Review Index:

Samsung 960 EVO 250GB and 1TB M.2 NVMe SSD Full Review - TurboWrite Squared

Subject: Storage
Manufacturer: Samsung

Introduction, Specifications and Packaging


Since Samsung’s announcement of the 960 Series SSDs, I have been patiently waiting not for the 960 PRO (reviewed a few weeks back), but for the 960 EVO. It is the EVO, in my opinion, that is the big release here. Sure, it doesn’t have the quad Hexadecimal Die Packages, Package-on-Package DRAM and ultimate higher capacity of the PRO, but what it *does* potentially have is class leading performance / price in the M.2 form factor. Just as we all wanted lower cost SSDs in the 2.5” SATA form factor, M.2 is seeing greater adoption across laptops and desktop motherboards, and it’s high time we started seeing M.2 SSDs come down in price.

I know, don’t tell me, the Intel 600p carries a SATA-level cost/GB in an M.2 form factor. Sure that’s great, and while I do recommend that SSD for those on a budget, its caching scheme comes with some particularly nasty inconsistencies in sustained writes that may scare off some power users. Samsung 840/850 EVO SSDs have historically handled the transitions between SLC cache and TLC bulk writes far better than any competing units, and I’ve eagerly anticipated the chance to see how well their implementation carries over to an NVMe SSD. Fortunately for us, that day is today:

View Full Size


View Full Size

An important point to note in the performance specs - the lowest capacity model is the only one to see its performance significantly taper in stated specifications. That is because even with its 48-layer VNAND operating in SLC mode, there are only two packages on all 960 EVOs and the 250GB capacity comes equipped with the fewest dies to spread the work across. Less parallelism leads to lower ultimate performance. Still, it is impressive to see only 250GB of flash reaching near saturation of PCIe 3.0 x4 in reads.

I've appended the 'sustained' (TLC) performance specs at the bottom of the above chart. These 'after TurboWrite' figures are the expected performance after the SLC cache has been depleted. This is nearly impossible in actual usage scenarios, as it is extremely difficult for any typical (or even power user) desktop workloads to write fast and long enough to deplete such a cache, especially considering how much larger these caches are compared to prior models.


View Full Size

Samsung has carried forward their simple packaging introduced with the 960 PRO. The felt pad on the bottom of the installation guide is both functional and elegant, keeping the 960 Pro safely in place during shipment.

Read on for the full review of the 250GB and 1TB Samsung 960 EVO!

Video News

November 15, 2016 | 10:49 AM - Posted by Chaitanya (not verified)

That 1Tb looks like perfect drive for most users. Only downside is the 3 year warranty while 850Pro has 10 years at similar price point.

November 15, 2016 | 11:16 AM - Posted by Buyers

Article says 5 Year warranty.
"Warranty period is an outstanding 5 years rated at..."

November 15, 2016 | 02:33 PM - Posted by Anonymous (not verified)

5 years is hardly an outstanding warranty. You can even get HDDs with that

November 15, 2016 | 04:14 PM - Posted by John H (not verified)

Compared to it's peers:

Intel 600p/750 - 5 year warranty
Samsung PM961/951 - 3 year warranty
Toshiba RD400 - 5 year warranty

"Outstanding" may be a bit of a strong word for a 5 year warranty. Definitely a good warranty though, plenty of drives out there with 1-3 year warrantys, albeit at lower prices.. (i.e WD Red NAS drives.. 3 year, or SATA SSDs 1-3 yrs).

November 15, 2016 | 07:15 PM - Posted by fuersliph

Samsung's website says 3 years 100TBW, its under more specs. Not sure which is correct.

November 15, 2016 | 10:37 PM - Posted by Chaitanya Shukla

Samsung's website, press release, all the sellers and even the specifications on 1st page of review says its 3 year warranty. Article seems to have made a goof-up, Only 960Pro has 5 year warranty of the two 960 series drives.

November 15, 2016 | 10:59 PM - Posted by Allyn Malventano

Yup, I misread the slide I was referencing for the write-up. Fixed, and thanks!

November 15, 2016 | 11:29 AM - Posted by ole (not verified)

man Allyn you write some super detailed review and explain a lot into the smallest details. its been awhile since i have read so much. loved the trim portion and your test mythology average qd test.

November 15, 2016 | 11:17 PM - Posted by Allyn Malventano

Thanks! We've certainly put a lot of work into the new testing!

November 15, 2016 | 11:41 AM - Posted by Paul A. Mitchell (not verified)

Is the 960 Pro 2TB (grey line) approaching max sequential, given the 32 Gbps ceiling and allowing for controller overhead? Sequential READs at QD=16, 32 and 64 look very flat (grey line). Once again, very nice review. Your graphs are very sharp and easy to read.

November 15, 2016 | 11:18 PM - Posted by Allyn Malventano

Yup. It's pretty much saturating PCIe 3.0 x4 at those QD's.

November 15, 2016 | 12:18 PM - Posted by Patrik (not verified)


could you maybe start measuring power consumption of laptop with specific SSD, in some scenarios (video playback, general browsing, tranfering files, etc.).
Something like 850 EVO M.2 or 2,5″ could be the base one (it has quite low/normal notebook power consumption).

I'm mainly interested if the battery life of notebook would be affected significantly by adding something like 960 EVO or PRO vs. mentioned 850 EVO...

November 15, 2016 | 11:21 PM - Posted by Allyn Malventano

I've been working on a way to *properly* test this. The issue is that desktop motherboards are not a sure bet as far as support for the same lower power states implemented by laptop hardware. You guys know me - I won't publish results on something unless I'm sure it is being tested the right way.

November 15, 2016 | 12:51 PM - Posted by PCPerFan (not verified)

This drive fared much worse in the mixed workload tests by other reviewers (rhymes with mom's).

November 15, 2016 | 11:01 PM - Posted by Allyn Malventano

Other reviewers test with a saturated random workload, and do so with 4k random across the entire span of the SSD. For caching SSDs, this ends up saturating the SLC, meaning all writes go to TLC. This is (IMO) beyond unrealistic, which is part of the reasoning behind the need for this new testing.

November 16, 2016 | 01:19 PM - Posted by Penteract (not verified)

And this attention to detail and willingness to work very hard created information useful the the majority of your readers is what makes you my #1 source for SSD analysis. Thank you Allyn!

November 16, 2016 | 01:51 PM - Posted by Jann5s


November 15, 2016 | 12:56 PM - Posted by Anonymous (not verified)

great article.
the massive downgrade in warranty is just too great for me to move me off 850 pros.

allyn, could you see samsung releasing a 10 year warranty consumer product again [with these 960's or other] or is that an industry once off?

November 15, 2016 | 11:02 PM - Posted by Allyn Malventano

Most likely not. 10 years is just a long time for a warranty.

November 15, 2016 | 01:06 PM - Posted by Lemonsquare

Interesting... the 250 GB EVO received an abysmal review score at Tom's. The 1TB was excellent, though.

November 15, 2016 | 04:36 PM - Posted by John H (not verified)

There's quite a few tests it did worse on THG; mixed sequential and random, PCMark looks pretty bad, and their disk service time is also bottom of the pack. Interesting differences..

November 15, 2016 | 11:04 PM - Posted by Allyn Malventano

Yeah, I just don't see a *realistic* mixed workload exceeding 300 MB/s on writes, not on an actual system being used by an actual person, at least.

(300 MB/s is what would be needed to saturate the SLC on the 250GB model).

November 15, 2016 | 01:29 PM - Posted by Anonymous (not verified)

Still no word from Samsung regarding edrive support for either the 950 PRO or 960 PRO?

November 15, 2016 | 11:05 PM - Posted by Allyn Malventano

The story remains the same as with the 960 PRO, unfortunately.

November 23, 2016 | 02:10 AM - Posted by DiamondMines

Hi Allyn,

I was wondering when the 960 EVO M.2 will be out? I can't find it anywhere. Have a good day sir.

November 15, 2016 | 02:30 PM - Posted by Jordan Viray (not verified)

Great review!

The behavior of your disk trace in live systems matches my findings as well, i.e., most everything is QD1. That's why QD1 performance has been my main metric for SSD performance for some time now. It's strange that so many enthusiast site reviews give so much weight to benchmarks and unrealistic performance figures more relevant to server loads.

November 15, 2016 | 05:20 PM - Posted by bburnham37 (not verified)

Allyn, you mention that the 256GB version of the 960 EVO uses 48 layer v-nand like the other capacities. Other tech review sites have stated that the 256GB version used 32 layer v-nand, can you confirm which is correct?

November 15, 2016 | 11:15 PM - Posted by Allyn Malventano

Assuming you mean the 250GB version. It is not 32-layer, but it is a smaller capacity die. The 250GB model uses 128Gbit dies, while the 500GB and 1TB both use 256Gbit dies. They do this to help maintain parallelism (therefore performance) on the 250GB model.

Where did you see mention of 32-layer?

November 16, 2016 | 02:06 PM - Posted by bburnham37 (not verified)

It was in the Tom's article.

"The 960 EVO will appear first in three capacities that range from 250GB to 1TB. The initial release ships with 48-layer V-NAND using 256Gbit (32GB) die for the two largest capacities. The 250GB model uses Samsung's older 32-layer flash with 128Gbit die (16GB). We expect the series to grow when Samsung releases 64-layer V-NAND with 512Gbit (64GB) die in 2017.",4802...

And yes, I meant the 250GB version, not 256.


November 17, 2016 | 01:59 PM - Posted by Allyn Malventano

I'm awaiting final confirmation, but all replies I've received so far indicate Toms is incorrect.

November 15, 2016 | 07:10 PM - Posted by Milko Tsazov (not verified)

Very nice review as always mr. Malventano. I follow your storage reviews, but I have a little complaint about them - they are great, but I'm colorblind and it is hard for me (and probably others) to understand the graphs sometimes unless I really take a closer look with zoom in my browser and staring at the graphs for a few minutes. Please put a metric values in graphs if it is possible in your reviews or something else , because it is really hard to read the graphs (at least for me). Thank you and keep up the good work.

November 15, 2016 | 11:10 PM - Posted by Allyn Malventano

That was part of the reasoning behind the comparison bar charts, as they provide useful data (by distilling down via QD weighting). Unfortunately, the line charts must still be used in some places and they make it nearly impossible to have a useful number of comparison products (8) while also maintaining easy viewing without using color as a differentiator. It's tough to tell some of them even *with* the use of color. I'm totally open to input on this, if you know any tricks I can try that will help you.

November 16, 2016 | 02:02 PM - Posted by Jann5s

you can simulate colorblindness:

don't know if they are good approximations but the site seems respectable

November 16, 2016 | 01:48 AM - Posted by Anonymous (not verified)

Release date?

November 16, 2016 | 06:28 AM - Posted by Christopher (not verified)

Just lately I have seen friend's 2TBW 256GB 850 EVO "Stuck" on 50MBps writes - TLC is not an option.

November 16, 2016 | 11:02 AM - Posted by Allyn Malventano

What you are describing certainly sounds like a failure and is not representative of typical SSD performance. Your bud should RMA that drive. 

November 16, 2016 | 03:21 PM - Posted by Christopher (not verified)

Sadly, it seems that you've not verified how much of that is happening in the Internet, and most of those is related to TLC versions of those drives. This is something that I would not expect from 9 months of use of SSD.

November 17, 2016 | 02:14 PM - Posted by Allyn Malventano

If it is the epidemic you describe, then all of those impacted should RMA their drives. You're also the first I've seen reporting this specific issue in comments of any of our SSD articles. We do track these sorts of things.

November 16, 2016 | 08:36 AM - Posted by Anonymous (not verified)

Allyn, excellent testing and explanation as usual!
Any plans on test the MyDigitalSSD BPX line of M.2's?

November 16, 2016 | 09:35 AM - Posted by AMv8(1day) (not verified)

I second this. I'd really like the BPX get run through your gauntlet and see how it fares with sustained writes. I'm looking for a good, reasonably priced M.2 SSD for proxy caching duties and the 600p looks to be a bad option.

November 16, 2016 | 10:14 AM - Posted by Qrash

I third this request. The performance/price results at THG look good, but I would be more comfortable with Allyn's testing. Also, it looks like MyDigitalSSD doesn't have a driver for the BPX, so it runs on the Microsoft driver.

November 16, 2016 | 04:02 PM - Posted by Allyn Malventano

We're going to look at these as soon as we can get them in for testing!

March 9, 2017 | 02:46 PM - Posted by Anonymous (not verified)

Did you ever put this through your testing? It sounded like it from your comment on the last podcast when Josh had the MyDigitalSSD as his pick.
Also, do you publish the performance charts somewhere as you test more drives? I haven't seen any updated ones since this review.

November 16, 2016 | 08:31 PM - Posted by rock1m1

I have a Z87 MB which does not have NVMe. Can I use a PCIe add on card for this? Also I heard you can't boot directly to these cards when running in this fashion?

Thank you

November 17, 2016 | 02:16 PM - Posted by Allyn Malventano

Boot support requires BIOS support, which varies on the older boards. Best bet is to look for confirmation of support for the specific board you intend to use.

November 16, 2016 | 11:35 PM - Posted by Anonymous (not verified)

what i want to know would the 960 evo be a good first m.2 nvme drive or should i get the 950 pro

November 17, 2016 | 02:17 PM - Posted by Allyn Malventano

At this point the 960 EVO is a better deal for the capacity / performance, but it does have a 3-year warranty (950 PRO has 5-year).

January 30, 2017 | 09:18 PM - Posted by CubanLegend (not verified)

What I'd also like to know is, I've got a 960EVO 250GB on backorder from B&H but I have an Intell600p 256GB in hand. Should I wait for the 960EVO to build my system or will the performance difference in real world gaming/OS usage not matter and I should build my system now with the 600p as the boot drive?

November 16, 2016 | 11:43 PM - Posted by Anonymostly (not verified)

Really looking forward to seeing those X25-M results! :-D

November 18, 2016 | 06:10 PM - Posted by Anonymous (not verified)

Cannot wait to get my hands on my 1TB 960 EVO to replace my 850 pro. Sadly still stuck in pre-order purgatory with no idea when they'll actually ship.

Thanks for the great detail in the review and testing!

November 20, 2016 | 11:19 AM - Posted by vailr

B&H Photo shows a Dec. 8, 2016 shipping date for the 250GB 960 EVO.

November 19, 2016 | 09:59 PM - Posted by HERETIC (not verified)

Hi Allyn,
Would it be pedantic of me to ask for Trim speed charts
to be compiled best to worse?

Here's a classic case of how useless old synthetic benchmarks
can be-
Keep up the great work.

November 19, 2016 | 11:00 PM - Posted by Ladybug Crystal (not verified)

Allyn, are you going to test the 500GB version as well?
I really want to know its TLC speed when the cache (~26GB, right?) runs out.
I think 250GB is not quite enough and its normal TLC speed is a bit too slow for me, while I'm also not planning on dropping ~$500 on the 1TB version.

November 20, 2016 | 02:02 AM - Posted by Dark_wizzie

Hmm, maybe add 850 evo raid into the chart? Obviously testing every single drive in raid takes way too long, but raid 850 evo or raid of a budget drive seems like an interesting data point that doesn't take way too long.

November 22, 2016 | 09:00 PM - Posted by Chris C (not verified)

As tempting as it may be I cannot find myself ever buying a TLC drive for anything other than a scratch drive, cache drive etc.

November 28, 2016 | 03:13 PM - Posted by Clint.Merrell

Why? Because of the lifespan of TLC drives?

November 23, 2016 | 12:36 PM - Posted by Soylent

As an AMD owner, I'd love to see benchmarks of this on a motherboard with a PCIe 2.0 4x m.2 connector (like the Gigabyte 990fx-gamer). Does it completely saturate the 20Gb/s bandwidth?

November 24, 2016 | 08:45 AM - Posted by JoshD (not verified)

Wow, poor showing in the Intel 600p. I feel bad for those who upgraded to the Intel drive on the assumption NVME would be a big boost over a SATA6GB SSD.

The Client QD weighted chart makes it look like it might actually be worth upgrading from a SATA to NVME SSD for real world performance.

I'm curious where the OEM LiteON SSD in my Dell laptop would fit into the mix.

December 22, 2016 | 05:29 AM - Posted by Flisker (not verified)

Awesome review. Really like the detailed info about the topic. I would love to ask one thing about these drives. Will I gain any performance boost by moving from X79 (DMI 1) to Z170 (DMI 3) ?

Thanks ;)

January 6, 2017 | 07:24 AM - Posted by Anonymous (not verified)

question about endurance or TBW ssd 960 evo 250gb.
it's said 100 TBW but is it true
for for example i have ssd mx300 and according to crucial software
and other software like crstal disk the total write is 500000gb and life of the ssd around 15%, in the website of crucial the TBW of the ssd 160 TBW so it clean way above that.
i am interesting in buying the SSD 960 evo 250GB SO MY QUESTION IS :
WHAT IS THE REAL TWB OF THE SSD 960 EVO or the max write of the ssd until it die ?

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

This question is for testing whether you are a human visitor and to prevent automated spam submissions.