Samsung 950 Pro PCIe SSD: Not Performing as Advertised

Discussion in 'Hardware Components and Aftermarket Upgrades' started by moviemarketing, Oct 22, 2015.

Thread Status:
Not open for further replies.
  1. moviemarketing

    moviemarketing Milk Drinker

    Reputations:
    1,036
    Messages:
    4,256
    Likes Received:
    881
    Trophy Points:
    181
    Had high hopes for the Samsung 950 Pro PCIe SSD after reading the advertised transfer rates:

    [​IMG]

    Unfortunately it seems to fall rather short of these expectations in 4KB Random write:

    [​IMG]

    Hopefully this is something that might be fixed with firmware update. At this point, however, seems better to wait a bit longer for future PCIe SSD models from Mushkin or other manufacturers.
     
  2. John Ratsey

    John Ratsey Moderately inquisitive Super Moderator

    Reputations:
    7,162
    Messages:
    28,880
    Likes Received:
    1,950
    Trophy Points:
    581
    For me, the interesting part of the first table is how the power consumption of the PCIe x 4 interface drives compares to the SATA. The extra potential speed comes at a price and it will be interesting to see if any of the notebook manufacturers make specific provision for dissipating the extra heat.

    John
     
    TomJGX and moviemarketing like this.
  3. jaybee83

    jaybee83 Biotech-Doc

    Reputations:
    4,095
    Messages:
    11,564
    Likes Received:
    9,125
    Trophy Points:
    931
    maybe that new powersaving algorithm they implemented in the 950 pro in order to prevent the heat throttling seen in the sm951 is causing the lower performance values...lets call it artificial power throttling to avoid thermal throttling *lol*

    Sent from my Nexus 5 using Tapatalk
     
    TomJGX likes this.
  4. tilleroftheearth

    tilleroftheearth Wisdom listens quietly...

    Reputations:
    4,930
    Messages:
    12,318
    Likes Received:
    2,339
    Trophy Points:
    631
    The truth about these M.2 drives is that NVMe wasn't created for them...

    It was created for the X-Point Intel technology that will begin showing up in the next few months (if we're lucky).

    NVMe on anything other than an Intel drive today is just a check box for the spec-hungry, nothing else. Issues with lower than SATA 'scores', out of whack power requirements and built in throttling issues (to not only the SSD but to the rest of the components in a mobile system too...) are the signs that these in-between beta's should continue to be ignored by most users. And almost all mobile users, period.

    The final nail in the coffin for these useless examples of storage subsystem components in late 2015 is the capacities they are being offered in. 256GB and 512GB (nominal) capacity... and, more expensive too?

    Anyone can see that these are toys for people with more $$$ than brains.

    Having the latest does not always mean having the greatest when there is a new 'standard' to help confuse people with.

    If we were offered a 2TB+ M.2 SSD w/NVMe that cost less (per GB), stayed cooler, sipped less power and was as performance oriented as the SanDisk Extreme Pro 960GB (in actual, real world usage) then we could look forward to these drives - as the NVMe spec's could almost guarantee an increase in responsiveness from that gold standard of a workhorse SSD.

    But as it is, we are given the 950 Pro and all it can do is excel (only in the 256GB capacity) at short/bursty synthetic 'scores' that mean squat in actual workloads.

    Again; we've seen this all before - OCZ + SF based duracrap that limited the performance to preserve the integrity of the nand - at the expense of the promised performance that was never to be reached even once... Yawn. Been there, done that.

    Now, I'm waiting for the real reason NVMe was created.

    Come on Optane, where are you?
     
    Last edited: Oct 24, 2015
    Spartan and jaybee83 like this.
  5. superparamagnetic

    superparamagnetic Notebook Consultant

    Reputations:
    402
    Messages:
    252
    Likes Received:
    56
    Trophy Points:
    41
    The difference between advertised and measured speed is due to queue depth and differences in testing methodology. The advertised speed is at QD32, while the measured speeds are at QD1,2 and 4.

    Tomshardware has some nice plots of transfer speed vs QD
    http://www.tomshardware.com/reviews/samsung-950-pro-ssd,4313-3.html

    You can see that the 950 Pros hit close to their rated random writes once you go about QD2. Judging by the plots it seems like Anandtech's data is heavily weighted towards QD 1 (240 MBs = 60K IOPS).

    You'll also notice that all of Anandtech's numbers are quite a bit lower than Tomshardware too, which indicates differences in methodology.
     
  6. tilleroftheearth

    tilleroftheearth Wisdom listens quietly...

    Reputations:
    4,930
    Messages:
    12,318
    Likes Received:
    2,339
    Trophy Points:
    631
    We can analyze and agree on why there is a difference between advertised and the many ways of measuring the speeds... but that is partly my point. The advertised speeds don't matter on a workstation workload at a QD higher than 4 or 8, at the most. And matter less when comparing benchmark utility 'scores' with no correlation to the real world workout a storage subsystem gets.

    Real world workflows rarely mimic the synthetic (and simple for an SSD's controller to adapt to...) utilities used to test them 'like the manufacturers do'.

    Just like swimming is the best exercise to become better at swimming; testing a component in the actual workload it is supposed to give superior results in should be the standard testing method. Not a simple third party utility that can be programmed into the firmware to be 'told' what scores to report. Sigh...

    With SATA being a half duplex communication protocol and NVMe being full duplex, that should give us much improved (real world) performance in and of itself (more than 2x in real world usage). There is no NVMe solution (except maybe Intel's 750 series which for mobile use is not exactly a good match) that approaches that goal. Another indication of the failure of M.2 NVMe drives to deliver what they theoretically should be able to at this time.

    All in all, any non-Intel SSD w/NVMe capabilities is essentially a proprietary and one off example of that standard. That alone is enough for me to disregard it for my systems today. Couple that fact with the performance deficit, increased power and non-existent (in any off the shelf notebook) cooling requirements and the ridiculously low capacities being offered today... and you can see that not even the manufacturers believe in M.2 NVMe half as much as consumers want to...

     
  7. Peon

    Peon Notebook Virtuoso

    Reputations:
    406
    Messages:
    2,004
    Likes Received:
    128
    Trophy Points:
    81
    The Intel 750 suffers from the same bathtub curve as every other SSD out there when it comes to mixed read/write scenarios. As far as I'm concerned, it's just as big of a failure as every other NVMe SSD out there today.
     
  8. tilleroftheearth

    tilleroftheearth Wisdom listens quietly...

    Reputations:
    4,930
    Messages:
    12,318
    Likes Received:
    2,339
    Trophy Points:
    631
    No doubt. But as the initiator/designer of NVMe, Intel is the only one to really know it in depth and can use that to build the best NVMe hybrid (short of X-Point) drive today.

    Bottom line, in real world usage, Intel's 750 is still the most desirable option, if the platform can support it properly (basically; a desktop).


     
    TomJGX likes this.
  9. Palorim12

    Palorim12 Notebook Enthusiast

    Reputations:
    0
    Messages:
    27
    Likes Received:
    4
    Trophy Points:
    6
    Intel made it...but Samsung was the first to implement NVME in drives with the Samsung XS1715.

    The Intel 750 for the longest had terrible boot times with NVME. Don't you remember your best bud Maximus666, wait, he's Matrix Leader here, complaining about how the 750 is the worst drive in existence because of that issue?

    Yes, they patched it recently, but according to reviews the 950 Pro beats it out of the water when it comes to booting.

    Also, those worrying about Heat on the 950 Pro:

    http://www.pcper.com/reviews/Storag...-PCIe-SSD-Review/Thermal-Throttling-Conclusio

    "When Samsung announced the 950 PRO, several of you commented on potential thermal throttling due to heat generated in such a small package during heavy use. The following image represents a worst case scenario, with the 950 PRO being sequentially written with zero airflow across it:
    [​IMG]
    As you can see, you would have to write nearly 150GB at over 1.5GB/sec to get a 950 PRO to warm up enough to throttle, and when it does, the throttling is very minor, dropping to only 1.2GB/sec intermittently. The slightest airflow prevents this from happening at all, and even if there was zero airflow, the chances of maxing a 950 PRO out on writes for that long of a burst is extremely unlikely in even the most demanding consumer usage scenario."

    Basically, give it adequate air flow, and no throttling.
     
    TomJGX and jaybee83 like this.
  10. jaybee83

    jaybee83 Biotech-Doc

    Reputations:
    4,095
    Messages:
    11,564
    Likes Received:
    9,125
    Trophy Points:
    931
    thats what i was expecting, just by looking at the raw power consumption numbers of the 950 pro compared with the sm951 :) glad to see it in hard data as well!
     
Loading...
Thread Status:
Not open for further replies.

Share This Page