SAS vs SATA vs NVMe: Choosing Drives for Your Server Build

By kciservers.net Technical Team · 2026-07-27

You are configuring a server, and the storage section of the spec sheet is asking you to pick a drive type. SAS, SATA, NVMe -- three options, and the wrong one either wastes your budget or bottlenecks your workload. The good news is that the choice is not as open-ended as it looks. Your server's backplane and RAID controller already narrow the field. The question is which of the remaining options fits what you are actually going to run.

Here is how to think about it, from someone who configures these servers every day.

SATA: The Capacity and Value Play

SATA is the interface you already know. It runs at 6 Gb/s per lane -- about 600 MB/s of practical throughput after encoding overhead. It is half-duplex, meaning a drive can either read or write at a given moment, not both. It is single-port: one path from the controller to the drive, no redundancy.

None of that disqualifies SATA for server use. What it means is that SATA is the wrong pick when you need maximum IOPS or path redundancy, and the right pick when you need cheap capacity.

SATA drives come in two flavours that matter for server builds:

The practical constraint with SATA in enterprise servers is backplane compatibility. SATA drives plug into SAS backplanes -- a SAS controller can speak SATA -- but SAS drives will not work in a SATA-only backplane. So if your chassis has a SAS backplane, you can mix SATA and SAS drives freely. If it has a SATA-only backplane, you are locked to SATA.

SAS: The Enterprise Default

SAS is the interface that enterprise servers were built around. It runs at 12 Gb/s per lane in its most common current form, with a 24 Gb/s standard available on newer platforms. It is full-duplex, meaning simultaneous read and write. And it is dual-port: every SAS drive has two independent data paths, which means two controllers (or two paths from one controller) can address the same drive. If one path fails, the other takes over. In a dual-controller storage array, that is not a nice-to-have -- it is the reason the array survives a cable pull or a controller failure.

SAS drives also report more detailed SMART data, support longer error-recovery cycles (TCQ/SCSI command queuing), and are built to higher endurance and MTBF specs than consumer SATA drives. The trade-off is cost: SAS drives carry a premium over equivalent-capacity SATA drives, and the controllers that speak SAS (PERC, Smart Array, MegaRAID) cost more than basic SATA HBAs.

Where SAS earns its premium:

The backward compatibility point from earlier works in your favour here: a SAS backplane accepts both SAS and SATA drives. This is why most enterprise 2.5" chassis ship with SAS backplanes by default -- it gives you the flexibility to populate with cheap SATA for capacity or SAS for performance without changing the backplane.

NVMe: The Performance Tier

NVMe is not really a "drive type" in the same sense as SAS and SATA -- it is a protocol that runs directly over PCIe, bypassing the SCSI and ATA command layers that SAS and SATA use. The practical effect is lower latency, dramatically higher queue depth, and bandwidth that scales with PCIe generation rather than capping out at 12 Gb/s.

The numbers tell the story. SATA tops out at 6 Gb/s (~600 MB/s). SAS tops out at 12 Gb/s (~1,200 MB/s) on current-generation backplanes, with 24 Gb/s available on newer platforms. NVMe on a PCIe Gen4 x4 link can deliver around 8 GB/s of theoretical bandwidth, with real-world sequential reads in the 6–7 GB/s range. PCIe Gen5 x4 doubles that ceiling to roughly 16 GB/s. The queue depth difference is even more stark: SATA supports 32 outstanding commands and SAS a few hundred, while NVMe supports up to 65,535 queues with 65,535 commands per queue. For random-read-heavy workloads -- databases, virtualization, real-time analytics -- that parallelism is where the performance gain actually comes from.

NVMe drives in enterprise servers come in two physical forms:

The constraint with NVMe is that it requires a backplane wired for PCIe, not just SAS/SATA. You cannot drop an NVMe drive into a SAS-only backplane. On many modern servers, "universal" backplanes accept any of the three drive types -- but you need to confirm that before you buy, because a SAS/SATA-only backplane is still common on capacity-oriented chassis configurations.

The Backplane Constraint: Your Chassis Picks the Fight

Here is where the theory meets the metal. The drive type you can use is not a free choice -- it is constrained by the backplane configuration your server shipped with (or the one you select when ordering). Let's look at how this plays out across real platforms.

The Dell PowerEdge R740, a 14th-generation 2U rack server, ships in three front-bay configurations: 8x 2.5", 8x 3.5", and 16x 2.5". All three are SAS/SATA only -- none of them are NVMe-capable. If you buy an R740, your drive choice is between SAS and SATA. You cannot add front-bay NVMe drives without stepping up to a newer platform. The 8x 3.5" configuration supports up to 128 TB of raw capacity; the 16x 2.5" supports up to 122.88 TB. For a capacity-oriented build on the R740, that is the ceiling.

Step up one generation to the Dell PowerEdge R750, a 15th-generation 2U server, and the picture changes. The R750 offers an 8x 2.5" Universal backplane that accepts SAS, SATA, and NVMe drives in any bay. It also offers dedicated NVMe-only configurations (8x 2.5" NVMe, 16x 2.5" NVMe, 24x 2.5" NVMe) and mixed configurations like 24x 2.5" SAS/SATA with 8 universal NVMe slots. The 24x 2.5" NVMe configuration supports up to 720 TB of raw capacity. The 16x 2.5" NVMe configuration supports up to 480 TB. This is the flexibility you get from a newer platform: the same chassis can be a bulk SATA storage box or an all-NVMe performance machine, depending on which backplane you specify.

The Dell PowerEdge R760, a 16th-generation 2U server, goes further. It adds EDSFF E3.S support -- a 16x E3.S NVMe configuration that supports up to 1,024 TB of raw capacity in a 2U chassis. It also offers universal backplanes (8x 2.5" SAS/SATA/NVMe, 24x 2.5" with 8 universal slots) alongside dedicated NVMe and SAS/SATA configurations. The R760's 12x 3.5" SAS/SATA configuration supports up to 288 TB, while its 24x 2.5" SAS/SATA configuration supports up to 720 TB. The point is that on a 16G platform, you have the full spectrum: SATA for capacity, SAS for the enterprise middle ground, and NVMe (including E3.S) for performance -- all in the same 2U footprint, selected at the backplane level.

On the HPE side, the ProLiant DL380 Gen10 tells a similar story. Its SFF (2.5") chassis reaches NVMe not through a single tri-mode backplane but by fitting a dedicated 8 NVMe cage, or a Premium 6 SAS/SATA + 2 NVMe cage, in place of a drive box -- up to 20 NVMe drives in total. The 24 SFF configuration reaches up to 459 TB of raw capacity with 6 rear 2.5" bays. But the 8 LFF (3.5") and 12 LFF configurations are SAS/SATA only -- no NVMe. The 12 LFF supports up to 311.68 TB of raw capacity. If you need NVMe on the DL380 Gen10, you need an SFF chassis. The LFF chassis is a capacity play, and HPE did not wire it for PCIe.

The pattern is consistent across vendors: 3.5" chassis are SAS/SATA only (capacity), 2.5" chassis on newer platforms are often universal or NVMe-capable (performance), and the generation of the server determines whether "universal" is even an option.

The Controller Has to Match

The drive type you choose also dictates the controller you need -- and this is where refurb buyers get caught.

SAS and SATA drives need a SAS controller or HBA. On Dell 14G servers, that is typically a PERC H330 (entry-level, no cache, RAID 0/1/5/10 only) or a PERC H740P (8 GB cache, full RAID support). On Dell 15G and 16G servers, the options expand to include the PERC H755, PERC H755N (NVMe-specific, 8 GB cache), and the PERC H965i. On HPE ProLiant Gen10, the Smart Array line runs from the S100i SR (software RAID, no hardware cache) through the P408i-a and P408i-p (hardware RAID with cache) up to the P816i-a for high-density configurations. Broadcom MegaRAID options (MR416i, MR216i) are also supported on the DL380 Gen10.

NVMe drives are different. In their simplest form, NVMe drives do not need a RAID controller at all -- they connect directly to the CPU's PCIe lanes through the backplane, and the operating system sees them as individual PCIe devices. This is called "pass-through" or "non-RAID" mode. If you want hardware RAID on NVMe drives, you need a controller that speaks NVMe -- like the Dell PERC H755N or the PERC H965i. On the R750, the 16x 2.5" NVMe RAID configuration requires dual PERC H755N controllers, each driving 8 bays. On the R760, the H965i serves the same role. These controllers are not interchangeable with SAS/SATA PERCs -- an H740P will not manage NVMe drives.

For homelab and ZFS users, the controller question has an additional layer. If you are running TrueNAS, Proxmox with ZFS, or any software-defined storage, you want an HBA (host bus adapter) in IT mode -- passthrough, no hardware RAID. The drive presents as a raw block device to the OS, and ZFS handles redundancy. A Dell HBA330 or HBA355i in IT mode, or an HPE HBA in HBA mode, does this. Do not buy a hardware RAID controller if you are running ZFS -- you will pay for cache and a battery that ZFS does not use, and you will have to flash the card to IT mode anyway.

Which Drive Type for Which Workload

Here is the decision framework, stripped down:

Bulk storage / backup / archive / media: SATA HDDs in 3.5" bays. Cheapest per terabyte, and the workload does not need IOPS. A 12x 3.5" SAS/SATA chassis on an R750 gives you up to 216 TB; the same configuration on an R760 gives you up to 288 TB. On the DL380 Gen10, the 12 LFF configuration gives you up to 311.68 TB.

General-purpose virtualization / small business servers: SAS SSDs or high-end SATA SSDs in 2.5" bays. If your chassis supports NVMe and the budget allows, NVMe is better -- but SAS SSDs on a 12 Gb/s backplane will handle a dozen VMs without breaking a sweat, and they cost less.

Transactional databases / high-IOPS workloads: NVMe SSDs. The queue depth and latency advantage is not marginal -- it is an order of magnitude. If your platform supports NVMe (R750, R760, DL380 Gen10 SFF), use it. If it does not (R740, DL380 Gen10 LFF), SAS SSDs are your best option, and you should consider stepping up to a newer platform if the workload truly demands NVMe.

Mixed workloads (capacity + performance): A universal backplane with SATA HDDs for bulk and NVMe SSDs for hot data. The R750's 24x 2.5" SAS/SATA with 8 universal slots, or the R760's 24x 2.5" with 8 universal slots, is designed for exactly this. Put your hot data on the NVMe drives and your cold data on the SATA/SAS drives, all in the same chassis.

Homelab / ZFS / software-defined storage: SATA or SAS drives behind an IT-mode HBA. The interface does not matter much here -- ZFS cares about raw capacity and redundancy, not protocol. SATA HDDs for a storage pool, maybe a couple of SATA or SAS SSDs for L2ARC or SLOG. Do not overspend on NVMe unless you have a specific workload that benefits from it.

The Bottom Line

The drive type you choose is constrained before you ever look at a price list. Your server's generation determines whether NVMe is an option. Your backplane configuration determines which drive types fit which bays. Your controller determines whether you run hardware RAID, software RAID, or passthrough. Once you know those three things, the choice between SAS, SATA, and NVMe usually makes itself -- based on what your workload actually needs and what your chassis actually supports.

The most expensive drive is the one that does not fit your backplane. Figure out the backplane first.

*Written by kciservers.net Technical Team*

All posts · Configure a server online