Intel DC P3700 800GB NVMe SSD Review

[AkG]

Nearly a year and a half ago Intel introduced an entirely new way of thinking about Enterprise grade Solid State Drives: their Data Center series. During that first generation outing they focused in on steady-state performance and SATA based solutions, thus the 'S' in the DC S3700 and DC S3500 monikers. Put simply the DC S3xxx series introduced to the world the third generation Intel SSD controller, one which has recently cascaded down to the consumer grade 730 series. Now Intel has turned their attention to a more pressing issue: the underlying architecture that makes up a Solid State Drive. To showcase the fruits of their labor they have launched the DC P3700 series; and today we will be looking at the 800GB version.

The DC P3700 series is meant to highlight a number of advances within Intel’s newly revised SSD environment but it still targets the same data center-based market as the DC S-series. However, instead of utilizing a SATA interface, it uses the PCI-E bus to maximize performance. Therefore, the P3700, P3600 and P3500 aren’t meant as replacements to the slightly older but no less capable S3500 and S3700.

Like the Intel 910 series it obviously replaces, this new Data Center P3700 series will come in various capacities, but each and every one will have the same price per gigabyte ratio. While $3.02 per gigabyte does sound high (especially compared the DC S3700 series) this is a lot less than the 910 commanded when it was launched. For example a 800GB Intel 910 had an MSRP of $3,859 or $4.82 per GB whereas a 800GB DC P3700 will only set companies back $2,414. This is still high, but for the enterprise market it is much more palatable and should help the DC P3700 gain traction. However, lowered price is the smallest of the benefits the DC P3700 brings to the table.

In the past when dealing with either SATA or PCIE form factors the underlying foundation was the same: SATA and AHCI. For example, the 910 series may have used a PCI-E form factor but this was accomplished by the simple expedient of using a PCIe hub controller that was placed between the bus and the multiple SATA controllers that took care of the actual NAND. This was only a short term solution but suffice to say doing the hodge-podge of conflicting standards may have boosted performance beyond what SATA drives could handle but it also created nearly as many issues as it solved.

To eliminate the legacy issues that SATA carries from its hard drive roots, Intel helped found the Non Volatile Memory Express Workgroup in 2009. The NVM-EW's sole focus was to create a new Host Interface standard created by SSD manufactures specifically for solid state drives. This is why the heart and soul of Intel's latest DC P3700 is the Non-Volatile Memory Host Configuration Interface (also known as NVMe) and not Serial ATA's 'Advanced' Host Configuration Interface.

As the name suggests the DC P3700 uses a PCIe form-factor, but unlike the 910 the DC P3700 does not require a PCIe hub controller to connect to the PCIe bus. This in turn allows the DC P3700 to lower latency levels by 325% to an amazing 20 microseconds. Due to its NVMe foundation the DC P3700 also boasts an equally impressive 460K/180K IOP/s rating that is well in excess of what any classical PCIe SSD could do.

Mix in a maximum 25 Watts rating with real world average more in the 15 watt range and 10 full Drive Writes Per Day for five years (14.6 PetaBytes of writes in the 800GB's case), and on paper the DC P3700 is certainly going to do exactly what Intel wants it to do: get corporations’ attention and show what this new standard can offer end users.

 

[AkG]

Closer look at the DC P3700

Closer look at the DC P3700

​

Before we get to the drive itself, the DC P3700 800GB’s naming scheme requires a bit of explanation. The new name for enterprise grade devices begins with stating exactly what they are: DC or in this case, Data Center-focused. The third letter denotes form-factor. For example the DC S3 series is a SATA based series, and the new DC P3 series uses the PCIe standard. At some point (once 8639 SFF gains traction) we fully expect a DC N series to make its appearance that uses a truly native NVMe interface. The fourth is always a digit and denotes the generation or in this case '3' is for 3rd generation. The last three digits tell you if it is a 3 (entry or for nearly entirely read scenarios), 5 (mainstream or for mostly read-centric scenarios), 7 (high performance or for read/write scenarios), or 9 (ultra-high performance) device. So with this taken care of you can see that the DC P3700 800GB is an Enterprise 'Data Center' drive that makes use of the PCIe form factor, and is high performance model meant for demanding scenarios where the I/O requests could be either read or write.

To further help customers know where a given model will stack up against others of its generation, the color coding on the heatsink (or chassis) will tell you if it’s a high, moderate, or an entry level model. In the DC P3 series instance, the label on the DC P3700's heatsink is black with a blue pin-stripe. P3600's will have a gray with blue pin-striping label, and the entry model will be white. Since there are no new 9 series DC drives for the moment we can only assume it will be black with silver pin-striping.

​

In the past to hit the maximum capacity the older Intel 910 800GB required a main PCB, and two daughter cards whereas the new DC P3700 simply needs a single half-height, half-length PCB. Of course, just like the Intel 910 800GB, this moderate sized 800GB DC P3700 is literally covered with NAND ICs on both sides of the PCB and uses a large heatsink to keep the controller cool. This passive only heatsink requires active airflow of 300LPM.

​

While details are few and far between for the Intel CH29AE41AB0 chip, this native NVMe controller is recognized by the OS as a 'Xeon E3-1200 v3/v4' series management device. Regardless of this also being the reference designation for one of Intel’s Xeon CPUs, the controller is able to communicate directly with the system’s PCIe bus which helps lower overall latency.

The single controller design can handle much, much more NAND modules than any SATA/AHCI controller we know of. To be precise, the 800GB model makes use of 36 20nm HET NAND ICs. Considering there isn’t any more room on either side of the PCB for more NAND, we assume that higher capacity versions simply use higher density chips. The P3700 also happens to be the first Intel branded drive to make use of "High Endurance Technology" (HET) - or what everyone else calls e-MLC - NAND.

High Endurance Technology MLC NAND is approximately thirty times more durable than standard MLC NAND but not nearly as costly to manufacture as SLC-type modules. This in turn allows Intel to offer such massive drives for a relatively reasonable price.

Intel has actually been very conservative in switching over their enterprise grade drives to a new NAND process, so while standard 20nm MLC IMFT NAND has been available for quite some time now, 20nm HET IMFT NAND has only recently passed all of Intel’s various certification processes. So even though these use a relatively new fabrication process, these 20nm HET NAND ICs have the same 10 full drive writes per day for five years guarantee that the previous 910's 25nm HET had. To put that in more practical terms, Intel guarantees the drive’s NAND for over 14 petabytes of writes for the 800GB model and 36.5 petabytes for the massive 2TB model.

​

While the P3700 has been virtually covered in 20nm NAND modules, only half are directly in contact with heatsinks and the eighteen back-facing modules are bare. NAND may not need active cooling but these 18 ICs are more susceptible to ESD and even physical damage than the others that are protected by the heatsink.

In a very interesting turn, the DC P3700 is not recognized as four separate LUNs like the Intel 910, but rather is seen as one large single LUN. This means you don’t have to apply a 'software' RAID overlay to this drive to fully utilize it which in turn also lowers overall latency and optimizes CPU overhead.

​

Speaking of overhead, on POST the P3700 has almost no initialization wait time, can be booted from, and the Intel stock NVMe driver is simple to install. Simply power down the server, plug it into a free x4 or x8 PCIe 3.0 slot, power up the server, wait for it to be recognized and use the simple driver installation wizard. After this, you can go into device manager and format your new '800GB' of space just as if it was a SATA, SAS or any other typical drive. For system admins, this easy of setup will likely prove to be invaluable.

​

While the last generation Intel 910 series required a PCIe 2.0 x8 slot, the new DC P3700 only requires a PCIe 3.0 x4 slot and is not backwards compatible with the 2.0 standard. This halving of lanes not only allows servers to support more drives before running out of slots, but also makes finding a free slot on any supporting motherboard a lot easier. Obviously Intel feels that PCIe 3.0 has high enough market saturation for this step to be taken.

The Intel 910 series required only 2 RAM modules per controller whereas the DC P3700 has five Micron branded, DDR3-1600 SDRAM ICs. While this is technically a downgrade we doubt many will be unimpressed with the DC-P3700's 1.25GBworth of onboard cache. This reduction in number of module allotment also helps reduce overall power consumption, allowing the single controller to consume even more power while still staying within the DC P3700's low TDP envelope. Because there are fewer components to power, the number of capacitors has also been reduced over that of the Intel 910 but there’s still ample backup power to provide full Flush In Flight capabilities in the event of unexpected power loss.

​

The DC P3700 may use an entirely new controller and controller interface standard but the stock NVMe driver does support passing the TRIM command on to the drive. It may not sound like much but this is a major addition for long term drive maintenance.

 

[AkG]

NVMe & What it Really Means to You

NVMe & What it Really Means to You

When people first hear the acronym NVMe, many might assume that it stands for 'Non-Volatile Memory, Enterprise' and it is some new type of enterprise grade NAND. While a good and logical guess, it is not even close to what NVMe is and what it is meant to do. Non-Volatile Memory Express is actually an entirely new standard for SSDs and how they communicate with a system’s PCIe bus.

As most enthusiasts and professionals know, SATA based devices have a whole host of legacy issues that intrinsically hold Solid State Drives back. Put simply, the SATA standard was never designed with extremely high performance NAND based devices in mind. Instead it was designed with Hard Disk Drives in mind. While the governing body of SATA's standard certainly has done their best to improve it via SATA 6Gbps and the new SATA Express, at the end of the day it still relies upon the outmoded 'Advanced' Host Controller Interface that requires an intermediary controller between the CPU and the storage device.

By requiring an I/O Controller Hub (or PCH), AHCI - like IDE before it - allows the production of associated drives to be less expensive as the drive's controller doesn’t need to do the heavy lifting. Rather, the tertiary processing gets offloaded to the PCH / CPU combination. This increases latency and creates a performance bottleneck when dealing with ultra-high performance devices which require a massive amount of data to be processed off-device.

In this past this didn’t cause too much of an issue since spindle-based drives didn’t need to have extremely high performance controllers. However, solid state devices have seen their capabilities increase exponentially in a relatively short amount of time, vastly outstripping what even a 6Gbps SATA interface can provide. Even in a perfect world where SATA-IO could quickly implement upgraded standards there would still be a large latency bottleneck due to the fact that the SATA controller would still have to receive the commands from the CPU, pass them on to the SSD controller, receive the information and then retransmit.

Without scalability or a quick means to keep pace with SSD technology, many have concluded that SATA is a dead end with an outmoded standards process. Instead something different is needed, something built from the ground up with next generation storage performance as its primary focus.

While waiting for a new standard to emerge, Solid State Device manufactures turned to the PCIe bus to circumvent the SATA or SAS controller. Unfortunately, this in turn required a PCIe HUB controller and special proprietary drivers which allow a PCIe SSD to connect to the PCIe bus and effectively communicate with the system. For the most part these new overheads simply reduced instead of eliminated the underlying problems with existing SSD communication designs.

Obviously these issue have been known for some time, and back in 2007 (a mere four years after SATA was implemented) Intel helped create a new open source standard called the Non-Volatile Memory Host Controller Interface Specification (NVMHCI for short). After two years of work a consortium of over ninety companies founded the NVM Express Workgroup which would be in charge of developing NVMHCI into a workable, open source connectivity and communication standard. It is out of this workgroup that the standard which we now know as Non-Volatile Memory Express (NVMe) was created.

As previously stated NVMe has been designed from the ground up with the unique abilities and demands of Solid State Drives in mind. As such overall latency, available bandwidth, and scalability are the most important areas NVMe seeks to address. To minimize these issues, the NVM-EW opted to use PCIe as its foundation. However, instead of just making a hodge-podge standard that relies upon PCIe host bus adapters to work, NVMe compatible controllers will be able to talk directly to the CPU as they have to 'speak' PCIe.

By removing this middleman controller a lot of the latency issues associated with PCIe based Solid State Drives are also removed. Equally important, this also eliminates the need for custom drivers and their associated overhead, which will also further reduce latency as there will be fewer layers between the SSD controller and CPU. In the case of the Intel DC P3700 800GB, its NVMe design allows it to boast an impressively low 20 microsecond read and write latency.

As an added bonus NVMe based devices will require fewer controller chips on the device, which reduces power consumption and cooling requirements. This is why the Intel PDC P3700 consumers only 20-25 watts of power compared to the last generation, lower performing Intel 910 which required 25 to 28 watts of power.

Obviously NVMe solves the latency issue which was actually starting to bottleneck PCIe based drives but it also solves future performance issues as well. By using the PCIe bus the NVMe Workgroup is able to be meet emerging needs faster than SATA's or SAS' workgroups could ever hope to, as most of the hard work is done for them. For example PCIe 2.0 NVMe devices have access to a 2GB/s wide bus, whereas PCIe 3.0 NVMe devices will be able to hit nearly 4GB/s before saturating the bus, and future versions (PCIe 4.0 compatibility has already been announced) will have nearly 8GB/s of bandwidth to work with, and so on and so forth. Needless to say both performance and future proofing have been neatly taken care of as they are built directly into NVMe standards.

As an additional benefit from having no legacy issues to support, the number of channels and even number of NAND ICs per channel can be scaled up beyond what AHCI based devices can reasonably support. This allows increased capacity drives which not only reduces the cost per gigabyte of each NVMe device, but also allows for fewer NVMe drives to meet given capacity and performance requirements of a build. That’s something that will be of utmost importance to enterprise consumers. For example, the performance offered by one NVMe DC P3700 can replace up to eight Intel DC S3700 drives, while also offering increased steady state performance and decreased latency.

While NVMe is for the time being an Enterprise-only affair this should quickly change and workstation or even mass market NVMe based drives will likely start appearing in the near future. On the surface, the idea of NVMe may not appeal to consumer motherboard manufactures used to offering only 'SATA' compatibility, and the idea of finding room for another port standard certainly does not appeal to many engineers. Luckily NVMe has another ace up its sleeve: Small Form Factor 8639 specification and SATA Express.

​

In a bout of inter-bureaucratic cooperation rarely seen, SATA Express has been designed from the ground up to use either 'legacy' AHCI or NVMHCI as its standard. Of course, there will be a certain performance loss by using NVMe instead of AHCI ( at heart it is still a SATA-IO and not NVM-EW created standard) but this will allow an intermediary step between AHCI compliant solid state drives and NVMHCI complaint devices. Thanks to Intel pushing SATA Express (via their Z97 PCH controller) this ensures compatible ports will become standard on most Intel based consumer grade motherboards. On its own this would certainly help ease the inevitable transition from AHCI to the superior NVMe standard but the truly ingenious part is the new 8639 Small Form Factor specification.

8639 SFF is an emerging standard which takes the usual SATA / SAS / SATA Express power and data ports and converts it to also support NVMe based devices. To ensure there is no confusion, NVMe devices using SFF 8639 interface will have a slightly different pin out configuration. Much like SATA connectors will not work on SAS drives, NVMe connectors will not (or should not) work with the other types, thus eliminating the 'it fits, but doesn’t work' risk that could have otherwise cropped up.

This boon to workstation and home users is more a side effect as the SFF 8639 standard is meant to quickly allow servers to be upgraded to NVMe with just a backplane swap. For whatever the reason the end result is that NVMe will not just be replacing 'PCIe' Solid State Drives, but will also replace AHCI based 'SATA', and 'SAS' SSDs as well. Taken as whole NVMe is easily the biggest advancement in storage subsystem development since the creation of solid state drives and compatible devices are about to start pouring into the market.

 

[AkG]

Test System & Testing Methodology

Testing Methodology

Testing a drive is not as simple as putting together a bunch of files, dragging them onto folder on the drive in Windows and using a stopwatch to time how long the transfer takes. Rather, there are factors such as read / write speed and data burst speed to take into account. There is also the SATA controller on your motherboard and how well it works with SSDs & HDDs to think about as well. For best results you really need a dedicated hardware RAID controller w/ dedicated RAM for drives to shine. Unfortunately, most people do not have the time, inclination or monetary funds to do this. For this reason our testbed will be a more standard motherboard with no mods or high end gear added to it. This is to help replicate what you the end user’s experience will be like.

Even when the hardware issues are taken care of the software itself will have a negative or positive impact on the results. As with the hardware end of things, to obtain the absolute best results you do need to tweak your OS setup; however, just like with the hardware solution most people are not going to do this. For this reason our standard OS setup is used. However, except for the Vista load test times we have done our best to eliminate this issue by having the drive tested as a secondary drive. With the main drive being a Seagate 600 Pro 400GB Solid State Drive.

For synthetic tests we used a combination of ATTO Disk Benchmark, HDTach, HD Tune, Crystal Disk Benchmark, IOMeter, AS-SSD and PCMark Vanatage.

For real world benchmarks we timed how long a single 10GB rar file took to copy to and then from the devices. We also used 10gb of small files (from 100kb to 200MB) with a total 12,000 files in 400 subfolders.


For all testing a Asus PZ97 Deluxe motherboard was used, running Windows 7 64bit Ultimate edition. All drives were tested using AHCI mode using Intel RST 10 drivers unless the drive is a NVMe drive in which case the latest Intel IaNVMe drivers were used.

All tests were run 4 times and average results are represented.

In between each test suite runs (with the exception being IOMeter which was done after every run) the drives are cleaned with either HDDerase, SaniErase, OCZ SSDToolbox or Intel Toolbox and then quick formatted to make sure that they were in optimum condition for the next test suite.


Steady-State Testing

While optimum condition performance is important, knowing exactly how a given device will perform after days, weeks and even months of usage is actually more important for most consumers. For home user and workstation consumers our Non-Trim performance test is more than good enough. Sadly it is not up to par for Enterprise Solid State Storage devices and these most demanding of consumers.

Enterprise administrators are more concerned with the realistic long term performance of any device rather than the brand new performance as down time for TCL is simply not an option. Even though an Enterprise device will have many techniques for obfuscating and alleviating a degraded state (eg Idle Time Garbage Collection, multiple controllers, etc) there does come a point where these techniques fail to counteract the negative results of long term usage in an obviously non-TRIM environment. The point at which the performance falls and then plateaus at a lower performance level is known as the “steady state” performance or as “degraded state” in the consumer arena.

To help all consumer gain a better understanding of how much performance degradation there is between “optimal” and “steady state” we have included not only optimal results but have rerun tests after first degrading a drive until it plateaus and reaches its steady state performance level. These tests are labelled as “Steady State” results and can be considered as such.

While the standard for steady state testing is actually 8 hours we feel this is not quiet pessimistic enough and have extended the pre-test run to a full ten hours before testing actually commences. The pre-test or “torture test” consists of our standard “NonTrim performance test” and as such to quickly induce a steady state we ran ten hours of IOMeter set to 100% random, 100% write, 4k size chunks of data at a 64 queue depth across the entire array’s capacity. At the end of this test, the IOMeter file is deleted and the device was then tested using a given test sections’ unique configuration.

Processor: Core i7 4770

Motherboard: Asus Z97 Deluxe

Memory: 32GB G.Skill TridentX 2133

Graphics card: NVIDIA GeForce GTX 780

Hard Drive: Seagate 600 Pro 400GB SSD, Intel 910 800GB PCI-E SSD, OCZ RevoDrive 450 800GB

Power Supply: EVGA SuperNova 1000P2

Case: Cooler Master Storm Trooper

Special thanks to Intel for their support and supplying the i7 4770 CPU.
Special thanks to G.Skill for their support and supplying the TridentX Ram.
Special thanks to NVIDIA for their support and supplying the GTX 780s.
Special thanks to EVGA for their support and supplying the SuperNova PSU.
Special thanks to Cooler Master for their support and supplying the CM Storm Trooper

Below is a description of each SSD configuration we tested for this review:

Intel 910 800GB (Single Drive) HP mode: A single LUN of the Intel 910 800GB in its High Performance Mode

Intel 910 800GB (Raid 0 x2) HP mode: Two of the Intel 910 800GB SSD LUN's in High Performance Mode Configured in RAID 0

Intel 910 800GB (Raid 0 x4) HP mode: All four of the Intel 910 800GB SSD LUN's in High Performance Mode Configured in RAID 0

Intel DC S3500 480GB: A single DC S3500 480GB drive

Intel DC S3500 480GB (RAID 0): Two DC S3500 480GB drives Configured in RAID 0

Intel DC S3700 200GB: A single DC S3700 200GB drive

Intel DC S3700 800GB: A single DC S3700 800GB drive

Intel DC S3700 200GB (RAID 0): Two DC S3700 200GB drives Configured in RAID 0

Intel DC S3700 800GB (RAID 0): Two DC S3700 800GB drives Configured in RAID 0

Intel 710 200GB (RAID 0): Two 710 200GB drives Configured in RAID 0

Intel DC P3700 800GB: A single DC P3700 800GB drive

 

[AkG]

ATTO Disk Benchmark

ATTO Disk Benchmark

<i>The ATTO disk benchmark tests the drives read and write speeds using gradually larger size files. For these tests, the ATTO program was set to run from its smallest to largest value (.5KB to 8192KB) and the total length was set to 256MB. The test program then spits out an extrapolated performance figure in megabytes per second. </i>

<div align="center"><img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/atto_r.jpg" border="0" alt="" />
<img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/atto_w.jpg" border="0" alt="" /></div>

As expected the new DC P3700 scales very nicely with larger sized data. The combination of NVMe and a PCIe 3.0 interface almost assured that this was going to be the case. What was not entirely expected was the small file performance of this new drive where it not only outperforms a quad controller based Intel 910 800GB, but every enterprise drive we have tested to date.

 

[AkG]

Crystal DiskMark / AS-SSD

Crystal DiskMark

<i>Crystal DiskMark is designed to quickly test the performance of your hard drives. Currently, the program allows to measure sequential and random read/write speeds; and allows you to set the number of tests iterations to run. We left the number of tests at 5 and size at 100MB. </i>

<div align="center"><img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/cdm_r.jpg" border="0" alt="" />
<img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/cdm_w.jpg" border="0" alt="" /></div>

The merely 'outstanding' single queue depth 4k results are obviously because this device has been tuned to deep queue depths. At deep queue depths this drive is so fast that even the Intel 910 it replaces appears to be laggardly slow. To put this another way the P3700's small file performance is noticeably higher than Plextor's M.2 M6e's sequential file performance which brings new meaning to the word fast.

AS-SSD

<i>AS-SSD is designed to quickly test the performance of your drives. Currently, the program allows to measure sequential and small 4K read/write speeds as well as 4K file speed at a queue depth of 6. While its primary goal is to accurately test Solid State Drives, it does equally well on all storage mediums it just takes longer to run each test as each test reads or writes 1GB of data.</i>

<div align="center"><img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/asd_r.jpg" border="0" alt="" />
<img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/asd_w.jpg" border="0" alt="" /></div>

Once again this drive simply outclasses any device tested to date. Stellar sequential file performance, outstanding small file performance and nearly surreal deep queue depth performance is one hell of a combination. We also feel it bears mentioning that while the Intel 910 800GB's latency ranged from 80 microseconds (single LUN) to over 200 microseconds (4 LUN software RAID) the new DC P3700 was a <i>flat</i> 20 microseconds. That is basically a10x performance increase in one generation and this lowered latency will significantly help the IOMeter results. Though to be fair, once you start RAID'ing multiple DC P3700 together this advantage will noticeably shrink.

 

[AkG]

IOMETER: Our Standard Test

IOMETER: Our Standard Test

<i>IOMeter is heavily weighted towards the server end of things, and since we here at HWC are more End User centric we will be setting and judging the results of IOMeter a little bit differently than most. To test each drive we ran 5 test runs per device (1,4,16,64,128 queue depth) each test having 8 parts, each part lasting 10 min w/ an additional 20 second ramp up. The 8 subparts were set to run 100% random, 80% read 20% write; testing 512b, 1k, 2k,4k,8k,16k,32k,64k size chunks of data. When each test is finished IOMeter spits out a report, in that reports each of the 8 subtests are given a score in I/Os per second. We then take these 8 numbers add them together and divide by 8. This gives us an average score for that particular queue depth that is heavily weighted for single user environments and workstation environments.

In the first chart we have used our standard 1 client with 1 worker. In our second chart we have used two clients, each with two workers or four times the concurrent operations at a given queue depth. </i>

<div align="center">
<img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/iom.jpg" border="0" alt="" />
</div>

On first glance at the numbers enthusiasts and IT admins may feel a touch let down by these results. We must admit that we too were hoping for an even higher performance increase; however once you actually look at what this <i>single</i> drive with <i>single controller</i> is doing to a four controller based device the result do become impressive. When you then realize that the only numbers even coming close to the DC P3700 is the <i>overclocked</i> Intel 910 results, these results go from great to downright jaw droppingly great. We just wish we could overclock this beast like the last generation as then the results would probably have been giggle inducing!

 

[AkG]

IOMeter: File, Web, & Email Server Testing

IOMETER: File Server Test

<i>To test each drive we ran 6 test runs per device (1,4,16,64,128,256 queue depth) each test having 8 parts, each part lasting 10 min w/ an additional 20 second ramp up. The 6 subparts were set to run 100% random, 75% read 25% write; testing 512b, 4k,8k,16k,32k,64k size chunks of data. When each test is finished IOMeter spits out a report, in that reports each of the 6 subtests are given a score in I/Os per second. We then take these 8 numbers add them together and divide by 6. This gives us an average score for that particular queue depth that is heavily weighted for file server usage.

In the first chart we have used our standard 1 client with 1 worker. In our second chart we have used two clients, each with two workers or four times the concurrent operations at a given queue depth. </i>

<div align="center"><img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/iom_f.jpg" border="0" alt="" />
</div>

IOMETER: Web Server Test

<i>To test each drive we ran 6 test runs per device (1,4,16,64,128,256 queue depth) each test having 8 parts, each part lasting 10 min w/ an additional 20 second ramp up. The 8 subparts were set to run 100% random, 95% read 5% write; testing 512b, 1k, 2k,4k,8k,16k,32k,64k size chunks of data. When each test is finished IOMeter spits out a report, in that reports each of the 8 subtests are given a score in I/Os per second. We then take these 8 numbers add them together and divide by 8. This gives us an average score for that particular queue depth that is heavily weighted for web server environments.

In the first chart we have used our standard 1 client with 1 worker. In our second chart we have used two clients, each with two workers or four times the concurrent operations at a given queue depth. </i>

<div align="center">
<img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/iom_web.jpg" border="0" alt="" /> </div>

IOMETER: Email Server Test

<i>To test each drive we ran 5 test runs per drive (1,4,16,64,128 queue depth) each test having 3 parts, each part lasting 10 min w/ an additional 20 second ramp up. The 3 subparts were set to run 100% random, 50% read 50% write; testing 2k,4k,8k, size chunks of data. When each test is finished IOMeter spits out a report, in that reports each of the subtests are given a score in I/Os per second. We then take these numbers add them together and divide by 3. This gives us an average score for that particular queue depth that is heavily weighted for email server environments.

In the first chart we have used our standard 1 client with 1 worker. In our second chart we have used two clients, each with two workers or four times the concurrent operations at a given queue depth. </i>

<div align="center"><img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/iom_e.jpg" border="0" alt="" /> </div>


Thanks to NVMe and the lack of RAID overhead the new Intel DC P3700's finally answers the question that the Intel 910 posed: how much performance was being lost to overhead from software RAID and inefficient drivers? The answer is <i>a lot</i>.

At lower queue depths the differences are not all that apparent but once again only the Intel 910 800GB is capable of keeping up, but once the queue depths start to get deep not only is the DC P3700 able to easily outclass all devices but shows no sign of slowing down.

 

[AkG]

Steady State Testing: Standard, File, Web, & Email Server

IOMETER: Our Standard Steady State Test

To test each drive we first ran our default IOMeter test for 10 hours. At the end of this time period the device is in a steady state and ready for the actual testing to begin. To test each drive we ran 5 test runs per device (1,4,16,64,128 queue depth) each test having 8 parts, each part lasting 10 min w/ an additional 20 second ramp up. The 8 subparts were set to run 100% random, 80% read 20% write; testing 512b, 1k, 2k,4k,8k,16k,32k,64k size chunks of data. When each test is finished IOMeter spits out a report, in that reports each of the 8 subtests are given a score in I/Os per second. We then take these 8 numbers add them together and divide by 8. This gives us an average score for that particular queue depth that is heavily weighted for single user environments and workstation environments.

In the first chart we have used our standard 1 client with 1 worker. In our second chart we have used two clients, each with two workers or four times the concurrent operations at a given queue depth.

​

IOMETER: File Server Steady State Test

To test each drive we first ran our default IOMeter test for 10 hours. At the end of this time period the device is in a steady state and ready for the actual testing to begin. As with our standard IOMeter File Sever test, 6 test runs per device (1,4,16,64,128, 256 queue depth) each test having 8 parts, each part lasting 10 min w/ an additional 20 second ramp up. The 6 subparts were set to run 100% random, 75% read 25% write; testing 512b,4k,8k,16k,32k,64k size chunks of data. When each test is finished IOMeter spits out a report, in that reports each of the 6 subtests are given a score in I/Os per second. We then take these 8 numbers add them together and divide by 6. This gives us an average score for that particular queue depth that is heavily weighted for file server usage.

In the first chart we have used our standard 1 client with 1 worker. In our second chart we have used two clients, each with two workers or four times the concurrent operations at a given queue depth.

​

IOMETER: Web Server Steady State Test

To test each drive we first ran our default IOMeter test for 10 hours. At the end of this time period the device is in a steady state and ready for the actual testing to begin. As with our standard IOMeter Web Server test, 5 test runs per drive (1,4,16,64,128 queue depth) each test having each test having 8 parts, each part lasting 10 min w/ an additional 20 second ramp up. The 8 subparts were set to run 100% random, 95% read 5% write; testing 512b, 1k, 2k,4k,8k,16k,32k,64k size chunks of data. When each test is finished IOMeter spits out a report, in that reports each of the 8 subtests are given a score in I/Os per second. We then take these 8 numbers add them together and divide by 8. This gives us an average score for that particular queue depth that is heavily weighted for web server environments.

In the first chart we have used our standard 1 client with 1 worker. In our second chart we have used two clients, each with two workers or four times the concurrent operations at a given queue depth.

​

IOMETER: Email Server Steady State Test

To test each drive we first ran our default IOMeter test for 10 hours. At the end of this time period the device is in a steady state and ready for the actual testing to begin. As with the standard IOMeter email server test, this consists of 5 test runs per drive (1,4,16,64,128 queue depth) each test having 3 parts, each part lasting 10 min w/ an additional 20 second ramp up. The 3 subparts were set to run 100% random, 50% read 50% write; testing 2k,4k,8k, size chunks of data. When each test is finished IOMeter spits out a report, in that reports each of the subtests are given a score in I/Os per second. We then take these numbers add them together and divide by 3. This gives us an average score for that particular queue depth that is heavily weighted for email server environments.

In the first chart we have used our standard 1 client with 1 worker. In our second chart we have used two clients, each with two workers or four times the concurrent operations at a given queue depth.

​

This new DC P3700 is as close to the Energizer rabbit as you are likely to find: it just keeps going and going and going. Intel has made long term-performance stability a priority and this in combination with drastically reduced latency makes for what is easily the best SSD we have ever seen. Not only is it more capable than any drive tested before, it is also much more stable and predictable in its long-term performance.

While watching the various tests run, there weren't any sudden peaks or valleys like we usually see; the sheer reliability of the performance was almost mesmerizing.

 

[AkG]

Adobe CS5 Load Time / Firefox Portable

Adobe CS5 Stead State Load Time

<i>Photoshop is a notoriously slow loading program under the best of circumstances, and while the latest version is actually pretty decent, when you add in a bunch of extra brushes and the such you get a really great torture test which can bring even the best of the best to their knees. To make things even more difficult we have first placed the devices into a steady state so as to help recreate the absolute worst case scenario possible </i>

<div align="center"><img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/adobe.jpg" border="0" alt="" />
</div>

Firefox Portable Offline Steady State Performance

<i>Firefox is notorious for being slow on loading tabs in offline mode once the number of pages to be opened grows larger than a dozen or so. We can think of fewer worse case scenarios than having 100 tabs set to reload in offline mode upon Firefox startup, but this is exactly what we have done here.

By having 100 pages open in Firefox portable, setting Firefox to reload the last session upon next session start and then setting it to offline mode, we are able to easily recreate a worse case scenario. Since we are using Firefox portable all files are easily positioned in one location, making it simple to repeat the test as necessary. In order to ensure repetition, before touching the Firefox portable files, we have backed them up into a .rar file and only extracted a copy of it to the test device.

As with the Adobe test, we have first placed the devices into a steady state.</i>

<div align="center"><img src="http://images.hardwarecanucks.com/image/akg/Storage/DC_P3700/ff.jpg" border="0" alt="" /> </div>

This device is so fast that we can honestly say that even a potent combination of i7 4770 and DDR3-2133 is the bottleneck with these results.