Buying an NVMe SSD is mostly an exercise in ignoring the number on the box. Every
drive here is rated somewhere between 6,000 and 15,000 MB/s, and almost nobody
reading this will ever see the difference between them in normal use. The things
that separate a good drive from a bad one are less advertised: whether it holds
its speed when it is nearly full, whether it has a cache of its own or borrows
your system memory, and whether the firmware has had enough years of attention
to be trustworthy.
What follows is eight drives ranked with that in mind, from the one that has the
fewest compromises to the one that makes the most sensible compromises for the
money.
How these were judged
Nothing here was tested on a bench by this site. There is no lab, no pile of
review samples, and no equipment to measure sustained throughput or power draw.
Pretending otherwise would be the easiest thing in the world and it would be a
lie, so: every speed figure below is the manufacturer’s own published
specification, and every judgement about how a drive behaves in sustained use
comes from reading across published testing rather than from running it here.
Where that evidence is thin, the entry says so instead of inventing a winner.
The affiliate disclosure explains what this site earns
and what it does not influence.
What actually matters
Sustained write speed, not peak. Almost every consumer SSD writes into a
fast SLC cache first and moves the data to slower storage afterwards. While that
cache has room, the drive hits its advertised number. When it runs out — a large
game install, a video export, a big file copy — the drive falls back to its real
write speed, which can be a fraction of the figure on the box. This is the single
biggest difference between a cheap drive and an expensive one, and it is the
number nobody prints.
How full you keep it. SSDs get slower as they fill, because the cache the
drive uses to absorb writes is carved out of free space. A 1TB drive kept at 95%
capacity is meaningfully slower than the same drive at 50%. This is a better
argument for buying 2TB than any benchmark.
Whether it has DRAM. A drive needs somewhere to keep the map between logical
addresses and physical flash. Drives with their own DRAM keep it on board; drives
without it use Host Memory Buffer, borrowing a small slice of system RAM over the
PCIe bus. HMB has got good enough that most people will never notice, and half
this list uses it. It still gives ground under sustained load.
PCIe 5.0: who it is actually for
Gen5 drives roughly double sequential bandwidth, and for gaming that is close to
irrelevant. Games load by issuing many small random reads, which is the workload
sequential bandwidth helps least. Independent testing has consistently found
load-time differences between good Gen4 and Gen5 drives measured in fractions of
a second.
Gen5 earns its price when you routinely move very large files — video work,
large datasets, machine-learning checkpoints. It also demands a PCIe 5.0 slot and
real cooling, and it draws more power. If you are building a machine that does
that work, the 9100 PRO is the drive. If you are building a gaming PC, spend the
difference on capacity.
A note on heatsinks
Most desktop motherboards now ship with M.2 heatsinks, and for a Gen4 drive that
is enough. Buy the bare version and use the board’s cooler rather than paying
for a heatsink you will remove. The exceptions are real: a PS5 requires one, a
Gen5 drive needs one, and a laptop or handheld has room for neither — which is
why efficiency, not peak speed, is the specification that matters there.
