Four-Corner Performance Testing
Comparison Products
NVMe was quickly adopted in the enterprise SSD space but manufacturers have been slow to adapt the technology for client use. There are only three retail NVMe products currently shipping: the Samsung 950 Pro, Intel SSD 750 and the Zotac SONIX. The SONIX SSD is not in our comparison charts today, as its Phison E7 NVMe controller will receive a new firmware update in the coming days that increases performance.
We included the largest and smallest capacity Intel SSD 750 NVMe Add-In Card (AIC) SSDs in the tests. You can read about the mid-range 800GB drive here.
The OCZ RD400 is a direct competitor to the Samsung 950 Pro. Both drives feature the M.2 form factor and operate in desktops and high-performance notebooks. We included both existing 950 Pro capacity sizes, and Samsung relayed to media that a larger 1TB 950 Pro would emerge in early 2016. The product has yet to launch.
We also included the 480GB Kingston Predator in the tests. The Predator employs a Marvell PCIe 2.0 x4 flash controller and serves as a representative of previous generation M.2 SSDs that utilize the AHCI protocol.
To read about our storage tests in-depth, please check out How We Test HDDs And SSDs. Four-corner testing is covered on page six of our How We Test guide.
Sequential Read Performance
The OCZ RD400 aims for the consumer and workstation markets, and your workload dictates how far up the queue depth range you should focus. Regular desktop and notebook users should focus on very low queue depths, between 1 and 4, and workstation users should look slightly higher, between 2 and 8.
All three OCZ RD400 SSDs deliver very good low queue depth sequential read performance. The RD400 provides slightly more bandwidth than the Intel SSD 750 products, but less than Samsung's 950 Pro. Operating system and software limitations serve to level out performance between the products because the software has not evolved enough to take advantage of the available performance during many common workloads. We will look at real-world application performance later in the review, which will highlight how much the legacy software limits bandwidth.
Sequential Write Performance
The two largest RD400 SSDs feature a “wavy” performance profile during the test, which is what we would normally expect during thermal throttling. We tested the drives without the add-in card adapter, which adds a small heat-absorbing element. The results would lead you to believe the RD400 has a throttling issue, but that is not the case. We will cover that in more detail in the next section.
The two largest RD400 SSDs (512GB and 1TB) deliver the highest sequential write performance of any consumer SSD we have tested. The 256GB RD400 trails slightly behind its bigger brothers, but matches the sequential write performance of the 1.2TB Intel SSD 750. It also outperforms the lower capacity 750 and 950 Pro.
Sequential Thermal Throttle Test
It is important to understand how we drove the SSD into the thermal throttle condition. We automated our synthetic SSD tests so they can run back-to-back. The intense tests take more than three hours to complete, which is not a realistic timespan for a typical client workload, especially if we are attempting to identify thermal throttling during normal use. In addition, the random workloads occur prior to the sequential workload, and we did not observe a throttling condition during the random portion of the test regimen.
In the chart above, we tested the RD400 1TB in three configurations to look for signs of thermal throttling. The first test (red line) is the M.2 drive without any thermal transfer material. The second test (gray line) utilizes the OCZ add-in card with thermal transfer material between the card and the RD400. The third test (blue line) utilizes an Angelbird Wings PX1 M.2 to PCIe adapter, which features a large half-height, half-length heatsink.
The preconditioning phase of the test is just as important as the end result. We filled the drive with data, and then let it rest for a half hour to normalize the controller temperature. We ran Iometer with 100% sequential 128 KB writes for two hours after the idle period, recording performance every second.
Only the base M.2 should exhibit throttling issues during the two-hour workload, but it does not have a throttling issue. The add-in card experienced a few significant performance dips, but only for a brief time. I believe the dips stem from internal wear leveling or other background activity, and are not the result of a throttling issue. The Angelbird Wings PX1 adapter test should display the least amount of variation due to its large heatsink, which is in direct contact with the controller. However, the PX1 results show the same variation as the base M.2 test.
The tests only indicate that thermal throttling is not an issue with very large sequential write workloads. It does not mean a user cannot force a throttle condition with small-block random write data. Random writes take place in very short bursts under typical conditions. An extended random write workload is so far outside of the RD400’s design parameters (and typical workloads) that we are not concerned enough to test it.
Random Read Performance
We analyze random small-block size performance at a queue depth of 1 to separate mainstream products from high performance models, and all of the SSDs on the chart are premium SSDs. 10,000 random read IOPS is the normal crossover point between high-performance and mainstream products, and all of the SSDs in our test surpass that mark.
The OCZ RD400 products trail the Samsung 950 Pro at low queue depths, but the performance difference is very small. All three of the OCZ drives scale well up to queue depth 64. We see a slight decline at queue depth 128, but your workload will never reach that peak for it to become a performance issue.
Random Write Performance
The Intel SSD 750 has a low queue depth advantage with small-block random write workloads over the Samsung 950 Pro. The advantage stems from the power limitations of M.2 compared to a discrete add-in card form factor; PCIe cards have a higher power budget than M.2 SSDs. The M.2 power specification limit is around 8 watts, whereas the PCIe add-in card limit is 25 watts.
The OCZ RD400 SSDs fall in between the two existing NVMe products during random write measurements. I never expected to see a true M.2 SSD reach the high peak random write numbers that the RD400 1TB achieved at queue depth 4 with the current power limitations. The two small capacity RD400 SSDs also set new records for consumer M.2 performance.
High random write performance reduces system latency and increases the user experience. The OCZ RD400 provides the superior low latency feel you want, but does so with very low power consumption.