>> A further layer of distinction involves queue depth. That’s
a metric of the number of parallel storage requests or threads
in operation at any one time. In a desktop context, a typical
application will generate a single thread, requesting a read or a
write, waiting for the result, then following with another. Server
systems often support multiple applications or clients and
operate with queue depths of 32 or more. But for desktop PCs,
queue depths in the range of one to four are more realistic. IOPS
or IO operations is another measure of the same performance.4
ANVIL’S STORAGE UTILITIES
Our weapon of choice for measuring 4K random
performance is Anvil’s Storage Utilities, as it allows for
tests at the queue depths most relevant to desktop computing.
>> Download Anvil version 1.1.0, fire it up and then select
“Benchmarks” from the top menu. Choose “Threaded IO / write.”
At the top of the pop-up window is a slider, with a selection of queue
depths. Choose queue depth one, then hit “Start” below. Note the
result in MB/s, then repeat the process at QD2 and then QD 4.
>> Next, go through the same process with reads [Image C],
selecting “Benchmarks” and then “Threaded IO / read,” before
once again testing performance at QD1, QD2, and QD4. You
should find that QD1 is a worst-case scenario, with performance
increasing as you increase the queue depth. Inevitably, most
drive manufacturers quote 4K random or IOPS performance at
QD32. That’s a best-case scenario that isn’t hugely relevant to
desktop as opposed to server applications. Consequently, we
focus our testing on shorter queue depths.5
FILE TRANSFER
When it comes to emulating our SSD test routine, the
trickiest element is our 30GB internal file copy. The idea is
to provide a real-world snapshot of sustained performance andDCuncover any glaring limitations that can occur, such as
the cache limitations and chipset overheating.
>> For our testing, we use a 30GB tranche of game
files, chosen for a mix of large and small file sizes. For
obvious copyright reasons, we can’t share the precise
files we use. However, select 30GB of mixed files
from any modern game installation and you will have
something that’s reasonably similar, albeit not accurate
enough for direct comparisons of performance.
>> The test routine is straightforward. It’s conducted
internally on the drive to exclude the influence of a
secondary drive. First, copy your test files to the target
drive. Then create a new folder on that same drive and
paste a new copy of those files into the folder. Record
the time it takes to complete the copy. Again, it’s worth
keeping an eye on the Windows Explorer progress graph
[Image D] to look for any obvious drops in performance.6
COMPARING PERFORMANCE DRIVES
So, you’ve installed the software and run the
numbers. But what does it all mean? As we
implied earlier, much depends on the type of SSD
in hand. SATA-based drives don’t just have the peak
sequential limitations mentioned earlier. They’re also
compromised by the AHCI protocol, which was originally
conceived for mechanical drives with magnetic platters.
Newer M.2 drives benefit from both the PCI Express
interface for improved peak bandwidth, and the NVMe
protocol, which was expressly optimized for use with
modern solid-state storage tech. For quad-lane M.2
drives operating on the PCIe 3.0 standard, maximum
theoretical read and write speeds are in the region of
4GB/s, with real-world performance topping out at
around 3.5GB/s. For the latest PCIe 4.0 drives, you can
double that to 8GB/s as a maximum theoretical, with
early drives hitting around 5GB/s in practice. As for 4K
per formance, that’s more of a leveller, with few drives of
©^ any type scoring above 200MB/s at QD1.
MI
CR
OS
OF
T
To extract accurate benchmark numbers from any SSD,
you need the right configuration and motherboard settings.
For the most part, this isn’t complicated, and default
motherboard settings will typically be optimal. But that’s
not always the case. There are a few basics to be aware of.
What’s more, if you have a range of BIOS tweaks, you may
not wish to lose them with a global rollback to default.
First, refer to your motherboard’s documentation to
ensure that you have the drive connected to the optimal slot
or socket. With many motherboards, overall bandwidth
is shared between M.2 and SATA resources, so not all
M.2 slots or SATA ports are equal. Depending on how
you populate them, you may not be able to access a given
drive’s full performance.
With that ticked off, double-check your BIOS settings.
For a SATA drive, ensure you are running in AHCI mode. For
a quad-lane M.2 drive, make sure that PCIe x4 is enabled
where available. Just remember, on some motherboards
selecting x4 mode for an M.2 slot can force a pair of SATA
ports (typically ports five and six) to be disabled, again
because bandwidth is shared across storage interfaces.SSD SETUP
maximumpc.com SEP 2020 MAXIMUM PC 51