Storage is one of the factors with the greatest impact on real-world server performance. It does not matter how much RAM or how many CPU cores your machine has: if the disk subsystem is slow, the entire chain grinds to a halt. The difference between a mechanical HDD at 7,200 RPM and a latest-generation NVMe can mean a leap from 100 MB/s to over 7,000 MB/s in sequential reads, and from 100 IOPS to over 1,000,000 IOPS in random operations.
However, faster does not always mean better for every scenario. An efficient storage strategy combines different technologies according to the workload: HDD for cold data and mass archival, SATA SSD for general-purpose applications, and NVMe for workloads demanding minimal latency and maximum throughput. In this article we analyse each technology with real data and help you choose the optimal configuration for your infrastructure.
HDD (Mechanical Drives): the workhorse of mass storage
Hard disk drives (HDDs) use magnetic platters that spin at high speed (5,400 or 7,200 RPM in consumer models, up to 15,000 RPM in enterprise SAS models) while a read/write head physically moves across the surface. This mechanical architecture imposes inherent speed and latency limitations, but offers one decisive advantage: the lowest cost per terabyte on the market.
A 3.5" enterprise HDD at 7,200 RPM delivers sequential read speeds of 150-250 MB/s and between 80-200 IOPS in random 4K operations. Typical latency is 4-8 milliseconds. Current models from Seagate Exos and Western Digital Ultrastar reach capacities of up to 24 TB per unit, making it possible to build mass storage servers with hundreds of terabytes in a single chassis.
HDD advantages:
- $ Unbeatable cost per TB: from 15-20 EUR/TB on enterprise models
- + Massive capacities: up to 24 TB per drive in 3.5" form factor
- + Excellent for sustained sequential writes (streaming, backup)
The main disadvantages are high latency, sensitivity to vibrations (especially in dense racks with many drives) and higher power consumption per usable TB. Additionally, HDDs have moving parts subject to mechanical wear, which reduces their long-term reliability compared to flash technologies. In our storage servers, we use enterprise HDDs with 2.5 million hour MTBF and anti-vibration technology to maximise durability.
SATA SSD: the balance between performance and cost
SSDs (Solid State Drives) based on NAND flash memory completely eliminate moving parts. Data is stored on non-volatile memory chips, enabling virtually instantaneous access without waiting for a head to seek into position. In their SATA version (AHCI interface), SSDs use the same connector as 2.5" HDDs, making migration extremely straightforward.
An enterprise SATA SSD such as the Samsung PM893 or the Micron 5400 PRO achieves sequential read speeds of 550 MB/s and writes of 520 MB/s, fully saturating the SATA III bus limit (6 Gbps). In random 4K operations, a SATA SSD delivers between 75,000-100,000 IOPS, representing a 500x improvement over an HDD. Latency drops dramatically to 50-100 microseconds.
The 2.5" form factor allows up to 24 units in a standard 2U server, achieving very attractive capacity and performance combinations. For most web applications, medium-sized databases (MySQL, PostgreSQL), mail servers and general virtualisation environments, a SATA SSD provides more than sufficient performance at a reasonable price. This is the option we recommend on our dedicated SME servers that need good performance without an excessive budget.
NVMe: performance without compromise
NVMe (Non-Volatile Memory Express) is not a type of memory but a communication protocol designed specifically for flash storage. While SATA SSDs use the AHCI protocol (originally designed for mechanical drives and limited to a single command queue with 32 entries), NVMe communicates directly with the CPU via the PCIe bus, supporting 65,535 queues with 65,536 commands each.
This architectural difference is what allows NVMe SSDs to achieve performance that multiplies SATA SSDs by 10 or more. A PCIe Gen4 x4 NVMe drive such as the Samsung PM9A3 or the Micron 7450 PRO delivers sequential read speeds of 7,000 MB/s and writes of 5,000-6,000 MB/s. In random operations, the figures are equally impressive: 1,000,000+ read IOPS and 600,000+ write IOPS, with latencies of just 10-20 microseconds.
The PCIe Gen5 generation, already available on the latest platforms, doubles the bandwidth again: SSDs like the Samsung PM1743 and the Kioxia CM7 reach 14,000 MB/s in sequential reads and over 2,500,000 IOPS. These numbers make Gen5 NVMe the mandatory choice for demanding workloads such as high-volume transactional databases, AI model training, HPC (High Performance Computing) and high-frequency trading.
NVMe Gen4 vs Gen5:
- Gen4 x4 = 8 GB/s theoretical bandwidth (~7,000 MB/s real-world)
- Gen5 x4 = 16 GB/s theoretical bandwidth (~14,000 MB/s real-world)
- Gen5 reduces latency by up to 30% compared to Gen4 under mixed workloads
Comparison table: HDD vs SSD vs NVMe
The following table summarises the key specifications for each technology. Values correspond to representative enterprise models on the current market.
| Specification | HDD (7200 RPM) | SATA SSD | NVMe Gen4 | NVMe Gen5 |
|---|---|---|---|---|
| Sequential read | 150-250 MB/s | 550 MB/s | 7,000 MB/s | 14,000 MB/s |
| Sequential write | 150-200 MB/s | 520 MB/s | 5,000-6,000 MB/s | 12,000 MB/s |
| Random IOPS (4K) | 80-200 | 75,000-100,000 | 1,000,000+ | 2,500,000+ |
| Latency | 4-8 ms | 50-100 us | 10-20 us | 7-15 us |
| Max capacity | 24 TB | 8 TB | 8 TB | 32 TB |
| Price per TB | 15-25 EUR | 60-100 EUR | 80-150 EUR | 150-300 EUR |
| Power consumption | 6-10 W | 2-4 W | 5-8 W | 10-25 W |
| Endurance (DWPD) | N/A | 1-3 DWPD | 1-3 DWPD | 1-3 DWPD |
| Ideal use case | Backup, archive, cold storage | Web, apps, mid-size DB | Heavy DB, VMs | AI, HPC, trading |
Use cases: when to choose each technology
HDD: backup, archive and cold storage
HDDs remain the most cost-effective option for storing large volumes of infrequently accessed data. They are ideal for backup repositories with S3 Object Storage, log archival, surveillance video storage and cold storage in general. A storage server with 12 x 20 TB drives in RAID 6 provides approximately 200 TB of usable capacity at a very low storage cost.
SATA SSD: web applications and mid-size databases
The SATA SSD is the sweet spot for most enterprise workloads. Web servers running WordPress, Magento or PrestaShop, MySQL/PostgreSQL databases up to several hundred gigabytes, mail servers and general virtualisation environments all benefit enormously from the HDD-to-SSD upgrade without requiring the additional cost of NVMe.
NVMe: mission-critical databases, AI and HPC
When every microsecond counts, NVMe is the only option. Transactional databases such as Oracle, SQL Server or MongoDB handling thousands of queries per second, Elasticsearch clusters with real-time search, machine learning model training reading massive datasets randomly, and algorithmic trading platforms requiring minimum latency are all scenarios where NVMe makes a measurable difference to business performance.
RAID configurations by drive type
RAID configuration is a critical complement to the drive type. Each RAID level offers a different balance between performance, usable capacity and fault protection:
- check_circle RAID 1 (Mirror): duplicates data across two drives. Ideal for the operating system on SSD or NVMe. Loses 50% of capacity but provides maximum safety and read speed.
- check_circle RAID 5 (Distributed parity): requires a minimum of 3 drives, tolerates 1 failure. Good balance for general-purpose HDD storage. Usable capacity is N-1 drives.
- check_circle RAID 6 (Double parity): tolerates 2 simultaneous drive failures. Recommended for large HDD arrays (8+ drives) where a rebuild can take hours. Usable capacity: N-2.
- check_circle RAID 10 (Mirror + Stripe): combines RAID 0 speed with RAID 1 safety. Optimal for NVMe on high-performance databases. Loses 50% of capacity but maximises IOPS and minimises write latency.
Practical recommendation:
For a database server with NVMe, the typical configuration is 2x NVMe in RAID 1 for the OS + 4x NVMe in RAID 10 for data. For a mass storage server with HDD, we recommend 12x HDD in RAID 6 with a hot spare, combined with 2x SSD in RAID 1 for the operating system.
Storage at EasyDataHost
At EasyDataHost we offer all storage options, tailored to every need:
Storage Servers (HDD)
From 288 TB to 2 PB+ on our storage servers. Ideal for mass backup, archival and cold storage with 20-24 TB enterprise drives.
NVMe Servers
All our dedicated servers are available with PCIe Gen4 and Gen5 NVMe for high-performance workloads. Configurations from 2x 1 TB to 8x 8 TB in RAID 10.
S3 Object Storage
Our S3 storage service combines HDD for capacity and NVMe for metadata, delivering high throughput with no traffic or request charges.
Synology NAS
Managed NAS solutions with SSD cache and mass HDD storage. From 100 TB to 1 PB with replication and backup included.
Conclusion: choose according to your workload
There is no single storage type that is best for every scenario. The key is to analyse the data access pattern of your application and size the storage accordingly:
- arrow_right Cold data, archival and mass backup: HDD in RAID 5/6.
- arrow_right Web, apps and mid-size databases: SATA SSD or entry-level NVMe.
- arrow_right Transactional databases, Elasticsearch, VDI: NVMe Gen4 in RAID 10.
- arrow_right AI, HPC and high-frequency trading: NVMe Gen5 at maximum density.
- arrow_right Hybrid combinations: use NVMe for hot data and HDD/S3 for cold data, with automatic tiering.
If you need advice on choosing the optimal storage configuration for your project, our engineering team can help you design a bespoke solution. Get in touch and we will analyse your case at no obligation.