Performance Results and Potential Issues

Given the higher performance of these newer Thunderbolt 3 capable parts, it seemed fitting to apply a more demanding workload, so I pointed our PCPer Test Suite at the round of test samples. This suite is actually well suited for external use, as the sequential test was tailored to replicate typical file copies, and the random and mixed performance metrics help demonstrate application loading and heavier usage with background writes taking place. I've left an intentional break in the results near the bottom. Results above the break are the external drives tested on an ASUS X370 Prime with the Thunderbolt EX3 adapter card installed. Results below the break are NVMe and SATA parts directly connected to our standard X99 storage testbed, included to give a matter of perspective vs. natively connected storage devices. I'm skipping the usual test suite introduction for this abbreviated review, but details of the test suite are available here.

Sequential

sequential.png

Regarding sequential read performance, the X5 is the king of the hill, coming in 1.5x faster than the next fastest TB3 product, nearly doubling the USB 3.1 Gen2 products, and quadrupling the USB 3.1 Gen1 Samsung T5. Writes are also impressive, just slightly less so.

Comparing with a native 970 EVO, a few hundred MB/s are lost along the way. Still not bad at all for an external storage device.

Random

random.png

Random performance is indeed impressive, especially random reads, which some within a few percent of a natively-connected 970 EVO.

Mixed Burst

mixed.png

Read throughput is the pain point for this particular test, as the write rate is throttled to 20 MB/s. The X5 does great here, showing that it is even fast enough to run applications, VM's, and other heavier loads directly off of the external SSD with performance similar to an internally mounted device. Read service times paint the rest of this picture:

mixed-2.png

Overall an impressive showing from the Samsung X5.

Caching

cache-1.png

This cache test is meant for desktop use, but we hit the X5 with it in order to help evaluate possible thermal throttling during heavier duty cycles. The test issues 60 second bursts of continuous writes. Idle times are progressively shortened (+600s, +300s, +60s, etc) during the first 6 passes, then increased for the last two passes in order to evaluate proper cache recovery.

The X5 did well in this test. When sufficient idle time was present, available cache remained at 40GB (the test runs at half-filled capacity, and Samsung's caching routine allows for dynamically greater available cache so long as the SSD is not completely filled). During the heavier cycles with shorter idle periods, speeds remained between 1.0 and 1.2GB/s. As soon as the X5 was given a bit of breathing room, the cache immediately snapped back to the full available 40GB of 1.7GB/s write throughput.

Potential Issues

A few things to consider before moving on. First:

usb-attempt.png

Above is what happens when connecting the Portable SSD X5 into a standard (non-Thunderbolt) USB 3 port. Part of the advantage of having a portable SSD is the ability to use it to move data between multiple systems. While the SSD T1/T3/T5 were able to fall back to USB 2.0 and even 1.1 if needed for compatibility, the X5 is bound by the limitations of current Thunderbolt 3 controller hardware, which is unable to fall back to the USB protocol. This may change in the future, as more advanced controllers are released (hopefully with the ability to translate from USB UASP to NVMe), but we're just not there yet.

Thermal Throttling

The X5 has what is basically a hunk of aluminum acting as a thermal buffer. Short bursts of heavy activity are easily absorbed and prevent thermal throttling, but sustained continuous usage at the maximum throughputs possible will eventually slow things down a bit. In addition to the caching test above, I was able to write continuous and at maximum speed for greater than 5 minutes straight before any throttling was noted. Due to the insulative nature of the solid (not vented) plastic housing, the exact amount of time before throttling varies greatly by previous activity, room temperature, and even the type of surface the X5 is sitting on (wood desks are more thermally insulative than metal ones).

« PreviousNext »