Samsung’s 950 Pro 512GB SSD reviewed

Hardware innovation doesn’t usually happen overnight. The inertia inherent to component manufacturing prevents sweeping changes from coming around too often. The storage realm is no exception. SATA 6Gbps SSDs have been around since 2009, and they still have yet to be displaced as the staple of any mainstream enthusiast build.

However, a storage revolution has been brewing steadily for last few years. PCIe drives have long been available to those with thousands to spend on storage, but costs have come down considerably. These days, any mid-grade to high-end laptop worth its salt comes with a PCIe storage option, and M.2 slots can be found on nearly every modern Intel motherboard. Blessedly, notebook and motherboard manufacturers seem to have settled on a single physical form factor and connector (M.2 2280) to dedicate to consumer PCIe storage.

Intel’s Skylake platform takes full advantage of these developments. The Z170 chipset offers an abundance of PCIe Gen3 lanes for next-generation storage devices, allowing builders to run graphics cards and PCIe SSDs in tandem without cannibalizing precious bandwidth.

The next piece of the puzzle is NVMe. Poised to replace the aging AHCI standard, the NVMe protocol is engineered to scale much better with the broadly parallel NAND configurations in most SSDs nowadays. Windows 8.1 and 10 both support NVMe natively for secondary storage. Additionally, most Z97 and Z170 boards now support booting from an NVMe drive, either out of the box or with a firmware update.

So we have the interface, we have the connector, we have the protocol, and we have the software to support it all. The final piece of puzzle is an NVMe SSD—one built in the M.2 2280 form factor and driven by four PCIe Gen3 lanes. Samsung has already released one such drive: the NVMe version of its SM951 SSD. The SM951 wasn’t meant for consumers, though. It lacked support for Samsung’s Magician utility, and it was primarily focused towards OEMs.

Today, we have Samsung’s first consumer-friendly next-gen SSD. Say hello to the company’s latest and greatest, the 512GB version of the 950 Pro SSD. This drive represents the intersection of two novel technologies: V-NAND and the NVMe protocol. Let’s see if they play nice together.

Samsung has done a good job making this drive stand out aesthetically. Every other M.2 drive we’ve seen is just a wee sliver of green PCB with a sticker covering its naughty bits. The 950 Pro is a similar bite-sized chunk, but it sports a black-on-black look that lends a little personality to the otherwise drab hardware.

Underneath the sticker, we find the drive’s controller, its memory, and two NAND packages, which are loaded with Samsung’s 32-layer, 128Gb MLC V-NAND chips. This is a Pro-series drive, after all, so purists can rejoice. There’s no TLC flash here. The brain of the drive is the same UBX controller that Samsung developed to power the SM951 NVMe.

Capacity Max sequential (MB/s) Max Random (IOps) Price $/GB
Read Write Read Write
256GB 2200 900 270k 85k $199.99 $0.78
512GB 2500 1500 300k 110k $349.99 $0.68

Aside from the 512GB drive we have on hand, the 950 Pro also comes in a smaller 256GB variant, which is a bit slower in theoretical terms and more expensive per gigabyte. All 950 Pros will eventually have a full set of security features: 256-bit AES encryption is already implemented, and Samsung promises that Opal and eDrive support will be coming in a future firmware update. Additionally, the 950 series will launch alongside version 4.8 of the firm’s Magician software. This utility will have a few new tricks up its sleeve to handle drive maintenance NVMe-style. Samsung warrants the drive for five years or 400 total terabytes written.

A change in our test setup

Before we go any further, let’s talk about some changes to our test setup. Say goodbye to the old old TR storage rig, and say hello to the new old TR storage rig. Our beefy main storage-testing system is back in commission, so the Endurance Experiment box we brought out of retirement for the last couple of reviews can head back out to pasture.

Since this is the same hardware that we were using to test SSDs not so long ago, we have access to a much larger set of data to compare fresh drives against. The backbone of the rig, the Asus Z97-Pro, has been updated to the latest firmware so that we can boot from NVMe PCIe drives. We’ve done extensive validation to ensure our secondary storage testing results have not changed between firmware revisions.

That said, the boot performance has changed enough that we can’t recycle the old numbers. Therefore, you will notice that while our IOMeter and RoboBench result sets are chock full of data, our boot and load time results include a much smaller set that we re-tested.

Now that we’re all ready to test the 950 Pro, let’s get down to it.

 

A slight hitch?

As we put the 950 Pro through our standard storage test suite, we noticed a few surprising results. It seems our test unit is performing less well than one might expect in certain IOMeter workloads compared to Samsung’s claimed performance specifications, especially those heavy on random writes. We aren’t sure of the exact reason for the numbers we saw, but there are a number of potential culprits—thermal throttling, a bug in Samsung’s beta NVMe driver, or perhaps an issue with our Windows 8.1 test image. We also can’t rule out the possibility that the drive is simply performing as expected.

We’ve brought our random-write results to Samsung’s attention, and we’ll update the review when we hear back. In the meantime, we’re forging ahead, since our other results for the drive appear to be OK. Without further ado, let’s move on to the benchmarks.

IOMeter — Sequential and random performance

IOMeter fuels much of our latest storage test suite, including our sequential and random I/O tests. These tests are run across the full capacity of the drive at two queue depths. The QD1 tests simulate a single thread, while the QD4 results emulate a more demanding desktop workload. For perspective, 87% of the requests in our old DriveBench 2.0 trace of real-world desktop activity have a queue depth of four or less. Clicking the buttons below the graphs switches between results charted at the different queue depths.

Our sequential tests use a relatively large 128KB block size.



The 950 Pro posts healthy sequential numbers. Its read speeds at both QD1 and QD4 are right up there with the greats, trailing just a little behind Intel’s excellent 750 Series drive (the only other consumer NVMe drive in this dataset). Samsung’s SM951, an ACHI-based drive, still leads the pack at QD1, however, and it bests the 950 Pro at QD4. The 950 Pro’s write numbers are good, too. This drive ends up comfortably ahead of the SATA pack, but it’s not quite as strong as Intel’s NVMe offerings.

Next, we’ll turn our attention to performance with 4KB random I/O. The tests below are based on the median of three consecutive three-minute runs. SSDs typically deliver consistent sequential and random read performance over that period, but random write speeds worsen as the drive’s overprovisioned area is consumed by incoming writes. We’ve reported average response times rather than raw throughput, which we think makes sense in the context of system responsiveness.



The 950 Pro pulls ahead head of the pack with random-read response times at both queue depths, beating out even Intel’s datacenter NVMe drive, the P3700. On the other hand, its random-write response times are lackluster, losing to a number of cheaper, older SATA drives. This is an example of where we believe the drive is stumbling for some reason, but like we said, these could be expected results. The SM951 doesn’t do any better here.

As we noted, the preceding tests are based on the median of three consecutive three-minute runs. SSDs typically deliver consistent sequential and random-read performance over that period, but random-write speeds worsen as the drive’s overprovisioned area is consumed by incoming writes. We explore that decline in the next set of tests.

Sustained and scaling I/O rates

Our sustained IOMeter test hammers drives with 4KB random writes for 30 minutes straight. It uses a queue depth of 32, which should result in higher speeds that saturate each drive’s overprovisioned area more quickly. This lengthy—and heavy—workload isn’t indicative of typical PC use, but it provides a sense of how the drives react when they’re pushed to the brink.

We’re reporting IOps rather than response times for these tests. Click the buttons below the graph to switch between SSDs.


Not exactly a revolutionary showing here. This test is another example of our seemingly odd write results. The drive’s peak and steady-state rates are about on par with most of the SATA drives we tested.

To show the data in a slightly different light, we’ve graphed the peak random-write rate and the average, steady-state speed over the last minute of the test.

As we said, the 950 Pro falls close to the middle of the SATA pack. That’s a bit frustrating, given how well it performs in our read synthetics.

Our final IOMeter test examines performance scaling across a broad range of queue depths. We ramp all the way up to a queue depth of 128. Don’t expect AHCI-based drives to scale past 32, though; that’s the max depth of their native command queues.

We use a database access pattern comprising 66% reads and 33% writes, all of which are random. The test runs after 30 minutes of continuous random writes that put the drives in a simulated used state. Click the buttons below the graph to switch between the different drives. And note that the P3700 plot uses a much larger scale. We’ll compare several of the PCIe drives on that scale in a moment.


The 950 Pro’s performance flattens out around a queue depth of eight. It certainly scales better than the AHCI SM951 before it, but Intel’s 750 Series drives have no trouble scaling even further still. Puzzlingly, the performance issue extends to random reads as well as writes this time around. In general, we would expect NVMe drives to scale well beyond a queue depth of just eight.

Below, we use the same data to plot the 950 Pro against the two Intel powerhouses, as well as Samsung’s SM951. Click to toggle between read, write, and total IOps.


These graphs really put a point on the 950 Pro’s scaling performance. It flatlines not all that far above the AHCI SM951. Meanwhile, Intel’s NVMe drives rocket skyward as the queue depth increases, scaling all the way to a queue depth of 128.

 

TR RoboBench — Real-world transfers

RoboBench trades synthetic tests with random data for real-world transfers with a range of file types. Developed by our in-house coder, Bruno “morphine” Ferreira, this benchmark relies on the multi-threaded robocopy command build into Windows. We copy files to and from a wicked-fast RAM disk to measure read and write performance. We also cut the RAM disk out of the loop for a copy test that transfers the files to a different location on the SSD.

Robocopy uses eight threads by default, and we’ve also run it with a single thread. Our results are split between two file sets, whose vital statistics are detailed below. The compressibility percentage is based on the size of the file set after it’s been crunched by 7-Zip.

  Number of files Average file size Total size Compressibility
Media 459 21.4MB 9.58GB 0.8%
Work 84,652 48.0KB 3.87GB 59%

The media set is made up of large movie files, high-bitrate MP3s, and 18-megapixel RAW and JPG images. There are only a few hundred files in total, and the data set isn’t amenable to compression. The work set comprises loads of TR files, including documents, spreadsheets, and web-optimized images. It also includes a stack of programming-related files associated with our old Mozilla compiling test and the Visual Studio test on the next page. The average file size is measured in kilobytes rather than megabytes, and the files are mostly compressible.

RoboBench’s write and copy tests run after the drives have been put into a simulated used state with 30 minutes of 4KB random writes. The pre-conditioning process is scripted, as is the rest of the test, ensuring that drives have the same amount of time to recover.

Let’s take a look at the media set first. The buttons switch between read, write, and copy results.



These RoboBench tests are where the 950 Pro really starts to shine. Working on the media set with a single thread, it beats Intel’s 750 in read, write, and copy alike. The 950 Pro even edges out the pricey P3700 in the read test. When eight threads are in the picture, the 750 regains some ground, but the 950 Pro’s numbers are impressive nonetheless.



The work set is a harder beast to tackle, with much smaller file sizes and less of a sequential workload compared to the media set. The 950 Pro chews through this workload with aplomb,  topping the charts with one thread. At eight threads, it doesn’t take first place, but it still puts on a good show, staying within a 10% striking distance of the dominant P3700. The SM951 remains a blazingly quick drive here.

Whatever specter was haunting our 950 Pro during our IOMeter synthetics has not hindered it at all in our RoboBench suite. Samsung’s first consumer NVMe drive manages to trade blows with Intel’s finest—but the AHCI SM951 still puts in a good showing.

 

Boot times

Thus far, all of our tests have been conducted with the SSDs connected as secondary storage. This next batch uses them as system drives. As we mentioned on the first page, we updated the firmware of our ASUS Z97-Pro motherboard, changing the boot characteristics of the system. Therefore, we have an abbreviated set of results from drives we re-tested for this review. This time around, we used Intel’s latest firmware for the 750 series, specifically crafted to improve its rather pokey boot times.

We’ll start with boot times measured two ways. The bare test depicts the time between hitting the power button and reaching the Windows desktop, while the loaded test adds the time needed to load four applications—Avidemux, LibreOffice, GIMP, and Visual Studio Express—automatically from the startup folder. Our old boot tests focused on the time required to load the OS, but these new ones cover the entire process, including drive initialization.

Intel’s new firmware does seem to have helped the 750 along, but it still boots quite a bit slower than its contemporaries. The 950 Pro boots promptly, as fast as any run-of-the-mill SATA drive. NVMe may not make your computer boot any faster, but the 950 Pro is proof that it doesn’t have to slow you down, either.

Load times

Next, we’ll tackle load times with two sets of tests. The first group focuses on the time required to load larger files in a collection of desktop applications. We open a 790MB 4K video in Avidemux, a 30MB spreadsheet in LibreOffice, and a 523MB image file in GIMP. In the Visual Studio Express test, we open a 159MB project containing source code for the LLVM toolchain. Thanks to Rui Figueira for providing the project code.

I’m running out of ways to say that modern SSDs all perform mostly the same in real-world load time tests, even when their scores in synthetics vastly differ. The 950 Pro does a fine job loading our productivity programs. Let’s move on.

Similarly, level load times are all within a few percentage points of each other. The 950 Pro is as good an SSD for gaming as any other.

Ordinarily, we’d devote some space here to discuss power consumption, but unfortunately we don’t have a process in place to measure the draw of PCIe-based storage yet. The method we use for SATA drives doesn’t carry over. As a result, that’s it for performance testing—read on for a breakdown of our hardware and test methods.

 

Test notes and methods

Here’s are the essential details for all the drives we tested:

  Interface Flash controller NAND
Crucial BX100 500GB SATA 6Gbps Silicon Motion SM2246EN 16-nm Micron MLC
Crucial MX200 500GB SATA 6Gbps Marvell 88SS9189 16-nm Micron MLC
Intel X25-M G2 160GB SATA 3Gbps Intel PC29AS21BA0 34-nm Intel MLC
Intel 335 Series 240GB SATA 6Gbps SandForce SF-2281 20-nm Intel MLC
Intel 730 Series 480GB SATA 6Gbps Intel PC29AS21CA0 20-nm Intel MLC
Intel 750 Series 1.2TB PCIe Gen3 x4 Intel CH29AE41AB0 20-nm Intel MLC
Intel DC P3700 800GB PCIe Gen3 x4 Intel CH29AE41AB0 20-nm Intel MLC
Plextor M6e 256GB PCIe Gen2 x2 Marvell 88SS9183 19-nm Toshiba MLC
Samsung 850 EV0 250GB SATA 6Gbps Samsung MGX 32-layer Samsung TLC
Samsung 850 EV0 1TB SATA 6Gbps Samsung MEX 32-layer Samsung TLC
Samsung 850 Pro 500GB SATA 6Gbps Samsung MEX 32-layer Samsung MLC
Samsung 950 Pro 512GB PCIe Gen3 x4 Samsung UBX 32-layer Samsung MLC
Samsung XP941 256GB PCIe Gen2 x4 Samsung S4LN053X01 19-nm Samsung MLC
Samsung SM951 512GB PCIe Gen3 x4 Samsung S4LN058A01X01 16-nm Samsung MLC
Samsung 850 Pro 500GB SATA 6Gbps Samsung MEX 32-layer Samsung MLC
OCZ Vector 180 240GB SATA 6Gbps Indilinx Barefoot 3 M10 A19-nm Toshiba MLC
OCZ Vector 180 960GB SATA 6Gbps Indilinx Barefoot 3 M10 A19-nm Toshiba MLC

All the SATA SSDs were connected to the motherboard’s Z97 chipset. The M6e was connected to the Z97 via the motherboard’s M.2 slot, which is how we’d expect most folks to run that drive. Since the XP941 and 950 Pro requires more lanes, they were connected to the CPU via a PCIe adapter card. The 750 Series and DC P3700 were hooked up to the CPU via the same full-sized PCIe slot.

We used the following system for testing, if you missed its turn on the runway on the first page:

Processor Intel Core i5-4690K 3.5GHz
Motherboard Asus Z97-Pro
Firmware 2601
Platform hub Intel Z97
Platform drivers Chipset: 10.0.0.13

RST: 13.2.4.1000

Memory size 16GB (2 DIMMs)
Memory type Adata XPG V3 DDR3 at 1600 MT/s
Memory timings 11-11-11-28-1T
Audio Realtek ALC1150 with 6.0.1.7344 drivers
System drive Corsair Force LS 240GB with S8FM07.9 firmware
Storage Crucial BX100 500GB with MU01 firmware

Crucial MX200 500GB with MU01 firmware

Intel 335 Series 240GB with 335u firmware

Intel 730 Series 480GB with L2010400 firmware

Intel 750 Series 1.2GB with 8EV10171 firmware

Intel DC P3700 800GB with 8DV10043 firmware

Intel X25-M G2 160GB with 8820 firmware

Plextor M6e 256GB with 1.04 firmware

OCZ Vector 180 240GB with 1.0 firmware

OCZ Vector 180 960GB with 1.0 firmware

Samsung 850 EVO 250GB with EMT01B6Q firmware

Samsung 850 EVO 1TB with EMT01B6Q firmware

Samsung 850 Pro 500GB with EMXM01B6Q firmware

Samsung 950 Pro 512GB with 1B0QBXX7 firmware

Samsung XP941 256GB with UXM6501Q firmware

Power supply Corsair AX650 650W
Operating system Windows 8.1 Pro x64

Thanks to Asus for providing the systems’ motherboards, Intel for the CPUs, Adata for the memory, and Corsair for the system drives and PSUs. And thanks to the drive makers for supplying the rest of the SSDs.

We used the following versions of our test applications:

Some further notes on our test methods:

  • To ensure consistent and repeatable results, the SSDs were secure-erased before every component of our test suite. For the IOMeter database, RoboBench write, and RoboBench copy tests, the drives were put in a simulated used state that better exposes long-term performance characteristics. Those tests are all scripted, ensuring an even playing field that gives the drives the same amount of time to recover from the initial used state.

  • We run virtually all our tests three times and report the median of the results. Our sustained IOMeter test is run a second time to verify the results of the first test and additional times only if necessary. The sustained test runs for 30 minutes continuously, so it already samples performance over a long period.

  • Steps have been taken to ensure the CPU’s power-saving features don’t taint any of our results. All of the CPU’s low-power states have been disabled, effectively pegging the frequency at 3.5GHz. Transitioning between power states can affect the performance of storage benchmarks, especially when dealing with short burst transfers.

The test systems’ Windows desktop was set at 1920×1080 at 60Hz. Most of the tests and methods we employed are publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

 

Conclusions

The Samsung 950 Pro, much like the Intel 750 Series drives before it, makes a splash with its combination of PCIe Gen3 x4 performance and the NVMe protocol. In our review of the Intel 750 Series earlier this year, we called it the future of solid-state storage. The 950 Pro doesn’t wrest that crown from Intel’s drives, but it still makes a compelling case for itself. Its sequential performance and real-world file I/O rates rival the 750 Series drives, and its boot times blow Intel’s out of the water.

There are a couple of flies in the ointment, however. The 950 Pro doesn’t scale as well in our synthetic random workloads as the Intel drives do, and it delivers middling performance that’s comparable to SATA SSDs in our sustained random write tests. Samsung’s own SM951 is often faster—sometimes, a lot faster—in some tests, as well, even if it doesn’t scale as well as the 950 Pro in others. We’ve pinged Samsung to ask whether our results with the 950 Pro in these areas are somehow atypical, and we’ll update this review with more information when we have it.

 

One solid advantage the 950 Pro has over the 750 Series is its form factor. The Intel drive comes in a half-height PCIe card, forcing you to give up one of your PCIe 3.0 slots in the name of storage. Intel also makes a version of the 750 Series drive that comes in a 2.5″ package, but that version requires an uncommon U.2 connector or an M.2-to-U.2 adapter card to work. The 950 Pro’s M.2 form factor is far more convenient, since this slot is ubiquitous on newer motherboards.

So what do you have to shell out to take the svelte 950 Pro home with you? The 512GB version we tested carries a suggested price of $350, or $0.68 per gigabyte. Meanwhile, the 400GB Intel 750 Series costs $360 ($0.90 per GB), while the 1.2TB version costs $900 ($0.75 per GB). In both performance and price, the 950 Pro ends up at the lower realms of the high end, and that’s not a bad place to be.

Though those prices are far lower than what PCIe storage has commanded in the past, it’s still hard to recommend these fancy NVMe drives to the average enthusiast. SATA drives perform just as well in everyday productivity and gaming scenarios, despite their theoretical AHCI handicap. Good SATA SSDs also sell for just 30 to 40 cents per gigabyte, so buying drives that cost twice that amount just isn’t worth it for most ordinary workloads. Those folks dealing with heavily storage-bound use cases are really the only ones who’ll see a good return on the higher investment.

On the other hand, if you were already sold on Samsung’s previous Pro series drives, then the 950 Pro is a great buy. This new drive is priced below the 850 Pro 512GB’s launch price of $400, yet it makes use of PCIe and NVMe. It may be a better deal than any Samsung Pro SSD yet.

The 950 Pro is good news for the rest of us if only because it gives Intel some much-needed competition. As more PCIe x4 and NVMe drives enter the market, prices should start to fall to more palatable levels. Looking ahead, the PCIe drive scene will hopefully see some more competition soon, as OCZ readies its NVMe RevoDrive 400 for release later this year.TR

Comments closed
    • odizzido
    • 4 years ago

    That sticker is so sexy. I wonder if they sell them separately?

    • Legend
    • 4 years ago

    XPoint next year?

      • Firestarter
      • 4 years ago

      according to micron, yes somewhere in 2016

      [url<]http://i.imgur.com/bCedW5B.png[/url<] [url<]http://i.imgur.com/RiU3OxS.png[/url<]

        • UberGerbil
        • 4 years ago

        At the kind of prices only big enterprise customers will buy. Like everything else, it’ll take a while after general commercial introduction before prices drop even into the “sane enthusiast” price range. Remember when 64GB SSDs were over a thousand dollars?

        • Legend
        • 4 years ago

        Awesome!

    • hasseb64
    • 4 years ago

    Absolutly NO real world benefits with this drive compared to an old sata drive, why am I waiting for this tech?
    Thank you for good review!

      • joselillo_25
      • 4 years ago

      As I have said below I think this drives could shine in a low RAM environment, so a benchmark on this situations could be a nice review. For example, test window page file speed in a computer with only 2GB of RAM or in Photoshop when all the RAM is used.

      Please Tech Report, give us one of this if possible.

        • Firestarter
        • 4 years ago

        that’s highly contrived and completely irrelevant for just about anyone who can afford this drive. For the difference between this drive and a regular SATA6 SSD, you can upgrade your RAM capacity by quite a lot

          • joselillo_25
          • 4 years ago

          I have deleted many 25GB temp photoshop files for myself, so I do not think is ridiculous, specially if you got a lot of UNDO states in many documents at the same time. A easy test will be opening huge amounts of images and apply them modifications to increase the size of the page file and later see if there is any difference in the speed photoshop recharges the images.

          Also with RAW images or huge 32 bits images. I think is difficult to store all of this in RAM, windows page file and adobe scratch are going to show no matter how many RAM you got.

          My statement can be controversial but do not think is ridiculous, when you work with this huge images you aslo use your computer for other tasks so windows page file always page to let room for other programs.

            • Firestarter
            • 4 years ago

            with files that large, wouldn’t you want socket 2011 and a ridiculous amount of RAM? Until X-point memory comes along I’d say anything that isn’t RAM is just too slow if you’re looking for productivity

        • Meadows
        • 4 years ago

        That’s absolutely ridiculous. The people who buy these kinds of things are the same people who would never skimp on RAM either.

        [i<]With that said[/i<], Photoshop [i<][b<]can and does fill[/b<][/i<] 8 GiB or even 16 GiB setups, so it needs a "scratch disk" set at all times. For what it's worth, testing performance on a several hundred megabyte many-layered 16-bpc image (which fills anywhere from 10 to 20 GiB of memory) could potentially have useful results. Still, it's only relevant to a small subset of readers. (Then again, so is shotgun proteomics with regards to CPUs, but they still tested that too.)

      • Freon
      • 4 years ago

      For the most part, true, real-world affects are minimal, but not zero as you attempt to state.

        • hasseb64
        • 4 years ago

        I like your name! I have 50kg R12, NEW! Been around since late 80-ties

      • kuttan
      • 4 years ago

      In real world scenario performance of SSD drives were mainly depended on the application you use. For example when you are installing a big game, here install speed is really depended on the installers number of CPU cores/threads use, the type of load the installer puts on to the CPU core(s) etc determines installation time.

      If the installer uses only one CPU core then you are really CPU bottlenecked than your SSD or interface speed. This is why many reviews of such high speed SSDs like this one needs to use a RAM drive to achieve maximum possible transfer speed form the drive, no other real world use can get that peak 1 or 2 GB/sec speed. One notable advantage with NVMe PCI-E drives I saw is having ultra low access latency that may offer noticeable performance difference though.

    • Mr Bill
    • 4 years ago

    Which AMD chipsets able to support 4 lanes of PCIe Gen3?
    Answer: Only the A88X (FM2+)

    Since we seem to have a reading comprehension vs inductive reasoning fail problem here I will try to make this more clear.
    I maintain there is a distinction between “provide” and “support”, because the word “provide” implies furnishing, supplying, sourcing. Wheras the word “support” implies bearing, carrying, assisting, aiding. This was my intent in using those words.

    Does the A88X “provide” PCIe Gen3 to the motherboard… NO, it is not present in the silicon of the chipset, and so PCIe Gen3 is not “provided” to the PCIe connector on the motherboard.

    Does the A88X “support” PCIe Gen3 on the motheboard… YES, if and only if a Kaveri APU is present in the socket (which “provides” PCIe Gen3 via logic in the APU), then the A88X chipset “supports” PCIe Gen3 by passing it through from the APU to the PCIe connector on the motherboard.

      • chuckula
      • 4 years ago

      None that I’m aware of.

      The Kaveri APUs do support PCIe Gen 3 but only using the integrated PCIe controller. The southbridge parts [which are effectively identical to the southbridge parts in use going all the way back to Llano in 2011] do not appear to do so.

        • Mr Bill
        • 4 years ago

        So, not even the FM2+ motherboards? Bleh.

        Edit: Ah! But some can!

        • Mr Bill
        • 4 years ago

        The A88X (FM2+) chipset supports PCIe Gen 3 and only supports Gen 3 with the Kavieri APU. Earlier generation APU’s fault back to Gen 2. Good decision to go with an Asus A88X-Pro motherboard in my latest build (e.g. 2 x PCIe 3.0 x16 (x16 or x8/x8)).

          • chuckula
          • 4 years ago

          Does it?

          Remember: There are PCIe 3.0 lanes [b<]on the Kaveri APU itself[/b<] but that doesn't mean the [b<]chipset[/b<] has PCIe 3.0 support. Here's a listing from Newegg that seems to indicate the chipset lacks PCIe 3.0 since only 16 lanes (the ones from the APU) are listed: [url<]http://www.newegg.com/Product/Product.aspx?Item=N82E16813132056[/url<] Here's a review that says pretty much the same thing: [url<]http://www.hardwaresecrets.com/asus-a88x-pro-motherboard/2/[/url<] Remember that if you use PCIe 3.0 lanes from the APU then you have at best an 8-lane connection for any external GPU. Which isn't the end of the world, but it is a sacrifice.

            • Mr Bill
            • 4 years ago

            [url=http://www.gamersnexus.net/guides/1442-amd-apu-chipset-comparison-a88-a85-a78<]gamersnexus chipset comparison[/url<] and [url=https://www.asus.com/us/Motherboards/A88XPRO/<]Asus A88X-Pro (FM2+) motherboard[/url<]

            • chuckula
            • 4 years ago

            I don’t think you really understand what I’m saying.
            Both of those links you posted back up my point 100%.

            The A88X [b<]CHIPSET[/b<] supports exactly zero PCIe 3.0 lanes. Your first link goes out of its way to point out that it supports exactly four PCIe [b<]2.0[/b<] lanes. Once again, the Kaveri APU is AMD's only product on the market that supports PCIe 3.0 [b<]from pins located on the APU itself[/b<], but as I mentioned, any peripheral that uses any of those lanes takes away from the potential for a discrete GPU to use them.

            • Mr Bill
            • 4 years ago

            The second table in that first link I posted shows that the PCI-e Gen for A88X is 3.0 and that it supports 1X16 or 2X8. [quote<]A88X on the FM2+ platform will support PCI-e Gen3 slots when coupled with a Kaveri processor, but only when paired with Kaveri; although Richland is supported on A88X, it will not make use of PCI-e Gen3 capabilities and will instead revert to Gen2. PCI-e lane allocation is the same on A88X as it is on A85X: 1x16 or 2x8 can be configured, so two GPUs is the best you'll do.[/quote<] [quote<]Key Differences: A88X and A85X are each the top of the line for their respective platforms. They're not meant as direct competitors, so you're basically either choosing A85X or A88X based on whether you're using Richland or Kaveri (respectively). Other than that, [b<]A88X has added full PCI-e Gen3 support when used with Kaveri and has the upgraded AHCI/RAID controller, but that's about it for differences.[/b<][/quote<] Your own second link says...[quote<]The ASUS A88X-PRO comes with two PCI Express 3.0 x16 slots, one PCI Express 2.0 x16 (that always works at x4 speed), two PCI Express 2.0 x1, and two PCI slots. The PCI Express 3.0 x16 slots share bandwidth, which means that if the second one is unused, the first one works at x16 speed, but if both are populated, they will both work at x8. You also need to keep in mind that PCI Express 3.0 is only supported if an FM2+ processor is installed. With an FM2 CPU, the slots only support PCI Express 2.0.[/quote<] So the chipset supports it if the processor supports it (e.g. Kaveri APU). I have an A10-7850K on an A88X-Pro motherboard which uses the A88X chipset.

            • chuckula
            • 4 years ago

            Dude, reading comprehension fail. That table just says that the [b<]KAVERI APU PART[/b<] has some PCIe 3.0 built in. I already knew that, and the table DID NOT contradict anything I've said. In fact, it proves I'm right. I need you to conduct this mental exercise and then come back with a coherent technical explanation of why you think you are right after you have thought things through: 1. Understand that Kaveri is an APU. It is a piece of silicon that goes in a socket on a motherboard. 2. Understand that as part of being an APU, Kaveri includes some PCIe 3.0 pins ON THE PHYSICAL PIECE OF SILCON ITSELF. These pins can be connected to up to two of the PCIe slots on the motherboard, typically for GPUs. 3. Understand that the A88X [b<]IS A COMPLETELY DIFFERENT PIECE OF SILICON SOLDERED TO A DIFFERENT PART OF THE MOTHERBOARD[/b<]. There are [b<]EXACTLY ZERO[/b<] PCIe 3.0 lanes on this chipset, and literally every link you have posted verifies exactly what I said, all the way down to the block that says "PCIe 2.0 4x" that is literally staring you right in the face in the very link you posted. 4. Understand that except for the PCIe slots leading back to the APU directly, ALL of the PCI express lanes on the motherboard are version 2.0 and lead back to the CHIPSET which is a COMPLETELY DIFFERENT PIECE OF SILICON. 5. Understand that this repetitious point you keep posting about how the A88x "magically gains PCIe 3.0 support with Kaveri" [b<]actually proves everything I have said to be correct[/b<]. The [b<]KAVERI APU[/b<] has some PCIe 3.0 ports that are best used for a GPU. The A88X [b<]completely different piece of silicon[/b<] has a grand total of ZERO PCIe 3.0 lanes. ZERO. The reason that these motherboards "magically" gain PCIe 3.0 support only with Kaveri is because the CHIPSET which is a SEPARATE PIECE OF SILICON does NOT have any PCIe 3.0 support AT ALL. That's why if you don't plug Kaveri into the motherboard, you magically lose the PCIe 3.0 connectivity THAT IS NOT PRESENT IN THE SEPARATE PIECE OF SILICON CALLED THE A88X. Can you answer me this simple question: If the A88X has this magical PCIe 3.0 connectivity already, why the hell should it be necessary for a completely separate piece of silicon on the motherboard to be present to "activate" it? The chipset either has the support or doesn't have the support, and as anybody with a modicum of technical skill and reading comprehension understands, you have posted multiple links that all agree about the A88X having ZERO PCIe 3.0 support.

            • Mr Bill
            • 4 years ago

            Dude, inductive reasoning fail. Here is the [url=http://support.amd.com/TechDocs/Bolton_D2-D2H-D3-D4_Databook.pdf<]A88X Bolton FCH Databook[/url<]. On page 16 it says there is only a PCIe 2.0 controller built into this FCH. And, as you sarcastically point out, ther is no PCIe 3.0 controller is in this silicon. My intent was to show that motherboards built with the A88X can have PCIe 3.0 support in their PCIe slots which is good enough to know, regardless of which piece of silicon is actually providing the PCIe 3.0. However, none of the previous generation APU "chipsets" or hub or FCH, whatever, can either provide or "support" PCIe 3.0 regardless of APU that is present. So, the A88X has some sort of "support" as in enabling PCIe 3.0 via the APU but not "support" as in having a PCIe 3.0 controller.

            • Mr Bill
            • 4 years ago

            This is not a reading comprehension failure. The only way I can see that your argument is correct is if you are suggesting that the PCIe 3.0 traces go from the APU socket directly to the PCIe connector. Since the A88X is providing the PCIe 2.0 lines; this would mean that there are two sets of lines going to the PCIe connector on the motherboard. This seems unnecessarily complex.

            Inductive reasoning says that it would be simpler if the PCIe 3.0 traces go through the A88X and logic within the A88X detects whether a Kavari GPU is providing PCIe 3.0 and then passes it through the appropriate traces to the PCIe connector; or it detects no Kavari GPU and instead “provides” PCIe 2.0 from its own silicon.

    • WaltC
    • 4 years ago

    Just curious…why 8.1 & not Win10? My Evo 850 boots a bit faster in 10, unquestionably.

    Let me guess: is it because of the Win10 Magician software all of us Evo owners are still waiting on…? Samsung is really, really dragging its feet on its supporting Win10 software for the Sata EVO series. Great boot drive, and sure, I can use it in Win10 without specific Magician support (well most of Magician support works in 10, just not all of it.) But I still have to wonder…working on 4 months without specific Win10 support…and Samsung still isn’t saying much about it except…”soon.”

      • Krogoth
      • 4 years ago

      A.) Windows 10 is the new baseline that Damage and crew are using for benches.

      B.) Windows 10 has native NVM Express support while everything else older requires updates via stripstreaming or is EOL (2003 or older).

        • Meadows
        • 4 years ago

        They’re not using Windows 10.

    • DarkMikaru
    • 4 years ago

    Not a bad showing at all Samsung. I’m thinking of finishing off my new 6600k build with an NVMe based SSD for the OS and the 250GB version for 199 ain’t bad. Which means if they release a 128GB version it will probably be between 130 – 160.. I might bite at that price.

    My only concern with M.2 is you give up 2 Sata ports when you do so. And with only 6 available that worries me a bit for future expansion. M.2 would be sweet if it were in addition to the 6 ports on board. What do you guys think? Should I just stick with good old traditional ports or give the M.2 a spin?

    • UnfriendlyFire
    • 4 years ago

    There should’ve been a comparison with a HDD just to see how all of the SSDs compare to it, and to give a perspective of the slowest SSDs still being faster than the fastest HDDs.

    Unless if you came across one of those magical SSDs that are only as fast as a 5400 RPM HDD. Notebookcheck dinged a certain laptop model after finding that the company was using two different SSD models (same capacity) for the same laptop model, and one of them was horribly slow.

    • webkido13
    • 4 years ago

    After reading the review I’m now torn. Is it worth waiting for retail availability of the 950 Pro or should I just go for a NVMe SM951? The NVMe SM951 just became available a couple of weeks ago here (small country surrounded by EU countries). Any advice?

    • joselillo_25
    • 4 years ago

    a good test will be checking perfomance using windows page file or Photoshop scratch disk. So try to use a system with only 2 of 4 RAM and make some tests to see if we can see the teorical NVMe perfomance increase in some real world tasks.

    • Mr Bill
    • 4 years ago

    Vapor chamber cooler plates are now available in a memory form factor. It would be interesting to see a review of some examples of this new storage medium using those cooler plates. [url=http://celsiainc.com/blog-vapor-chambers-heat-pipes-cool-performance-fbdimms/<]one example[/url<]. [url=http://celsiainc.com/blog-integrating-vapor-chambers-into-heatsinks/<]A better example with a picture[/url<]

    • cynan
    • 4 years ago

    It would be nice to see a non-Samsung PCIe M.2 comparison. Any chances of getting something like the 480GB a Kingston HyperX Predator in for testing?

    • flip-mode
    • 4 years ago

    [quote<]Samsung has done a good job making this drive stand out aesthetically. [/quote<] LOL, that's just dandy. At this point I am probably the only one that doesn't give a damn.

      • Meadows
      • 4 years ago

      No you’re not. You have my sword.

        • chuckula
        • 4 years ago

        And my axe!

          • Meadows
          • 4 years ago

          Silly Chuckula, bow comes second.

            • chuckula
            • 4 years ago

            Yeah, but I look more like Gimli than Orlando Bloomolas.

            • Meadows
            • 4 years ago

            Disturbing mental imagery.

      • WaltC
      • 4 years ago

      Good point, and I thought the same thing. Who gives a “flip” (sorry) what the damned card looks like superficially? It’s not living-room furniture–I mean, gee, what do they want? For internal cards to be festooned with bright-red Christmas-tree LEDs?…;) [I lament at how far the geek world has fallen! The superficial has gained so much traction–it’s all a part of the Apple poison polluting the tech tree, don’t’ya know…;)]

        • Meadows
        • 4 years ago

        [quote<]"Who gives a "flip" (sorry)"[/quote<] Is your name Maurice Moss, perchance?

    • terminalrecluse
    • 4 years ago

    Glad to see QD1 results. For desktop usage most things are QD1 or so making this review very much relevant.

    • MaXGTS
    • 4 years ago

    They used a black PCB. YAY!

    • wierdo
    • 4 years ago

    Would be nice if they slap some heatsinks on these things, I gather they get pretty hot, not sure how that translates in terms chip lifespan and throttling.

    • nanoflower
    • 4 years ago

    One minor editing issue I noticed on page one. The line ” The piece of puzzle is an NVMe SSD” should read ” The final piece of puzzle is an NVMe SSD”. Otherwise a fine job overall.

      • Jeff Kampman
      • 4 years ago

      Good catch, thanks.

    • HisDivineOrder
    • 4 years ago

    Doesn’t surprise me Samsung is having problems with its drives. It’s just good that finally it’s captured in a review before people buy one. Because if the last few releases are anything to go by, you will have problems eventually.

    Better you know ahead of time.

    This is coming from someone with a Samsung 830 bought back when I was trumpeting Samsung as the only company to buy a SSD from.

    Look how far they’ve fallen. Unreliable.

      • Jeff Kampman
      • 4 years ago

      It’s way premature to say this drive has a “problem.”

      • Mr Bill
      • 4 years ago

      The “Pro” versions never had a problem. None of the six I have have had problems.

        • Pizzapotamus
        • 4 years ago

        Unless you count drives bricking on a firmware update a problem. [url<]http://www.guru3d.com/news-story/after-samsung-840-evo-issues-now-the-850-pro-has-issues-too.html[/url<]

          • sustainednotburst
          • 4 years ago

          That issue was caused by how Magician 4.5 applied the update. Not the FW update itself. I did that FW update through ISO, no issues to date on my system.

            • Pizzapotamus
            • 4 years ago

            But given that it’s Samsung software and one of the options they gave for firmware updates, it’s certainly still an issue that occurred with one of their pro drives.

      • bennyg
      • 4 years ago

      Can’t speak for others but my 840 evo has not caused any problems or been unreliable.

      I only noticed the issue on a bench result of speed vs data age. It took 1 hour to refresh data before the firmware fix was released. 15 minutes to update firmware. Contrast an hour a year with how many *days* the 1tb mech hdd has spent defragging.

      What I’m far more concerned with was the hundreds of Mb of stealthily background uploaded data that Samsung Magician sends to some Samsung employee in Korea. It did not inform or ask permission to do this

      • Freon
      • 4 years ago

      It seems the 850 EVO has been free from issue. I’ve been very cautious and erring on the side of buying BX100 drives, but now that EVO has been out for quite a while (over a year?) I’m thinking we’d know by now.

      My only concern over the Samsungs has been the TLC NAND anyway, and this is their “Pro tier” MLC drive so I don’t think there is much reason for concern.

      The whole Algolia thing turned out to be a bug in Linux, not any fault of Samsung’s.

      • Jigar
      • 4 years ago

      I can confirm 850 EVO is issue free

    • DPete27
    • 4 years ago

    This article shouldn’t have been published until the random write issue was confirmed to be accurate or not. The initial review carries much more weight than an updated “oops we messed up” article (if that’s what ends up happening)

      • Jeff Kampman
      • 4 years ago

      We’re still waiting on an official statement from Samsung, but our results are consistent with those that other reviewers are seeing. Given that, it’s probably safe to say that the sustained random write and performance scaling results are just how the drive performs.

    • DrDominodog51
    • 4 years ago

    What versions of GIMP and Libreoffice are you using?

      • weaktoss
      • 4 years ago

      LibreOffice 4.3.2
      GIMP 2.8.14

      These are listed on the “Test methods” section (always the second-to-last page).

        • DrDominodog51
        • 4 years ago

        Oh. Thanks. I only had time to scan the article and for some reason I thought the test methods are on the second page.

    • Chrispy_
    • 4 years ago

    Regarding the test rig, you guys are using an adapter card – Is that because onboard M.2 slots only work in SATA/AHCI mode?

    I haven’t jumped on the NVMe bandwagon yet, mainly because I’m waiting for it to become a default standard (which I’m sure it will) and not require stupid adapters – that, and that for low queue depth workloads used by most client PCs, even SATA has loads of headroom to spare. It’s still practically impossible to get QD1 4K read rates higher than 55MB/s which makes SATA3’s 550MB/s bottleneck seem like a distant and irrelevant bottleneck for a consumer desktop/laptop.

      • Freon
      • 4 years ago

      I can’t recall seeing a single full ATX Z170 board while I was shopping that failed to have at least one m.2 4x slot. NVMe support is baked in AFAIK, maybe short of the manufacturer really screwing up the BIOS. It’s only when you look at the mini- and micro- boards that you may lose the native m.2 4x slot, or may have a shorter m.2 variant.

      Actual PCIe 4x cards and m.2 seem to be the easy upgrade standards for desktop. m.2 has an edge in that it is already very common for laptops.

      U2 seems the odd one out requiring some special adapters. U2 could become common, but for now it is not, and is fighting a bit uphill since desktop is its only home.

      • Jeff Kampman
      • 4 years ago

      Judging by Asus’ 10Gbps spec on the Z97-PRO’s M.2 slot, I don’t think it has the full complement of PCIe lanes we want for testing NVMe drives.

      • weaktoss
      • 4 years ago

      As Jeff said, it’s actually because our board only serves two lanes to the onboard M.2 slot, and the drive needs four to hit its peak speeds. A lot of M.2 slots on Z97 boards are like this. The situation is a bit better on Z170 boards, but I’d definitely recommend hunting through the specs to find out what exactly you’re getting before buying anything.

        • Chrispy_
        • 4 years ago

        Of course, I’d forgotten that board makers weren’t obliged to link all four lanes to the slot – presumably a side effect of trying to divide the Z97’s 18 Flex lanes between SATA3, USB3, PCIe3 and other things ending in 3 that I can’t think of right now 😉

        I’d certainly be happy with “only 10Gb/s” SSD performance compared to my SATA drives but yeah – you guys can’t afford to hamper your test rig. I’m guessing real-world speeds of this thing in a “typical” x2 slot would be useful to add to the review, but I think we can work it out; it’s going to look exactly the same but all the graphs that exceed 1.6GB/s transfer rates are going to get clipped to 1.6GB/s, right?

    • chuckula
    • 4 years ago

    Hi Tony,

    Great review. [s<]Did you notice any potential thermal issues with these drives under load?[/s<] Ah, I see potential issues in the synthetics that might be thermally related. Lemme rephrase that: Do you have any ability to monitor the actual drive temperature during operation to diagnose situations where there is the potential for thermal throttling?

    • Freon
    • 4 years ago

    Very interesting stuff. I’m now watching the m.2 space closely and hope to pick up one for boot duty and to displace a BX100 in maybe the next 12 months. This sort of m.2 4x NVMe drive seems to be the right type of product. It’s just a matter of seeing the one that hits all the right notes in terms of performance, price, and features.

    Glad to see the regular slew of drives’ data for comparison purposes.

    • chrcoluk
    • 4 years ago

    is a shame intel refused to provide bios update code for z87 boards (which are almost the same as z97 but no support for nvme). According to asus intel is to blame as they saying to people asking why no nvme support on z87.

    Since my 850 pro at sata gen 3 is blazing fast already, I would never upgrade my board just for nvme, but I would buy an nvme drive if my board was compatible.

    So for this reason I think techreport claiming almost any modern intel board has support is somewhat a lie, unless you dont consider haswell architecture modern. Remembering the man on the street isnt blessed with freebie goodies when new models are released, instead people have to spend their hard earned cash and will rightly wait for the cost to be justified and hence not change their chipset on an annual basis.

    • TruthSerum
    • 4 years ago

    Now the only question is can I get it to boot drive status in UEFI…

      • Forge
      • 4 years ago

      Not a problem on any decent UEFI. SecureBoot is a little trickier, but can be done as well.

      • Bauxite
      • 4 years ago

      Z97, X99 and newer: yes with recent firmware

      X79: *maybe* someday with a couple models (asus hinted at it)

      Anything else: nope

        • Freon
        • 4 years ago

        I would assume pretty much any Z170 (maybe even any 100-series that has the slot) should be good to go, no?

          • Bauxite
          • 4 years ago

          “and newer”

        • Krogoth
        • 4 years ago

        Any UEFI compliant motherboard would support it as a boot device. That would be anything Post-SB and 6 series chipset.

        As for full NVMe support, you are limited to 9 and 10 series motherboards.

          • Bauxite
          • 4 years ago

          nope.jpg

            • Krogoth
            • 4 years ago

            You are confusing UEFI booting with NVM Express support.

            UEFI booting and Secure Boot have been working since Sandy Bridge days. It would require a simple update to get NVMe devices to boot on older boards.

            9 and 10 series boards just have out of the box support at this time.

            • Ewitte12
            • 4 years ago

            Good thing my z87 board suports it 😉 after injecting the modules from the z97 version at least.

            • Krogoth
            • 4 years ago

            Exactly, it is purely a software limitation not a hardware limitation. 😉

            Vendors are going to try to upsell their newer models when older stuff can technically work without any issues.

    • emorgoch
    • 4 years ago

    Looks like you’ve got the wrong drive highlighted for the “TR RoboBench – Read Speed – 8T – Media” test. The Samsung XP941 is highlighted instead of the 950 Pro.

      • Forge
      • 4 years ago

      And then the highlights are all over the place on the Boot time page. Reported, likely fixed shortly.

        • weaktoss
        • 4 years ago

        Whoops! RoboBench graph fixed, boot graph updates are on the way.

    • thill9
    • 4 years ago

    Hmm, people actually do use the retail Intel coolers.

    • Krogoth
    • 4 years ago

    Server and workstation-tier SSD performance on the cheap.

    It is overkill for 99% of the non-server/workstation market.

      • chuckula
      • 4 years ago

      Krogoth is not impressed.

      WITH YOUR PUNY WORKLOADS!

      • jessterman21
      • 4 years ago

      Yep, everyone else, get yourself a nice BX100…

      • derFunkenstein
      • 4 years ago

      I dunno, man. Robobench seems like the sort of thing people would actually do, and the 950 Pro shows a big difference. The last test with “copy” (from one place to another on a drive) in particular.

        • Krogoth
        • 4 years ago

        Not under a non-workstation and non-server workload.

        Any decent 2.5″ SATA SSD works just about as well for less. The reason? You are CPU-bound (single-threaded performance)

          • Meadows
          • 4 years ago

          Robocopy is multi-threaded, you tourist.

        • Meadows
        • 4 years ago

        We actually use Robocopy at work on a daily basis, but have yet to upgrade to SSDs at all. Some of our inhouse software can be under continuous development and it is policy not to have a copy on one’s computer but always pull the latest build from the intranet server at HQ.

      • BIF
      • 4 years ago

      I see you are not impressed.

      Neither am I. This is UNDERKILL.

      512 GB? Seriously, why are drive manufacturers putting out such apologetic, anemic capacities?

      Get with it. 1TB should at least be an option.

      • jihadjoe
      • 4 years ago

      But mediocre when given an actual server/workstation workload, though. Look how hard this gets pwned by the Intel 750 in database IOPs, never mind the DC-P3700.

      Overkill for desktop, crap for enterprise. A solution in search of a problem, really.

    • PrincipalSkinner
    • 4 years ago

    Crucial MX200 FTW.

    • southrncomfortjm
    • 4 years ago

    None of this makes any sense without those world famous scatter plots.

      • Freon
      • 4 years ago

      I’d say it makes sense without them, but I do kind of like the sort of 1000 ft. view gut check as part of the conclusion.

Pin It on Pinterest

Share This