Aplicata Quad M.2 NVMe SSD PCIe x8 Adapter Review

Aplicata Quad M.2 Performance Testing

Comparison Products

Loading...

We have a few similar products in the lab, but some come with difficult compatibility requirements and only work in a small number of branded systems. Others are only certified with specific NVMe SSDs. For our first NVMe HBA review, we're using the Aplicata Quad x8 with three sets of SSDs. We tested the drive with Samsung 960 Pro 1TB SSDs, older 950 Pro 512GB SSDs, and finally, a set of low-cost 240GB BPX drives from MyDigitalSSD.

We put the three Aplicata configurations up against a single Samsung 960 Pro 2TB NVMe SSD and an enterprise drive from Kingston.

Kingston developed the DCP1000 in conjunction with Liqid Technology and Phison. The drive uses four Phison PS5007-E7 SSDs with a healthy dose of overprovisioning (512GB raw/400 GB usable) to ensure strong application performance during heavy workloads. Kingston told us last year that the DCP1000 is an enterprise SSD that you can also use for demanding workstation applications.

The Test System

We're introducing a new test system that will pop up in our future reviews. Intel sent us a new Core i9-7900X 10-core processor and Crucial loaded us up with several kits of high-speed DDR4 to fill the eight memory slots. The Asrock X299 Taichi motherboard is the star of the show and supports Intel's new vROC feature. vROC allows the processor to act as a RAID controller, similar to Rapid Storage Technology (RST), but supports up to twenty drives in a single array. All we need is a compatible vROC dongle. Without a dongle, only Intel-branded SSDs work with the vROC RAID feature.

Testing Configuration

With our vROC configuration in limbo, we're forced to test the Aplicata Quad M.2 NVMe SSD PCIe x8 Adapter with Windows software RAID. We used Windows 10 to build a four-drive RAID 0 array. The array is not bootable, which vROC will eventually resolve. You don't have to use all four drives in the same array. You can designate one drive as a boot volume, and then use any combination of the remaining three drives in a secondary storage array.

Windows RAID has a glaring hole; it doesn't pass the TRIM command. We suggest increasing the amount of overprovisioning to give the drives more spare area to work through background processes.

Sequential Read Performance

To read about our storage tests in-depth, please check out How We Test HDDs And SSDs. We cover four-corner testing on page six of our How We Test guide.

It isn't surprising to see the two Samsung NVMe arrays provide explosive performance at low queue depths. Samsung's low-QD performance is at least a generation or two ahead of the other manufacturers. The two Samsung arrays deliver a significant performance increase over a single 960 Pro.

The more modest array with four MyDigitalSSD BPX 240GB NVMe SSDs provides a nice increase over a single drive. The BPX drives in the Aplicata adapter match Kingston's sequential read performance.

Sequential Write Performance

Sequential write performance is off the charts! We actually had to go in and fatten up the chart's axis to show the RAID 0 arrays.

10-gigabit Ethernet delivers just over 1000 MB/s of throughput. A single Samsung 960 Pro can take in your data from across the network, but the Aplicata is fast enough to soak up 56Gb/s Infiniband traffic with commodity SSDs.

The sequential write test shatters our hopes of matching the Kingston DCP1000 with low-cost consumer SSDs. We are using smaller and slightly slower E7 SSDs, but the DCP1000 provides a big boost over our DIY version.

Random Read Performance

We saw a modest improvement with the Aplicata arrays at low queue depths during write workloads, but that didn't carry over to the random read tests. The arrays provide similar performance up to QD8, but we see some separation beyond that point. The arrays follow the same trajectory that we would see with single devices, just amplified.

Random Write Performance

The random write results show lower performance than expected. This may be due to the Windows software RAID overhead. This is one area we expect Intel's vROC will increase performance. Surprisingly, the 960 Pro array results are lower than a single 960 Pro.

80% Mixed Sequential Workload

We describe our mixed workload testing in detail here and describe our steady state tests here.

There is a massive performance delta between the single 960 Pro and the four-drive 960 Pro RAID 0 array. The array doubles throughput performance at QD2, and by QD32 it offers triple the performance. Tasks like video editing will see a big performance increase if your processor is strong enough to take advantage of the additional throughput.

80% Mixed Random Workload

The Windows software RAID hurts random performance. Don't get me wrong, the array results are still impressive even by today's NVMe SSD standards, but without vROC, these products are best for sequential workloads.

Random Steady-State

Random write performance looks a bit different when the drives are in a steady state. The 960 Pro SSD array outperforms the single drive by a large margin. When Samsung released the 960 series, it was very difficult to find drives in stock. We heard rumors that data centers were buying them in large numbers because they scaled so well and provided an excellent cost to performance ratio. This test shows why.

The Phison E7-based Kingston DCP1000 provides better performance than the Aplicata E7 array due to its extra spare area.

PCMark 8 Real-World Software Performance

For details on our real-world software performance testing, please click here.

Our desktop application tests show that performance is very similar with this class of products. The arrays don't always increase performance over a single large NVMe SSD, as shown by comparing the two 960 Pro configurations.

Application Storage Bandwidth

A Samsung 950 Pro array is actually faster than a single 960 Pro in this test. The same is true for the software RAID 0 array, but the results are dampened somewhat with the array. The single 960 Pro is still faster in these tests than the array due to the low-QD random accesses in these tests.

The two Photoshop tests use the most sequential data, but the performance is much closer than we thought it would be. A processor with higher single threaded performance, like a Core i7-7700K, may show a larger difference due to its higher clock speed. The Intel Core i9-7900X wields more cores than a Core i7-7700K, but its clock speed is much lower.

MORE: Best SSDs

MORE: How We Test HDDs And SSDs

MORE: All SSD Content

Create a new thread in the UK Article comments forum about this subject
This thread is closed for comments
1 comment
Comment from the forums
    Your comment
  • duncan_from_enfield
    The PCIe x16 adapter seems a good fit for Threadripper or EPYC systems