Samsung SSD 950 PRO Solid State Drive Review


<< PREVIOUS            NEXT >>

AIDA64 Disk Benchmark

Many enthusiasts are familiar with the Finalwire AIDA64 benchmark suite (formerly Lavalys EVEREST), but very few are aware of the Disk Benchmark tool available inside the program. The AIDA64 Disk Benchmark performs linear sequential tests on each drive, and does not require a partition to be present for testing so all of our benchmarks are completed prior to drive formatting.

Linear sequential disk tests scan from the first physical sector to the last, and could be considered the best measure of storage performance. A side effect of many linear write-performance test tools is that the data is erased as it writes to every sector on the drive. Normally this isn’t an issue, but it has been shown that partition table alignment will occasionally play a role in overall SSD performance (HDDs don’t suffer this problem).


AIDA64 Sequential Reads: Intel X99 Platform


AIDA64 Sequential Reads: Intel Z170 Platform

Lavalys AIDA64 Disk Benchmark utilized a 1MB block size to test the 512GB Samsung SSD 950 PRO. As illustrated in results above, read performance with 950 PRO on the X99 test system measured average speeds of 2310 MB/s while the Z170 platform produced 2166 MB/s.

Tests on the X99 system delivered consistent flat-line wave form results, allowing for an average read speed virtually identical to this drive’s maximum peak speed across the full capacity. Z170 results delivered rather inconsistent sequential performance, with the average and maximum speeds separated by 110 MB/s.

AIDA64 continuous sequential write-to tests were next…


AIDA64 Sequential Writes: Intel X99 Platform (DISKPART Clean)


AIDA64 Sequential Writes: Intel X99 Platform (Magician Secure Erase)


AIDA64 Sequential Writes: Intel Z170 Platform

The waveform charts above illustrate buffered data writes on 950 PRO, indicating linear write performance speeds saturated the transfer cache. The 512GB Samsung SSD 950 PRO solid state drive recorded an average linear write-to speed of 916 MB/s on X99 using DISKPART clean -all NAND prep and 978 MB/s using Magician Secure Erase. This is a remarkable deviation, as maximum performance reached 1492 MB/s for a difference of 576 MB/s.

Test performance on the Z170 system appeared less volatile as the write speeds averaged 1349 MB/s while maximum speed reached 1507 MB/s, for a difference of 158 MB/s.

In the next section we use PCMark Vantage to test real-world SSD performance…


<< PREVIOUS            NEXT >>


Skip to comment form

  1. Daniele Cosentino

    Hi. Why u used the adapter for the x99 deluxe and not the integrated M.2 Socket?
    I buught one of those but i couldn’t make it works on my x99 deluxe.

    It is not compatible with the M.2 X4 socket?

    1. Olin Coles

      Looking at their website, it doesn’t appear that the ASUS X99-DELUXE has an integrated M.2 socket. See for yourself: https://www.asus.com/us/Motherboards/X99DELUXE/

      1. Eric Anderson

        The x99 Deluxe very much does have an integrated m.2 socket, and it even ships with a secondary extended pci-e x4 m.2 bracket.

        I’m having the same issue of it not being recognized in the BIOS, but the device seems to be recognized in Windows (running Windows 10).

        1. mangel

          Did you ever get the 950 pro to appear in your bios. I have the similar problem with the X99 Deluxe (1st edition).. Also nothing appears under NVME in the bios. Does boot into windows 10.

          The intel 750 series NVME does appear in both the boot section as well as the NVME section although the latter has no configurable settings.

  2. Chad

    Why does Crystalmark show such low 4K read speeds vs some of the other benchmarks? I have the 950 pro installed on a z170 Asus Hero VIII and I have that same issue. In fact, Samsung’s magician also reports about 208K with 4K IOPS read.

    1. David Ramsey

      That is an excellent question, and the only answer I have is “I dunno.”

      Hard disk drive performance is extremely repeatable: the platters have defined sectors, the drive may have a few megabytes of cache, but in general I can run various read and write tests all day long and see very little variance in the results.

      SSDs are something else entirely. They’re little “storage computers”, with multi-core processors, varying amounts of cache, and different strategies and internal protocols striving to balance performance with NAND durability. When I send a command to a hard disk, about all the controller does after a cache check is figure out which platter, track, and sector I want to deal with; when I send a command to an SSD, well, who knows what’s going on? It’s an evolving system and the various vendors don’t disclose the operational details for competitive reasons. Samsung doesn’t publish any details on the “UBX” controller they use on this drive, so all I can do is speculate.

      What this boils down to is:

      A. Running the same test multiple times can yield different results. Simple example: run a linear write test on a “clean” SSD, followed immediately by another run of the same test. The second iteration will show significantly reduced performance since the controller will have to erase each “sector” before writing to it the second time.

      B. Performance during a test can vary. Look at the AIDA64 linear read test on the Z170 and X99 platforms. Why are the results “smoother” on the X99? Again, who knows?

      Complicating the issue is the fact that this is our first NVMe SSD. Is this a factor? Will all NVMe SSDs show this relative to AHCI SSDs? Stay tuned! In the meantime all we can do is run the tests and report the results.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>