Kingston's DCP1000 SSD sits under the companies Enterprise banner and has been designed for extreme performance in the flash-based data centre environment. When Kingston say extreme performance, they're not joking; this is a drive capable of delivering reads at close to 7GB/s and writes up to 6GB/s!
Developed in cooperation with Liqid Inc, masters of designing data centre architecture with Kingston providing the hardware and Ligid in charge of the firmware, the DCP1000 comes in three capacities; 800GB. 1.6TB (the drive submitted for review) and the flagship 3.2TB drive.
The quoted performance figures for the drive are absolutely crazy! The 1.6TB drive has Sequential read/write figures of 6,800MB/s and 6,000MB/s respectively, the 3.2TB and 800GB drives have the same read speed with the write speed of the 3.2TB drive the same as the 1.6TB drive while the 800GB drive writes fall back to 5,000MB/s.
Random read/write figures are mind boggling. The 1.6TB drive is rated up to 1.1m, yes that's 1.1 million IOPS for reads with writes at 200,000 IOPS. The flagship 3.2TB drive is slightly slower at 1m IOPS for reads and 180,000 IOPS for writes while the 800GB drive only manages 900,000 IOPS for reads and 145,000 IOPS for writes.
At the heart of the drive are four Kingston M.2 NVMe SSD's each with, in the case of the 1.6TB drive, 400GB of capacity. Each of the drives uses a PhisonPS5007-11 eight-channel controller along with eight packages of Toshiba 15nm MLC NAND. A PLX PEX8725 24-lane PCIe Gen 3 switch sits between the drives and the PCIe 3.0 x8 board connector.
The drives can be used in a number of ways including RAID arrays and the drive is also bootable. Kingston provide endurance figures the drive as a whole and for the individual drives. Stated endurance for the 1.6TB drive as a whole is 1820TB TBW with each of the 400GB drives rated at 375TB TBW.
Kingston back the drive with a 5 year warranty.
Physical Specifications:
Usable Capacities: 1.6TB
NAND Components: 4 x Kingston 400GB PCIe Gen 3 NVMe SSDs
NAND Controller: 4 x PhisonPS5007-1, PLX Technology PEX8725 PCIe 3.0 switch
Interface: PCIe 3.0 x8
Form Factor: HHHL (half-height, half-length)
Dimensions: 168 x 69 x 18 mm
Drive Weight: 209g
Firmware Version: E7FTD4.7
As it's a drive aimed at the Enterprise market segment, it comes as no surprise to find that the DCP1000 ships in a pretty plain brown box with just Kingston's branding on it, the drive name in the bottom right with a NVMe logo on the other side.
All that's in the box is a standard length PCI bracket (we swapped it out for the short one that comes attached to the drive, which is the one you can see on the box) and a tiny installation guide.
The DCP-1000 is built on an HHHL (half-height, half-length) format PCB measuring 168 x 69 x 18 mm. Practically all of the PCB is covered by a passive heatsink, the only part of the board not covered is where the nine power capacitors sit. These provide hardware data loss protection against power failure.
We haven't taken the drive apart to photograph the internals as we hope to do some further testing with it and didn't want to risk upsetting the thermal properties of the drive.
Looking after Kingston's SSDs is their SSD Manager utility.
Compared to some of its rival's it's a very simple affair. It supports firmware upgrades for the drive and the S.M.A.R.T section has a useful temperature readout.
For testing, the drives are all wiped and reset to factory settings by HDDerase V4. We try to use free or easily available programs and some real world testing so you can compare our findings against your own system.
This is a good way to measure potential upgrade benefits.
Main system:
Intel Core i7-7700K with 16GB of DDR4-3200 RAM, Sapphire R9 390 Nitro and an Asus Prime Z270-A motherboard.
Software:
Atto Disk Benchmark.
CrystalMark 3.0.3.
AS SSD.
IOMeter
All our results were achieved by running each test five times with every configuration this ensures that any glitches are removed from the results. Trim is confirmed as running by typing fsutil behavior query disabledeletenotify into the command line. A response of disabledeletenotify =0 confirms TRIM is active.
Kingston's DCP1000 drive can be set up in RAID 0 or RAID 1. We built each array and then tested the drive.
The ATTO Disk Benchmark performance measurement tool is compatible with Microsoft Windows. Measure your storage systems performance with various transfer sizes and test lengths for reads and writes. Several options are available to customize your performance measurement including queue depth, overlapped I/O and even a comparison mode with the option to run continuously.
Use ATTO Disk Benchmark to test any manufacturers RAID controllers, storage controllers, host adapters, hard drives and SSD drives and notice that ATTO products will consistently provide the highest level of performance to your storage.
Kingston quote Sequential Read/Write figures for the 1.6TB version of the DCP1000 at 6,800MB/s and 6,000MB/s respectively. Tested under the ATTO benchmark, the review drive even bettered those with reads at 7,011MB/s and writes at 6,511MB/s in RAID 0. When the DCP1000 is built as a RAID 1 array, these figures drop to 3,056MB/s for reads and 2,314MB/s for writes, still hugely impressive.
To give an idea of just how fast this drive is we've added scores from the Kingston KC1000 drive, which is another PCIe HHHL format card but with a single M.2 SSD attached to it and the Samsung SSD960 PRO, one of the fastest consumer M.2 drives currently available.
In our sustained Sequential reads versus block size test, once again the drive produced figures exceeding the official ones in RAID 0, ending the test at 7,284.14MB/s. In RAID 1 the drive peak drops to 5,871.62MB/s, still a very impressive figure.
When it comes to writes, in RAID 0 the drive peaks at 6,202.29MB/s at the 64K block size, again faster than the official figure before falling back to 5,027.52MB/s at the end of the test. In RAID 1 the writes drop considerably ending the test at a peak of 1,718.32MB/s. It may be a lot slower than RAID 0 but the performance is more stable throughout the test.
We tested the drive using an 8KB data block and 8 threads which is more typical of an enterprise enviroment. We used a 16GB span. In a RAID 0 array the drive's read performance peaked at a stunning 729,361 IOPS, 5,974.95MB/s and the write performance was also extremely good at 421,412 IOPS, 3452.1MB/s.
In RAID 1 it's the writes that have the advantage at the very beginning of the test, but its short lived as the write performance plateau's out all the way through the test from a start position of 841.87MB/s finishing at 856.26MB/s. Reads on the other hand start at a lower 546.63MB/s but climbed steadily to finish at 1,791.37MB/s.
Testing the drive with a 8KB data block and a 70/30 read/write split, again another test with the enterprise environment in mind saw the drive produce 340,150 IOPS in RAID 0 while in RAID 1 the figure was 94,248 IOPS, again extremely strong figures.
For the long term performance stability test, we set the drive up to run a 20-minute 4K random test with a 30% write, 70% read split, at a Queue Depth of 256 over the entire disk.
In RAID 0, the DCP1000 averaged 207,453 IOPS for the test with a performance stability of 97.08% which you would want and expect from a data center drive.
When the drive is in RAID 1 mode, it averaged 107,252 IOPS for the test with a performance stability of 71.3%
We also tested the drive in a number of workload scenarios that it might face in real life. The IOMeter settings we used to test with are listed below:-
Database
Transfer Size: 8K
Reads 67% Writes 33%
Random 100%
Boundry 4K Outstanding IO 64 Threads/Workers: 8
Exchange Server 2007
Transfer Size: 8K
Reads 67% Writes 33%
Random 100%
Boundry 8K Outstanding IO 60 Threads/Workers: 1
SQL Server 2008 OLTP
Transfer Size: 8K
Reads 70% Writes 30%
Random 100%
Boundry 8K Outstanding IO 60 Threads/Workers: 1
Web File Server 64KB
Transfer Size: 64K
Reads 95% Writes 5%
Random 75%
Boundry 4K Outstanding IO 64 Threads/Workers: 8
Web File Server 8KB
Transfer Size: 8K
Reads 95% Writes 5%
Random 75%
Boundry 4K Outstanding IO 64 Threads/Workers: 8
Web File Server 4KB
Transfer Size: 8K
Reads 95% Writes 5%
Random 75%
Boundry 4K Outstanding IO 64 Threads/Workers: 8
Media Streaming
Transfer Size: 64K
Reads 98% Writes 2%
Sequential 100%
Boundry 4K Outstanding IO 64 Threads/Workers: 8
Video On Demand
Transfer Size: 128K
Reads 100% Writes 0%
Random 100%
Boundry 4K Outstanding IO 512 Threads/Workers: 8
Decision Support DB
Transfer Size: 1MB
Reads 100% Writes 0%
Random 100%
Boundry 4K Outstanding IO 64 Threads/Workers: 8
Mail Server
Transfer Size: 32K
Reads 58% Writes 42%
Random 95%
Boundry 4k Outstanding IO 64 Threads/Workers:8
Search Engine
Transfer Size:4K
Reads 100% Writes 0%
Random100%
Boundry 4K Outstanding IO 64 Threads/Workers:8
OS Paging
Transfer Size:64K
Reads 90% Writes 10%
Sequential 100%
Boundry 4K Outstanding IO 64 Threads/Workers:1
Digital Video Surveillance
Transfer Size:512K
Reads 90% Writes 10%
Sequential 100%
Boundry 512K Outstanding IO 64 Threads/Workers:8
Workstation
Transfer Size: 8K
Reads 80% Writes 20%
Random80%
Boundry 8K Outstanding IO 64 Threads/Workers 4
The DCP1000 provided some stunning numbers in these workload scenarios. The Video On Demand and Decision Support Database tests produced MB/s figures that surpassed the official top end figure of 6,800MB/s. The Video On Demand test produced a figure of 7,163MB/s while the Decision Support Database gave up a score of 7,143MB/s
PCMark 8’s standard Storage test saves a large amount of performance data. The default test runs through the test suite of 10 applications three times. Here we show the total bandwidth performance for each of the individual test suites for the third and final benchmark run.
The drive provides very strong performance throughout the tests in both RAID 0 and RAID 1 arrays. Interestingly in both the PowerPoint and World of Warcraft tests the drive was actually faster when in a RAID 1 array.
To test real life performance of a drive we use a mix of folder/file types and by using the FastCopy utility (which gives a time as well as MB/s result) we record the performance of drive reading from & writing to a 256GB Samsung SSD850 PRO.
100GB data file.
60GB iso image.
60GB Steam folder – 29,521 files.
50GB File folder – 28,523 files.
12GB Movie folder – 24 files (mix of Blu-ray and 4K files).
10GB Photo folder – 621 files (mix of .png, raw and .jpeg images).
10GB Audio folder – 1,483 files (mix of mp3 and .flac files).
5GB (1.5bn pixel) photo.
The DCP1000 is much more adept at handling larger files than the small bity files of everyday office use in both RAID 0 and RAID 1.
Just to see how fast the drive could transfer files we tested transfers from folder to folder on the drive. Once the SATA interface of our usual Samsung transfer drive was out of the way the drive really showed its muscle in RAID 0 but once again only on large file sizes. Our 100GB data file was moved between folders at 2,572MB/s or give to a better idea it took just 39.8 seconds to do it. The 5GB photo took 1.7 seconds and the BluRay movie a mere 17.5 seconds.
The combination of Kingston and Liqid Inc was bound to produce something quite special and with the DCP1000 they have produced as, Kingston state, the fastest HHHL (Half-Height Half Length ) NVMe SSD seen to date. Aimed at the flash based data center, the drive has at its heart four NVMe SSD drives and it uses them to stunning effect. It goes without saying this is the fastest drive we have tested to date and then some. It is available in three capacities; 800GB, 1.6TB and 3.2TB.
The 1.6TB drive Kingston kindly supplied for review uses four 400GB M.2 NVMe SSDs each using a PhisonPS5007-11 8-channel controller and Toshiba 15nm MLC NAND. The drives are connected to a PLX PEX8725 24-lane PCIe Gen 3.0 switch which in turn connects to the PCB's PCIe 3.0 x8 interface. The drive is bootable and plug and play and is identified by the OS as four logical drives which can be used in a number of ways including as RAID arrays.
To say the performance is stunning is a major understatement. The official Sequential numbers for the 1.6TB drive are 6,800MB/s for reads and 6,000MB/s for writes. We even managed to better them with the ATTO benchmark, the review drive giving up a read figure of 7,011MB/s with writes at 6,511MB/s when the drives were built into a RAID 0 array.
4K Random read figures are even more impressive, with reads up to 1.1m, yep 1.1 million IOPS and writes at 200,000. With our 8K enterprise tests we couldn't reach those dizzy heights for reads, the best being 729,361 IOPS, still pretty impressive. However for writes we got a maximum of 421.412 IOPS.
The drive supports a broad spectrum of operating systems; Windows 8.1, Windows 10, Windows Server 2012 R2, Windows 7 and Windows Server 2008 R2 (via updates or Hotfix driver download), Linux Kernel 3.3 and higher, FreeBSD 10.x/11, VMware vSphere 6.0 (vSphere 5.5 as download driver).
We found the 1.6TB version of the DCP1000 available to pre-order on Insight for £1461 inc vat HERE
Discuss on our Facebook page, over HERE.
Pros.
- Extreme performance.
- Flexible drive setups.
Cons
- Didn't do so well in some of our tests with smaller files.
Kitguru says: Outside its natural environment the DCP1000 is a very niche drive but if you have the need to shift large sized files very, VERY quickly then it's all the drive you need.
Be sure to check out our sponsors store EKWB here