How much faster is NVMe than SATA SSD in dedicated server hosting?

Why is NVMe faster than SATA?

NVMe reaches the CPU over PCIe lanes, while SATA runs over a link designed for hard drives. SATA-IO's Revision 3.0, released in June 2009, raised SATA to 6 Gb/s, and no later revision went faster. Its 8b/10b line code leaves about 600 MB/s for data.

The command model differs even more. NVM Express describes SATA devices as taking up to 32 commands in a single queue. NVMe allows up to 64K queues with 64K commands each, so every CPU core can drive its own queue without locking.

NVMe is also lighter on the CPU. NVM Express states that its command set uses less than half the CPU instructions per I/O request of SAS or SATA.

Where does the gap show up in real work?

The gap shows up wherever many small reads and writes arrive at once:

• Databases. Random reads of index and table pages, plus commit latency on every write-ahead log flush. Sizing for that case is on the best dedicated server hosting for PostgreSQL.
• Builds and CI. A checkout touches thousands of small files, and parallel jobs each keep their own queue busy. See what hardware a build server needs.
• VM hosts. Dozens of guests issue I/O at the same moment. A 32-command queue fills up quickly; 64K queues do not.
• Restores and large copies. At 550 MB/s, reading 1.92 TB takes about 58 minutes; at 5,000 MB/s, about 6.

Analytics engines feel it too: ClickHouse reads compressed columns in large parallel blocks. More in what hardware ClickHouse needs on bare metal.

When does NVMe make no difference?

NVMe makes no difference when the network, the CPU or RAM is the real limit. A 1 Gbit/s port carries 125 MB/s at most, a quarter of what one SATA SSD can read.

So a server that only streams files to visitors is bound by its port long before its drives. Every Lumenfer plan lists its port in the traffic line, for example 60 TB a month at 1 Gbit/s.

The same applies to data that already sits in the page cache. Once a working set fits in RAM, reads never reach the drive.

Does RAID 1 slow NVMe down?

RAID 1 keeps the write speed of one drive and can raise read throughput. Two-drive plans ship with the operating system on a software RAID 1 mirror; ask support before delivery if you prefer RAID 0 or two separate drives.

The Linux md driver's manual explains the mechanics. Writes go to all devices in parallel, so a write completes at the speed of a single drive. Reads come from one device at a time, and the driver spreads separate requests across the mirrors.

The same manual warns that read balancing does not make RAID 1 behave like RAID 0. One sequential reader still sees one drive; several readers can use both.

Is a SATA SSD server still worth renting?

Yes, for bulk storage, backups and archives, where capacity per dollar matters more than IOPS. SATA SSDs still beat hard drives by a wide margin on random reads.

Lumenfer's price check of 30 October 2026 found NVMe-only servers with a 2022+ CPU at 13 of 18 no-KYC crypto hosts. They made up 113 of 229 comparable plans.

The other plans mixed in SATA drives, hard drives or older CPUs. Read the storage line of any offer, not just its capacity.

How do you check the drives on a rented server?

Check the drive type, the array and the health right after delivery:

lsblk -d -o NAME,MODEL,TRAN,SIZE,ROTA      # TRAN shows nvme or sata
nvme list                                    # model, firmware, capacity (nvme-cli)
cat /proc/mdstat                             # RAID 1 state and resync progress
nvme smart-log /dev/nvme0                    # wear, temperature, media errors

# 60-second random-read test on a 4 GB file (not on the raw devices)
fio --name=randread --filename=/root/fio.test --size=4G --rw=randread --bs=4k \
    --iodepth=64 --numjobs=4 --ioengine=libaio --direct=1 --runtime=60 --group_reporting
rm /root/fio.test

What storage does each Lumenfer plan have?

Every plan has two NVMe drives in RAID 1 and a CPU with PCIe 5.0 lanes.