MBR vs. GPT: partitioning very large disks
Kort antwoord
MBR (Master Boot Record) is the older partitioning scheme: it supports at most 4 primary partitions and can address at most 2 TiB of a disk, regardless of how much larger the disk actually is. GPT (GUID Partition Table) is the modern replacement, supports far more partitions and addresses disks well beyond that limit. Any disk larger than 2 TiB needs GPT, MBR silently wastes whatever capacity sits past the 2 TiB mark rather than raising a clear error.
What a partition table actually is
A partition table sits at the start of a disk and records how that disk is divided into partitions: where each one starts, how big it is, and what type it's marked as. Before a filesystem can be created and mounted, a partition has to exist for it to live in. MBR and GPT are the two schemes in common use for that partition table, and they differ in what they can address and how much information they can store.
MBR: the older scheme, with a hard ceiling
MBR dates back to the early era of PC BIOS and stores partition information in a single sector at the start of the disk. That compact layout brings two practical limits:
- At most 4 primary partitions. Working around this with extended and logical partitions is possible but adds complexity.
- A 2 TiB addressing limit. MBR records partition sizes using 32-bit sector addresses with a fixed sector size, and the arithmetic runs out at 2 TiB. A disk larger than that can be attached to a server without issue, but MBR simply can't reference storage past the 2 TiB mark.
That second point is the one that catches people out. Partitioning a 4 TiB disk with MBR doesn't produce an error. It produces a disk that appears to partition and format normally, with roughly half its actual capacity permanently out of reach. The space beyond 2 TiB isn't unusable in some recoverable sense, it's simply invisible to a scheme that has no way to address it. This is a common and confusing mistake precisely because nothing about the process looks wrong at the time.
GPT: the modern replacement
GPT (GUID Partition Table) replaces MBR's single sector with a more robust structure, including a backup copy of the partition table stored at the end of the disk. In practical terms, GPT removes both of MBR's limits: it supports far more partitions (128 by default under the common implementations), and it addresses disks well beyond any capacity in current use, using 64-bit sector addressing rather than MBR's 32-bit scheme. Once a disk exceeds 2 TiB, GPT isn't just the better option, it's required to use the full disk at all.
The link to boot mode
Partitioning scheme and firmware boot mode are separate concepts, but they're tied together in practice. Legacy BIOS boot generally expects an MBR partition table to find its bootloader, while UEFI is built around GPT. In practice that means a disk you intend to boot from is normally partitioned to match the boot mode the server actually uses: GPT for UEFI, MBR for legacy BIOS. See What happens when a server boots for how BIOS/UEFI and the boot chain fit together, including why UEFI has effectively replaced legacy BIOS on current hardware.
A non-boot data disk, one that's just mounted for storage rather than holding the OS the server starts from, doesn't have this constraint in the same way, but there's rarely a reason to choose MBR for a new data disk either. GPT has no meaningful downside on modern systems, and choosing it upfront avoids ever hitting the 2 TiB ceiling if that disk is resized or replaced with a larger one later.
Choosing partitioning scheme in practice
As a working rule: use GPT unless you have a specific reason not to, and treat MBR as something you'd only keep for compatibility with an existing legacy BIOS setup. When you're deciding on storage capacity and type in the first place, see Choosing storage media: NVMe vs. SSD vs. HDD by workload, partitioning scheme is a downstream decision once you know how large a volume you're actually working with.