Samsung 960 Pro M.2 NVMe 1TB SSD Review

Test machine

For this review I will be using a computer with the following


  • Motherboard: Asus Z170 Deluxe (Intel Z170 chipset)
  • Processor: Intel 6th generation Core i7 6700K
  • CPU cooler: BeQuiet Dark Rock Pro 2
  • RAM: 16GB Corsair Vengeance LPX 2666MHz DDR4 (dual channel)
  • GFX: EVGA GTX 1060 SC 6G
  • Sound: Onboard Realtek ALC1050 HD audio controller
  • Hard disk OS: Toshiba OCZ RD400 M.2 NVMe
    512GB SSD
  • Case: Antec Performance One P280
  • PSU: Antec True Power modular 550W
  • Display: Dell P2715Q 27” 4K widescreen IPS LCD (HDCP compliant)
  • Operating System: Windows 10 Professional 64bit
  • Power consumption testing equipment: Quarch Technology QTL1824-02 XLC
    Programmable Power Module

The NVMe drivers used to test the Samsung 960 PRO M.2 NVMe SSD
were Windows 10 native NVMe drivers.

CPU power saving states was disabled for consistency, and
all the SSDs in this article were tested with all CPU power saving states

Test applications

To test the performance of the Samsung 960 PRO M.2 NVMe  SSD,
I will be using the following test applications in this review.

Test procedures

I will start off our testing procedures explanation by
stating that I did not run many synthetic benchmarks on the Samsung 960 PRO M.2
NVMe SSD. You may ask why I have run so few synthetic benchmarks?

SSD technology has moved so fast in the last couple of years,
that basic synthetic benchmarks alone are now of very limited use, as they don’t
really tell us much about performance and how the drive will behave in the real
world. I have therefore decided to show some basic benchmarks of the Samsung 960
PRO M.2 NVMe SSD, and will complement this with advanced benchmarks using
IOMeter and AS SSD benchmark. I will also show how the Samsung 960 PRO M.2 NVMe
SSD performs in the real world with our Myce Reality Suite test.

The reality of SSD performance

Whilst I can easily show you which SSD is technically the
faster, when you use one of these modern SSDs as an operating system drive it
becomes very difficult to tell them apart as far as performance is concerned.

A typical use of a small capacity SSD at the moment is to
have your operating system and applications installed onto the SSD. The performance
difference compared to a traditional HDD is enormous, however when you start to
compare SSD to SSD the difference becomes almost impossible to detect.

Let’s look at why this is the case.

Drive A can boot to the desktop in 8.11 seconds, and drive B
can boot to the desktop in 8.12 seconds, the difference in time is
milliseconds, and can one really tell the difference?

The fact is, all modern SSDs are only ticking over when they
are only running the OS and launching applications, it’s only when you get to
some of the larger capacity SSDs, with enough free space to be able to hold the
actual data that you’re going to be working with, be that video, audio or
pictures, for example, that you actually get a tangible difference in
performance. This is where the SSDs with the better sequential performance start
to pull well ahead of the SSDs which have lower sequential read/write

Small file random IOPS vs sequential performance


This is a fairly complex subject, but I will do my best to
explain things in a manner that is easy to understand.

The term IOPS is the amount of input or output transactions
that can take place in a one second interval, so for example, if an SSD is
quoted as being able to cope with 20,000 4K random write IOPS, then the SSD
should be able to cope with 20,000 input transactions in a period of one
second. If the same SSD is said to be able to produce 20,000 4K random read
IOPS, then the same SSD should be able to produce 20,000 4K random read output
transactions in a one second interval.

Ok, now we have some figures to work with, the next question
is how many IOPS are actually required?

This will depend on your usage pattern. If you are a typical
desktop user who browses the internet, does some word processing or perhaps some
audio or video editing, and perhaps plays a few games, then in actual fact, you
don’t need to have massive 4K random read/write performance. The actual amount
of 4K random performance that is required for a fast and smooth running system
for a desktop user with a usage pattern similar to the above will be well under
1,000 4K IOPS.

On the other hand, if the SSD is being used for running a
large and complex database server, then 4K random performance is the absolute
measurement of how fast that server will run, as this type of application does
most of its input and output transactions in the 4K domain.

So why would I need an SSD with 80,000 4K IOPS for a

In fact you don’t need this type of performance for a
desktop, but an SSD which is capable of coping with 80,000 4K IOPS will be
faster than an SSD which can only cope with 20,000 4K IOPS.

OK, I just said if under 1,000 4K IOPS are actually required
for typical desktop usage, why is an SSD with 80,000 4K IOPS faster than an SSD
with only 20,000 4K IOPS, confused?

You may ask, if I only require 1,000 4K IOPS surely the rest
is wasted?

While you may never need 80,000 4K IOPS, IOPS is all about
latency. The reason that an SSD can cope with as much as 80,000 4K IOPS is
because latency in this domain is very low. With 4K files, even if you require
to process 500 of them at the same time, you are not talking about a huge
amount of data, it has far more to do with how long it takes the SSD to process
a single file, and the amount of time required to process a single 4K is all
about how long it takes for the SSD to access or store that data before it can
move on to the next transaction.

In other words an SSD with 80,000 4K IOPS performance will
handle those 500 files faster than the SSD with 20,000 IOPS.

So how will a desktop user even notice this faster speed if
so little 4K random IOPS and data are actually used?

Multitasking is a good example. The more tasks you run at
the same time, you more you will notice the speed difference.

Sequential performance

I have always maintained that sequential performance was
every bit as important as small random file performance for a desktop SSD. To
me this was always so obvious for a desktop user. For example, let’s say you
want to launch an application or game. Both have some fairly large files to
load, and also a great many small files, but the point is, even the smaller
files are sequential in nature. Now let’s say you’re into audio or video
editing. Video files tend to be huge, and the files are written or read sequentially.
Isn’t this how many users are using their PCs these days?


So how does this shape up in the real world? Which is
better, massive 4K IOPS or massive sequential performance?

In an ideal world you want both, as an SSD with massive
random 4K IOPS and sequential performance will always be faster than an SSD
that has high sequential performance and moderate 4K random IOPS performance,
and the same applies to an SSD that has massive 4K random performance and
moderate sequential performance. The SSD which has high performance in both
patterns will always be the faster SSD.

However, you can still have an SSD that is very fast for
desktop use that has moderate random 4K performance and massive sequential
performance, the same can be said about a drive having massive random 4K
performance and moderate sequential performance, as it is about getting the
balance right if you have to compromise on one or the other.

Test drives

  • Plextor M6e PCIe 256GB SSD
  • OCZ REVODrive 350 PCIe 480GB
  • Intel 750 PCIe NVMe 1.2GB SSD
  • Samsung 950 Pro M.2 NVMe
    256GB SSD
  • Samsung 950 Pro M.2 NVMe
    512GB SSD
  • Toshiba OCZ RD400 M.2 NVMe
    512GB SSD
  • Samsung 960 Pro M.2 NVMe 1TB


Drive preparation for running the tests

All the SSDs used in this article were in a clean and fresh
state when the testing period started. From then on, each drive had to rely on
its own NAND cleaning effectiveness for the remainder of the tests.

The Samsung 960 PRO M.2 NVMe SSD was connected to the native
M.2 socket on my test PC, and all tests were carried out the drive connected to
this socket. All SSDs used in this article had their partitions aligned to the
Windows 10 x64 defaults.

Where I use graphs in this article to display results, I
will use the following colours to make it easier, for our readers to see which drive
we are reviewing.

Samsung 960 Pro M.2 NVMe 1TB SSD Review Samsung 960 PRO M.2 NVMe 1TB SSD

Samsung 960 Pro M.2 NVMe 1TB SSD Review Comparison SSD


Now let’s head to the next page, where I look at some
basic benchmarks…