“NVMe” has become a headline feature in VPS marketing. NVMe drives are faster than SATA SSDs — but whether that difference shows up in your application depends on the workload. Here’s how to think about it.

The technical difference

Both are solid-state flash storage. The difference is the interface:

SATA SSD NVMe SSD
Interface SATA III (AHCI) PCIe
Max sequential throughput ~550 MB/s ~3,000–7,000+ MB/s (PCIe 3/4)
Random IOPS (4K) ~50k–100k Several hundred thousand+
Command queues 1 queue, 32 commands Up to 64k queues
Latency Low Lower

Both are vastly faster than spinning hard disks, which manage only ~100–200 random IOPS.

On a VPS, the drive isn’t the whole story

Inside a virtual machine, your storage performance also depends on:

  • Virtualization layer — VirtIO drivers, caching mode, and the storage stack on the host.
  • RAID configuration of the host’s drives.
  • Sharing — many VMs use the same drives; providers often apply per-VM I/O limits for fairness.
  • Network storage — some clouds use network-attached volumes, adding latency regardless of drive type.

So a VPS “on NVMe” doesn’t necessarily deliver raw NVMe numbers, and a well-run SATA SSD platform can be perfectly fast for most uses. Measure what you actually get.

Which workloads benefit from faster storage?

Noticeable gains:

  • Databases with working sets larger than RAM (lots of random reads).
  • Write-heavy databases and queues (many small fsync operations).
  • Builds and package installs with thousands of small files.
  • Search engines (Elasticsearch/OpenSearch), log ingestion.
  • Large file processing, backups, video work.

Little or no difference:

  • Websites with page caching — pages come from RAM.
  • Databases whose working set fits in RAM — reads hit the buffer pool, not the disk.
  • Network-bound tasks: VPN, proxies, bots, APIs waiting on other services.
  • CPU-bound tasks.

For most web workloads, enough RAM for caching has a bigger effect than the type of SSD. A database that fits its data in memory barely touches the disk for reads.

How to measure your VPS storage

fio gives realistic numbers:

sudo apt install fio
# Random 4K reads/writes, typical for databases
fio --name=randrw --filename=/root/fio.test --size=2G --bs=4k --rw=randrw --rwmixread=70 \
    --iodepth=32 --numjobs=2 --runtime=60 --time_based --direct=1 --group_reporting
rm /root/fio.test

Look at IOPS and latency percentiles (clat p99). Sequential throughput (--bs=1M --rw=read) matters for big files and backups. See the fio disk benchmark guide.

Watch real workloads with:

iostat -x 5

High %util and await with high CPU I/O wait (wa in top) indicate storage is your bottleneck. See iostat.

Making storage less important

  • More RAM for database buffers and the OS page cache.
  • Caching layers: Redis object cache, page cache, CDN.
  • Indexes so queries read less data — see database indexing.
  • Filesystem tuning: noatime, periodic TRIM — see mount options.
  • Batching writes in applications.

Bottom line

NVMe is a real advantage for storage-heavy workloads. For a typical website, app server or self-hosting box, SSD-backed storage of either type is fast enough, and RAM, caching and good queries matter more. Benchmark the plan you’re considering with fio and with your actual application.

VPS.DO plans use 100% SSD storage on KVM. With the 7-day money-back guarantee, you can run your own benchmarks before committing — see plans in USA, Hong Kong and Japan, or the Storage VPS for large disks.

  • VPS
  • Storage
  • Performance