expensive media (NVMe SSDs) and rarely accessed data on slow, cheap media (HDDs, tape, cloud archival). Getting this right saves money without sacrificing performance. Getting it wrong means either overspending on flash you don't need or making users wait for data that should be hot.

By TechCompare · Updated

Hardware tier
Storage
Persistent storage devices
Topic focus
Storage tier guide
storage-tiers
Try the live Latency Visualizer. See how a file moves across CPU cache, RAM, and storage.
Edit in the full tool

How this is calculated

Hot tier: NVMe SSD, data accessed multiple times per day. Database primary storage, active project files, container images. Warm tier: HDD or S3 Standard, data accessed weekly to monthly. Older logs, completed project archives, backup sets less than 30 days old. Cold tier: S3 Glacier or LTO tape, data accessed yearly or less. Compliance archives, raw data from completed research projects, backups older than 90 days. The latency difference between hot and cold can be 100,000x (100 µs vs 10+ seconds for retrieval from deep archive). Cloud providers automate tiering with lifecycle policies. On-premises, tools like automatic storage tiering in ZFS and Storage Spaces do the same.

Verdict

The 100,000x latency gap between hot and deep-archive tiers is what drives the economics. Hot on NVMe serves multi-daily access, warm on HDD or S3 Standard handles weekly-to-monthly loads, and cold on S3 Glacier or LTO tape covers yearly compliance pulls. Cloud lifecycle policies move data between tiers automatically, and on-prem ZFS and Storage Spaces do the same. Mixing tiers wrong either overspends on flash or stalls users on cold retrieval.

More Latency scenarios

L1 vs L2 cache
L1 cache is the fastest memory in a computer, typically 1 ns latency (3-5 CPU cycles) and 32-64 KB per core.
View details ➜
L3 cache vs RAM
L3 cache (also called Last Level Cache or LLC) is shared across all cores in a CPU chiplet, typically 16-96 MB, with latency of 10-15 ns.
View details ➜
DDR4 vs DDR5 latency
DDR5 roughly doubles the peak bandwidth of DDR4 (from ~50 GB/s to ~100 GB/s per module), but true latency measured in nanoseconds is nearly unchanged.
View details ➜

Related guides

RAID 5 vs RAID 6 parity
RAID tiers that map to storage-tier choices.
View guide ➜
Memory latency numbers
The latency ladder these storage tiers sit beneath.
View guide ➜

Frequently asked questions

What belongs in hot, warm, and cold storage?
Hot: data touched multiple times a day - database primary storage, active projects, container images - on NVMe. Warm: weekly-to-monthly data like older logs and recent backups, on HDD or S3 Standard. Cold: yearly-or-less data like compliance archives on S3 Glacier or LTO tape. The tier is set by access frequency, not by age alone.
How much money does storage tiering actually save?
The per-GB cost gap between NVMe and deep archival is roughly 50-100x, so a 100 TB dataset parked entirely on fast storage wastes thousands per month against a correctly tiered layout. Cloud lifecycle policies automate the aging rules, and on-prem, ZFS auto-tiering and Storage Spaces do the same job between a flash pool and spinning disks.
What happens if you put cold data on hot storage?
You overspend silently - a year's worth of untouched logs on NVMe costs 50-100x what Glacier would. The reverse mistake hurts more visibly: putting hot data on cold archive makes users wait 10+ seconds per retrieval, a 100,000x latency penalty they can feel. Both errors are why tiering policy belongs in the architecture doc, not in ad hoc decisions.

Related tools

RAM Latency Calculator

Convert DDR3/DDR4/DDR5 timings (CL, tRCD, tRP, tRAS) into true latency in nanoseconds.

Use tool ➜

RAID Calculator

Calculate usable capacity and fault tolerance for RAID 0, 1, 5, 6, and 10.

Use tool ➜

Display Bandwidth Calculator

Check if your HDMI/DP cable supports your resolution and refresh rate.

Use tool ➜