Internals, Testing Methodology and System Setup
Internals:
Cracking these open, I expected to see some sort of PCB changes as comapred to the 1TB and lower models of the 850 EVO and Pro. Let's see what has changed:
There have been some changes, but they are more minor than expected. The PCB is the same physical size as compared to prior high capacity models (more on that further down the page), however the layout has changed slightly. On the surface it looks like a simple revision to the PCB that was carried across both lines.
With the PCB's out, lets take a closer look at things here:
850 EVO 2TB:
Front and center we find Samsung's new MHX controller. This is a beefier version of the MEX (1TB) and MGX (120/250/500GB) controllers used in the smaller versions of the 850 EVO. As you can see, Samsung has had to step this controller up in two speed/capability grades as supported capacities have gotten larger.
Further around the controller we see necessary DRAM for the controller's use as well as four flash packages. With some decryption of the part numbers and math, these are 3-bit MLC (TLC) 16-die stacks.
At the rear we find four more pacakges of identical nomenclature, bringing the total count of TLC VNAND dies to 128 – a neat doubling of the 1TB model.
850 Pro 2TB:
On the 850 Pro we see the same layout as well as the same MHX controller (all smaller capacities of the Pro employ the MEX controller). We also see four flash packages with part numbers previously unseen. More on that after this pic of the rear:
An interesting note on the flash memory configurations of both the EVO and Pro is that previous models employed asymetric staggering of flash die counts per package within each series. The 840 EVO had staggered flash package capacities while the 840 Pro was symetrical. The (1TB and below) 850 Pro models had staggered capacities while the 850 EVO was symetrical. This flip was due to Samsung's VNAND die capacity being 50% away from an even power of two in MLC form (86 Gbit), while their older planar NAND used in the 840 series was 50% away from an even power of two in TLC mode.
If the above paragraph was confusing, it should be, and staggering die capacities to a connected SSD controller makes performance tuning potentially more complex.
- Possibility #1: With a total of 192 (!) VNAND dies in the 2TB 850 Pro, Samsung opted to spin a custom package type containing 24 86 Gbit dies. This way the die count per package falls between a power of two multiple, and the MGX controller can neatly address each package full of VNAND in an identical fashion.
- Possibility #2a: Going strictly off of the 'D' in the part numbers, which suggests the 2TB 850 Pro is still using 16 die stacks, that would suggest that Samsung has silently launched 128 Gbit capacity 32-layer VNAND dies.
- Possibility #2b: Still 16 die stacks, but Samsung might have increased capacity by instead moving to the same die footprint but a 48-layer VNAND stack.
As a bonus, partially because I got a bit too overzealous with the iFixit toolkit, here's the full capacity series of the 850 Pro:
The 850 EVO series is a bit different, mainly in the form of even smaller PCBs in the low capacity models, as only a single flash package of TLC VNAND is sufficient for the 120GB model.
Testing Methodology
Our tests are a mix of synthetic and real-world benchmarks. IOMeter, HDTach, HDTune, Yapt and our custom File Copy test round out the selection to cover just about all bases. If you have any questions about our tests just drop into the Storage Forum and we'll help you out!
Test System Setup
We have several storage testbeds. A newer ASUS P8Z77-V Pro/Thunderbolt and an ASUS Z87-PRO. Variance between both boards has been deemed negligible when testing SATA devices. Future PCIe and SATA device testing, including this review, take place on a new ASUS Sabertooth X99, which comes equipped with USB 3.1, M.2, and can also handle SFF-8639 devices with the proper adapter.
PC Perspective would like to thank Intel, ASUS, Corsair, Kingston, and EVGA for supplying some of the components of our test rigs.
| Hard Drive Test System Setup | |
| CPU | Intel Core i7 5820K @ 4.125 GHz |
| Motherboard | ASUS Sabertooth X99 |
| Memory | 16GB Micron DDR4 @ 3333 |
| Hard Drive | G.Skill 32GB SLC SSD |
| Sound Card | N/A |
| Video Card | EVGA GeForce GTX 750 |
| Video Drivers | GeForce Game Ready Driver 347.88 |
| Power Supply | Corsair CMPSU-650TX |
| DirectX Version | N/A |
| Operating System | Windows 8.1 Pro X64 (update) |
- PCPer File Copy Test
- HDTach
- HDTune
- IOMeter
- YAPT
At long last! $800 is about
At long last! $800 is about what I was hoping for (as a starting point anyway). We need more of these!
yeah cant wait to have all my
yeah cant wait to have all my storage behind my motherboard and not messing with my airflow. 🙂
While I get it, the price was
While I get it, the price was a deal breaker for me. It scales pretty much directly with their 1TB version. When the 1TBs came out, they were a noticeably better bang for the (GB/)buck.
Here’s hoping competition becomes more fierce and larger, because I ain’t touchin’ this at $800-1000. I have a feeling this may be very temporary price point.
Nice size. Just hope the EVO
Nice size. Just hope the EVO doesn't go slow on me like the 840 did. Firmware upgrade is just a temporary band aid. I'd like to see a rebate for a unit instead: Hand in your old 840 EVO with your new 2TB purchase, and get $100 off. In addition, perhaps a repeat review of the EVO unit will be necessary a year from now to draw a conclusion.
This is a different type of
This is a different type of flash entirely (VNAND). Also, it appeared Samsung delayed the 850 EVO launch right when the 840 EVO slow down issue was popular. I'd guess this was to make sure their new product wouldn't have the same issue.
Except its not a band-aid
Except its not a band-aid fix? Allyn has stated multiple times on Overclock that it actually fixes the issue with the 840 EVOs and the “band-aid fix” thing arised out of bad wording in the initial announcement of the new FW.
Great drive; no doubt. Here
Great drive; no doubt. Here are my main sticking points. One, with the latest kerfuffle about NANDs losing integrity if not powered up within a certain time is concerning (not that SSDs are usually used as long term storage without power). Two, these don’t address the elephant in the room NVMe. With NVMe poised to take the baton why invest in these? My next SSD will be NVMe not SATA.
For the NAND thing, don’t
For the NAND thing, don't believe the one blog about an SSD losing data in one week. That was more likely a defective SSD.
NVMe is great, but these will fit into the many laptops that are not NVMe capable.
True, they’ll fit into many
True, they’ll fit into many laptops, but most new laptops today have eschewed the 2.5 versions for the m.2. form factor as thinner laptops can be made, not to mention the massively increased performance on x4 slots. I think the writing is on the wall with the current SATA line. Intel has already demonstrated (as has Samsung with the 951) that SATA with its platter favoritism is too old and no longer is a viable solution for NAND based solutions.
Wonder how they look in RAID.
Wonder how they look in RAID. Keep 4TH mechanical drive as a backup and use 2 of these as data drive with SM951 as primary drive. That is expensive though.
Meant 4TB not 4TH
Meant 4TB not 4TH
I’ve always wondered, how
I’ve always wondered, how exactly does the SSD warranty work. Is it just for parts and labor or does the drive continence factor in, like for example if (you are an idiot and) you have your all your wedding photos on that drive without backup and the drive crashes before you can print them, OR say you are writing an educational text book and the drive crashes loosing all your “Revisions” notes, destroying hours and hours or days or weeks of hard work, is the drive manufacture liable for the monetary value of said photos? Or what if the drive crash causes surge damage to the mother board or other components (dont know if that is still a problem but it usta be)? Does that 5/10 year warranty make them liable for all the problems that would be created by a failing of that warranty, or is it simply “The physical drive will be replaced sorry bout the rest.”
And also 2tb SSD……. nice!
There is generally no data
There is generally no data recovery portion of an SSD warranty. If they have no backup, it's up to the customer to send it off to a data recovery place and try to get the data back *before* doing the simple RMA swap with the seller. Also, sellers will still accept the return if it has been opened by a data recovery place (they mark the drive as such). Same applies to hard drives.
RMA is a one for one exchange
RMA is a one for one exchange as always.. Data Recovery is and has always been a responsibility of the user. That is what backup’s are for.
Finally! I purchased 2 1TB
Finally! I purchased 2 1TB 840 EVO’s for about $1000 so 2TB for $800 is a decent price but I suspect it will come down in price even more with sales and competition.
Been HDD free for over a year. I will wait for this to come down in price a bit more.
If you have those two in
If you have those two in RAID-0, you're still getting 2x the performance of even the 2TB model, as a single SATA channel is clearly the bottleneck for such a large capacity SSD.
Guess I’ll be buying 2 of
Guess I’ll be buying 2 of these when my credit card empties.. Already filled the 2 1tb 850 evos with flight simulator x and steam sales.
Bye bye money.
Get another pair of 1TB’s and
Get another pair of 1TB's and quad RAID-0 it? (Also, back up!)
That might be a possibility
That might be a possibility if they drop the price of the 1TB’s to a reasonable degree. If there is only a $40-50ish difference between the 1TB and 2TB, I can’t say it would be worthwhile just to save $100ish.
Naw, he should just get 4 of
Naw, he should just get 4 of these. Then Quad RAID it. Why settle for less? 🙂
I hope they won’t get rid of
I hope they won’t get rid of their 128gb drives like how 64gb drives are disappearing, they worked great for boot drives esp for Linux distros which are rather small.
Anyway, exciting news hopefully prices keep getting driven down I want to replace my vertex 3 128gb next year.
It’s great to finally see 2TB
It’s great to finally see 2TB capacities at reasonable prices! Unfortunately I won’t be touching anything from Samsung until they address their TRIM related data loss issues: https://blog.algolia.com/when-solid-state-drives-are-not-that-solid/
Do you use a linux server in
Do you use a linux server in RAID on EXT4? If the answer to this question is no, then you have nothing to worry about with TRIM.
This is one blog reporting an issue they’ve noticed in their systems. No one else has reported anything similar. TRIM on SSDs have been working fine in Windows and OSX. Most of the listed Samsung SSDs have been out for several years. If there was a TRIM data loss issue with those drives, i’m pretty sure there would be widespread reporting on it.
The first site to report on this was Tech Report, and they reported it as if everyone’s Samsung SSD is gonna lose data, when its simply not true. In reality, as i stated above, it was one blog. If you look at Tech Report’s history on reporting Samsaung SSD “issues”, they seem to live off spreading widespread panic.
Yeah, that is specific to
Yeah, that is specific to queued TRIM, and is not an issue if you are on the current Kernel since they have that SSD blacklisted now. Also note that Samsung is not the only brand of SSDs present on that blacklist.
The Queued TRIM issue is a
The Queued TRIM issue is a separate issue from what Algolia is reporting.
Algolia specifically states that “UPDATE June 16:
A lot of discussions started pointing out that the issue is related to the newly introduced queued TRIM. This is not correct. The TRIM on our drives is un-queued and the issue we have found is not related to the latest changes in the Linux Kernel to disable this features.”
Yes, but the thing is there
Yes, but the thing is there are a lot of these drives out in the wild, and we would be seeing a lot of users with corrupted partitions or lost data if what they said in that update applied to everyone with these drives.
I wrote the post with the
I wrote the post with the mention of tech report.
I agree with you. It is annoying how some people are taking this as fact and are jumpin the bandwagon. Like i said, tech report’s article made it sound like is a fact that everyone is affected by this, when its just the algolia ppl who are affected.
Awesome, cant believe I’m
Awesome, cant believe I’m seeing this capacity now. When I move to a 4k workflow, I will definitely consider these drives! Thanks 4 the review Allyn.
“Pricing:
Intro MSRP’s
850
“Pricing:
Intro MSRP’s
850 Pro 2TB*: $1000 ($0.49/GB*) (Amazon)
850 EVO 2TB: $800 ($0.40/GB) (Amazon)”
The Amazon links don’t go to any product page. The drives don’t currently exist on Amazon’s website, so what are the links for?
The links are to the search
The links are to the search as they have not yet been listed for sale. They also include our affiliate code, which helps benefit the site should someone make a purchase as a result of using them.
Hey Allyn,
Is there an
Hey Allyn,
Is there an article that describes the differences between each of your SSD benchmarks? File server vs web server, etc. And latency? I’m not sure how that plays into performance either.
Also, it would be great if there was an article for traced-based analysis of gaming loads, or if you could recommend a program for us to test it with our own games. I was wondering if the faster SSDs (Intel 750, Samsung sm951 Nvme) make any difference at all for something like Skyrim with a million mods, running around outside… I think the CPU usage hits 25% during cell loading and I think the game might actually be CPU bottlenecked with 850 Pro since it’s Sata.
Or am I spitting out nonsence?
Anyways, keep up the good work. :p
We don’t spell out the exact
We don't spell out the exact workloads in order to keep these tests 'ours', but they are simply based on their descriptor (file server, etc).
We used to use PCMark for the trace based stuff, but it got to the point where there was little to no difference across batches of SSDs – even SSDs that we knew would be performing differently based on their other specifications / benchmark results. The issue was that PCMark is weighing the time where the storage is idle too heavily in its results. An extra 10 seconds of loading time out of a 10 minute gamimg session doesn't add a large percentage to the total, but a gamer is actually *waiting* on that load and would be much more inclined to note the performance difference. PCMark also does not correct or test for SSDs being filled to various capacities. We're working on something better, but we're not there yet.
As far as 'CPU bottlenecking because of SATA', that shouldn't really be a thing with a single SATA SSD (unless the game is requesting >100k IOPS, which is highly unlikely). It's more likely that the game is sharing a thread to both load and process the data being read. If that game is pegging a core when doing so with a SATA SSD, it's not likely to be any faster if loading from a different source. Situations like this are what would make SSD testing very game / CPU speed specific, and would rapidly add too many variables to justify the testing, as just testing some level loads with some games would not paint the full picture for someone on a slower CPU. That CPU dependency is also why I see trace-based game load testing as an 'incorrect' test. None of the trace playback utilities can emulate the actual game load thread's CPU dependency. An SSD 750 may play back a trace at 2x the speed of a SATA SSD, but if the game was actually CPU bottlenecked on the SATA SSD, it would not go any faster on the PCIe SSD when retested with the game itself.
Hey Allyn,
Thanks for the
Hey Allyn,
Thanks for the reply.
When I was talking about cell loading in Skyrim causing 25% CPU usage, it wasn’t a suggestion for a test because as you said, it would be way too hard to test. I was just curious whether I would see a gain if I decide to get an NVMe drive for myself.
So if that one maxed out single core is at 25%, it must be sending requests to the SSD and processing the rest of the game and whatever it gets from the SSD all in that one core, leading to the CPU core being pegged, totally busy. So then in theory an NVMe SSD would lower the CPU usage for the requests to the SSD so that the game has more CPU cycles to do its other tasks while loading? That would… in theory… mean that the game will load faster.
In theory, of course. 🙂
Correct me if I’m wrong, but wouldn’t traced based analysis help us see what type of loads a game demands from an SSD? For example, game A might request a ton of random 4k reads, game B might want 512kb random reads, but when starting the game up, request sequential reads, and saving a save file would mean doing a bunch of random and sequential writes, etc. I guess the data doesn’t really matter in the real world, and this would be more for curiosity’s sake.
2 TB is the sweet spot for
2 TB is the sweet spot for primary hard drive. Unfortunately, it’s still too expensive IMHO compared to SATA so I’ll wait until it’s down to $300. SATA is still good enough for me.
If I’m spending $800 on an
If I’m spending $800 on an SSD, I would want a superfast one like the Intel 750 series with a lower capacity. Also, I don’t know about most people here, but I wouldn’t be able to fit my steam library in a 2TB drive.. If I had two of these I would be able to fit most of my games.. But I’m okay with having most of my games on HDDs and only some of them on my SSD.
This could be really nice for laptops that don’t have mSATA or M.2…
Still a hard sell for $800 which is more than most people’s PCs.
I still think SSDs can’t fully replace a hard drive yet, they are just not cheap enough. They have come a long way, but we need more!
If I was going to pay a grand
If I was going to pay a grand for an ssd I’d much rather get an Intel 750 series 1.2TB.. at least I’d get a noticable performance boost along with the capacity upgrade.
Did Samsung ever explain/fix
Did Samsung ever explain/fix the slow I/O Meter performance you show in your July review of the 2TB EVO at low queue depth? Is it now equivalent to the PRO in low queue depth performance? Thanks.
Hi there
Thanks for this
Hi there
Thanks for this awesome benchmark! I’m also very interested to know if you have heard something from Samsung about the slow I/O Meter performance you show in your July review of the 2TB EVO?
Thanks a lot for your time.
BR