VPS SSD storage varies wildly: shared SATA SSD (cheap, oversubscribed), dedicated NVMe (fast, isolated), or local NVMe with no redundancy (fastest, data loss on hardware failure). The team that runs a database picks dedicated NVMe with provisioned IOPS; the team that runs a static site is fine with shared SATA SSD. The team that picks based on cost alone gets whatever the provider allocated, which may be oversubscribed shared storage with unpredictable performance.
Table of contents
- The three tiers of VPS SSD
- IOPS, throughput, and latency
- What VPS providers actually offer
- When SSD matters vs when it doesn’t
- FAQ
The three tiers of VPS SSD
Shared SATA SSD: cheap, oversubscribed. Multiple VPS instances share the same physical SSD. The team’s neighbors can slow down the team’s I/O. The team that uses shared SSD for a static site or small web app gets acceptable performance; the team that uses it for a database has unpredictable query times.
Dedicated NVMe: fast, isolated. Each VPS has its own NVMe drive or a slice of one with reserved IOPS. The team that uses dedicated NVMe has consistent performance - the neighbors cannot slow down the team’s I/O. The team that runs a small-to-medium database on dedicated NVMe has good performance.
Local NVMe: fastest, no redundancy. NVMe directly attached to the VPS with no replication. The team that uses local NVMe has the best performance (millions of IOPS) but data loss on hardware failure. The team that runs a database on local NVMe must implement application-level replication (PostgreSQL streaming replication, MySQL Group Replication) for durability.
IOPS, throughput, and latency
IOPS (I/O Operations Per Second): how many reads/writes per second. SATA SSD: 5,000-50,000 IOPS. NVMe SSD: 100,000-1,000,000+ IOPS. The team that has a database with high write rate needs high IOPS; the team that has a read-mostly workload needs less.
Throughput (MB/s): how much data per second. SATA SSD: 200-500 MB/s. NVMe SSD: 1,000-7,000 MB/s. The team that has a workload with large sequential reads (data warehouse, log processing) needs high throughput.
Latency (microseconds): time per operation. SATA SSD: 100-500 microseconds. NVMe SSD: 20-100 microseconds. The team that has many small random reads (OLTP database) needs low latency; the team that has large sequential reads is throughput-bound, not latency-bound.
What VPS providers actually offer
Entry-level VPS tiers ($5-20/month): shared SATA SSD or shared NVMe. The team that picks the cheapest tier gets shared storage. Acceptable for low-traffic sites, dev environments, small apps. Not acceptable for databases with significant load.
Mid-tier VPS ($20-100/month): dedicated SSD or NVMe, varying IOPS. The team that runs production databases on mid-tier VPS picks providers that specify ‘dedicated SSD’ or ‘NVMe’ explicitly. The team that picks based on disk size alone may end up on shared storage with poor performance.
High-performance VPS ($100+/month): dedicated NVMe with provisioned IOPS. The team that needs database performance on a VPS picks high-performance tiers or dedicated database hosting (e.g., DigitalOcean’s dedicated database, Linode’s dedicated database, AWS RDS).
Cloud-managed databases (RDS, Cloud SQL, Azure Database): backed by NVMe with replication. The team that runs production databases on managed services does not worry about the underlying storage - the provider handles replication, backups, and patching. Cost is higher than VPS; operational complexity is lower.
When SSD matters vs when it doesn’t
Matters for: databases, large file processing, anything I/O-bound. The team that runs PostgreSQL, MySQL, MongoDB on a VPS sees a clear performance difference between SATA SSD and NVMe. Queries that take 1 second on SATA SSD may take 200ms on NVMe. The team that benchmarks before committing picks the right tier.
Does not matter for: static sites, simple web apps, low-traffic services. The team that serves a static site or a small web app with low write volume gets acceptable performance on any SSD tier. The bottleneck is usually network or CPU, not disk.
Bench before committing. Most VPS providers offer trial periods or hourly billing. The team that runs sysbench or fio on the tier before committing knows the actual IOPS, throughput, and latency. The team that picks based on spec sheet may get different performance in practice.
FAQ
Is VPS SSD as fast as dedicated server SSD?
VPS SSD is virtualized - the team shares the underlying SSD with other VPS instances (or has a dedicated slice). Dedicated server SSD is direct hardware access. The team that runs benchmarks on both sees VPS slightly slower due to virtualization overhead; the team that uses dedicated SSD has predictable, maxed-out performance.
What is the difference between SSD and NVMe?
SSD is the storage technology (flash memory). NVMe is the interface/protocol (PCIe-based, designed for flash). NVMe SSDs are faster than SATA SSDs because the interface matches the storage. The team that runs high-performance workloads picks NVMe; the team with light I/O is fine with SATA SSD.
Should I use local SSD or network SSD?
Local SSD (directly attached): fastest, no network latency, but data lost on hardware failure. Network SSD (EBS, Azure Disk, GCP Persistent Disk): replicated, can be detached/reattached, but adds network latency. The team that needs the best performance picks local with application-level replication; the team that needs simplicity picks network SSD.
What is provisioned IOPS?
Provisioned IOPS is a cloud provider feature (EBS io1/io2, Azure Premium SSD) where the team pays for a specific IOPS level regardless of actual usage. The team that needs consistent database performance picks provisioned IOPS; the team that has bursty workloads picks general-purpose SSD.
How do I benchmark VPS storage?
Use fio (flexible I/O tester) or sysbench. Example: fio --name=randwrite --ioengine=libaio --direct=1 --filename=/tmp/test --bs=4k --size=1G --rw=randwrite --numjobs=4 --runtime=60 --group_reporting. Reports IOPS, throughput, latency. The team that benchmarks before committing picks the right tier.
Is SSD storage more reliable than HDD?
Yes, in most metrics. SSDs have no moving parts (failure mode: wear-out, controller failure). HDDs have moving parts (failure mode: mechanical failure, head crash). MTBF for SSDs is typically 1.5-2 million hours; HDDs are 0.5-1 million hours. The team that picks SSD for reliability picks the right answer; the team that picks SSD for performance alone picks the right answer for a different reason.
How this fits the rest of the stack
For a sense of what the full project costs before it commits, the RunxBuild hosting calculator shows the line items together. The API, the database, the storage, the worker, the bandwidth - each one is a separate number, and the team’s mental model for the platform is the sum of those numbers.
Useful related references: