Samsung SSD 960 Evo 500GB Review: The king of the TLC hill

[Steve]

We liked what we saw when we tested Samsung's SSD 960 Pro nearly a month ago. The only drawback to Samsung's new series so far appears to be the price and while not outrageous by any means, a cost of $0.64/GB for the cheapest model is still expensive. Remedying this concern, Samsung has unleashed its more affordable 960 Evo series, which is also made in a smaller 250GB capacity.

Like the 960 Pro, the 960 Evo is powered by Samsung's Polaris controller coupled with Samsung's 48-layer V-NAND. Although the 960 Evo has been outfitted with TLC memory instead of MLC, the claimed impact on performance is minor with sequential read speeds dropping from 3.5GB/s to 3.2GB/s and write throughput going from 2.1GB/s to 1.9GB/s.

In other words, you'd be paying over 20% less for the 960 Evo than the Pro and yet the impact on performance could be considerably less than that.

Read the complete review.

 

[BSim500]

The 960 Pro 512GB model for example sports an endurance rating of 400 terabytes written and that figure has been halved for the 500GB Evo model. As such, the warranty period offered by Samsung has been reduced from five years to three.

Wait a minute, wasn't the warranty for the 850 EVO's (TLC) 5-years vs 10-year for PRO (MLC)? So is this shorter 3 year warranty now an admission of the 960 EVO's being even less durable than the 850 EVO's? What cell size does it use, still 40nm or has that been shrunk? Why no mention of this stuff?

In as little as 8 weeks, many people saw the read speeds of the Crucial BX200 (TLC) halving (like the Samsung 840's) and yet none of the tech sites that reviewed it on freshly written data even noticed:-

^ The issue with TLC based drives is no longer "you probably won't wear it out" (as in maximum lifespan writes), it's measurable TLC-related read speed slowdowns of "stale" data over several weeks / months as processes shrink. Samsung 850's avoided it by using large 40nm cells, but if they're shrinking those to match the "race to the bottom" rat race of 15-16nm planar drives (that are dirt cheap for a reason), then it could rear its ugly head again.

We are way overdue for another serious long-term SSD roundup test measuring this stuff, especially given the same tech sites pumping "the death of the HDD" clickbait can't figure out how unsuitable the cheapest drives are for 99% unpowered backup drives. Tech sites all benchmark on 5 minute old data, and yet in real life SSD's for both system and backup drives are used in the exact opposite way - a lot of stuff gets written in the first week that starts degrading over time (due to the nature of the tech), which is why it's important to measure it and related countermeasures (ie, does the firmware constantly rewrite old data, if so how much does that reduce endurance. If not, how much do the files slow down), etc.

^ The first tech site that does this stuff will get permanently ad-block white-listed for life. CrystalDiskMark charts showing one drive 4x faster don't mean a damn thing for real world gaming load times. Nor does "real world" benchmarks which involve reading back only 5 minute old data convey that one drive may then sustain those read speeds over a year whilst another drive completely fall apart after 3 months (which is exactly what's happening to some TLC vs MLC drives in "the real world").

 

[Steve]

Wait a minute............................

WOW what a rant, that was a real work out for the first comment. FYI we condition all SSDs before testing, which is why the BX200 sucks so bad in our tests.

 

[BSim500]

WOW what a rant, that was a real work out for the first comment. FYI we condition all SSDs before testing, which is why the BX200 sucks so bad in our tests.

It's more of an expression of frustration, Steve, than a rant. Most sites that said "it sucked" did so not for the reasons mentioned (most of the same sites also gave glowing reviews to the Samsung 840 EVO's during launch month too). There have been large threads on other tech sites posting images of slowdowns over time of various TLC drives, there's a widespread "sour taste" in many people's mouths after the 840 debacle, and wanting long-term TLC reliability tests done as a response (not just SSD's but flash drives too) is hardly outrageous...

Easy to "shoot the messenger" but I notice the questions still remain unanswered : Why have all the warranties on the 960 drives almost halved vs the 850's (3yr TLC and 5yr MLC vs previous 5yr and 10yr)? Why is the 960 EVO now being treated by Samsung themselves as having "840 EVO Class reliability" based on warranty length (the ultimate "confidence litmus test")? What has changed technologically to cause this? What size process is the drive based on, 40nm? 25nm? Something else? Are we heading back downward in a "shrinking cell, reduced durability" trend again?

As you said yourself in your 850 PRO review "Power users concerned about the write limitations of any SSD will be pleased to learn that Samsung offers an incredible 10-year warranty with the 850 Pro SSD series and that right there is a very big deal." What's changed?

 

[Radical]

Wait a minute............................

WOW what a rant, that was a real work out for the first comment. FYI we condition all SSDs before testing, which is why the BX200 sucks so bad in our tests.

Steve, what does 'conditioning' actually entail?

This is just a casual thought, and definitely not a judgement as I think Techspot is a damn fine web paper, but perhaps there should be a section somewhere entitled 'How We Test', which includes a bit of insight into the various methodologies you guys employ, along with some of the inherent complexities involved in testing new tech.

Just a thought...

 

[Bigtruckseries]

I recently bought Samsung's 4TB SSD because I edit 4K video and I wanted large storage and fast transfer.

For people who are building gaming PC, I really recommend you purchase a 1TB nvme instead (to start with) and a large capacity HDD to store music, videos documents, etc.

Run your games from the SSD and your non-speed-necessary files from the HDD.

500GB just doesn't hold much in terms of gaming because these new game files are HUGE.
My Fallout 4 alone is damn near 80GB.


A smarter move is to buy only 1TB for your Desktop and then store all your non-speed-necessary files on a NAS server.

I bought a Western Digital My Cloud Mirror 12GB for home storage and the beauty is I can access the files using desktop apps from anywhere in the world. Documents, music, picture and small video files are quick and fast loaders over WiFi - but you still have access on LTE.

 

[Theinsanegamer]

What about power consumption with the samsung driver? Toms Hardware found that using the samsung driver instead of the microsoft driver significantly reduced battery life in laptops. This is something else that needs to be brought up under the cons list.

 

[ddferrari]

WOW what a rant, that was a real work out for the first comment. FYI we condition all SSDs before testing, which is why the BX200 sucks so bad in our tests.

It's more of an expression of frustration, Steve, than a rant. Most sites that said "it sucked" did so not for the reasons mentioned (most of the same sites also gave glowing reviews to the Samsung 840 EVO's during launch month too). There have been large threads on other tech sites posting images of slowdowns over time of various TLC drives, there's a widespread "sour taste" in many people's mouths after the 840 debacle, and wanting long-term TLC reliability tests done as a response (not just SSD's but flash drives too) is hardly outrageous...

Easy to "shoot the messenger" but I notice the questions still remain unanswered : Why have all the warranties on the 960 drives almost halved vs the 850's (3yr TLC and 5yr MLC vs previous 5yr and 10yr)? Why is the 960 EVO now being treated by Samsung themselves as having "840 EVO Class reliability" based on warranty length (the ultimate "confidence litmus test")? What has changed technologically to cause this? What size process is the drive based on, 40nm? 25nm? Something else? Are we heading back downward in a "shrinking cell, reduced durability" trend again?

As you said yourself in your 850 PRO review "Power users concerned about the write limitations of any SSD will be pleased to learn that Samsung offers an incredible 10-year warranty with the 850 Pro SSD series and that right there is a very big deal." What's changed?

What has changed is Samsung's wallet- they're poised to lose 5 billion dollars in total over the Note 7 debacle, and now they have another recall on their washing machines.

To my knowledge no one else offers a 10-year warranty, so perhaps Samsung decided they didn't need to either. Their performance is why they sell so well, and how many people keep the same drive for 10 years anyway? I'll field that one- almost no one. Reducing the warranty to simply match what the competition is offering- and saving money in the process- seems more likely than a reduction in cell size.

Samsung has already addressed the slower read times of old data:

"Finally, on April 23, 2015, Samsung released a new firmware version EXT0DB6Q via Magician, and it appears to greatly improve the read speeds of old data (from <150 MB/s to >350 MB/s) while doing some sort of calibration/rewriting in the background to slowly bring the speeds of old files up to a full 500 MB/s and keep old data fresh."

I just tested my three year-old, half full 256GB 840 Pro on 'SSD Read Speed Tester 2.04' and had a nice flat graph across, with an average read speed of 526 MB/s. Plenty of original data on there too.

The good news is that you can choose to stick with HDD if you distrust SSD's long-term performance, but if this problem was as common as you make it sound, surely more people would have heard about it by now.

 

[Lightspeed]

If my memory serves me right, I remember that user's who installed the Samsung 950 Pro as Windows boot drive saw similar & at times, slower performance than using a typical SATA SSD. I wonder if this 960 Evo & the 960 Pro is any different. ??
What I really want is a system drive that makes Windows load in seconds, shut down in seconds and makes everything Windows does snappier than sata SSD's.

 

[BSim500]

Reducing the warranty to simply match what the competition is offering- and saving money in the process- seems more likely than a reduction in cell size.

That may be the case but I'm simply looking for confirmation of what cell size is used instead of "probably's", "maybe's" and "hopefully's".

Samsung has already addressed the slower read times of old data:

"Finally, on April 23, 2015, Samsung released a new firmware version EXT0DB6Q via Magician, and it appears to greatly improve the read speeds of old data (from <150 MB/s to >350 MB/s) while doing some sort of calibration/rewriting in the background to slowly bring the speeds of old files up to a full 500 MB/s and keep old data fresh."

I just tested my three year-old, half full 256GB 840 Pro on 'SSD Read Speed Tester 2.04' and had a nice flat graph across, with an average read speed of 526 MB/s. Plenty of original data on there too

A "flat graph" is entirely expected for PRO's as they're all MLC drives which never suffered from the problem in the first place. The slowdown problem was caused by combination of small process node + TLC combination (8x voltage states instead of 4) proving significantly less durable than anticipated due to reduced "overhead" to maintain a charge vs voltage drift, ie, 0%/14%/25%/43%/57%/71%/85%/100% vs 0%/33%/66%/100%. Such a low overhead that only something like 3x atoms per voltage state were holding the charge. That's why Samsung went back to 40nm with 3D-NAND in the first place.

Samsung's fix workaround for the 840 EVO's merely "hid" the issue by constantly rewriting data every few weeks (which slashed endurance rating far below original advertised one). What annoyed many people however, is that both the 840 EVO's and 840 (vanilla) both suffered from the problem, but the "fix" was only issues to the EVO's leaving 840 owners stuck with a half-broken drive (something that only about 2x out of 20 tech sites could be bothered to follow up). And that's why people are now wary of tech sites that "review" drives with very few actual facts of the underlying process nodes of new drives...

The good news is that you can choose to stick with HDD if you distrust SSD's long-term performance, but if this problem was as common as you make it sound, surely more people would have heard about it by now.

I already own 2x SSD's and have for years. Both are MLC (like yours) however, for a reason...

 

[Steve]

If my memory serves me right, I remember that user's who installed the Samsung 950 Pro as Windows boot drive saw similar & at times, slower performance than using a typical SATA SSD. I wonder if this 960 Evo & the 960 Pro is any different. ??
What I really want is a system drive that makes Windows load in seconds, shut down in seconds and makes everything Windows does snappier than sata SSD's.

Never saw that with mine, it's still as fast as the day I installed it :S

 

[ypsylon]

If my memory serves me right, I remember that user's who installed the Samsung 950 Pro as Windows boot drive saw similar & at times, slower performance than using a typical SATA SSD. I wonder if this 960 Evo & the 960 Pro is any different. ??
What I really want is a system drive that makes Windows load in seconds, shut down in seconds and makes everything Windows does snappier than sata SSD's.

It really depends what you do with you PC.

If you're running one NVMe drive for everything then booting fast shouldn't be an issue.

However worst thing you can have in the system are classic HDDs. Recently I went completely HDD free with my main workstation (yeah including some gaming but nothing special outside ME trilogy). It was extremely expensive with all these 1 and 2TB SSDs but finally done it. RAID controller which normally loads kernel for 60 seconds is doing that in ~45 now; from Windows sign to the desktop I cut about 20 seconds. It took full minute on i750 to reach desktop sometimes ( with all peripherals activated - printers, scanners and stuff). Now it went down to 35-40 seconds at most.

NVMe is ALWAYS slower to boot than AHCI device. And by at least 30%. I have another workstation (Z97) with RAID10 (via intel chipset) made out of old intel 530 120 GB drives (rated at only 24k IOPS). It takes less than 30 seconds from hitting the power button to the desktop. Culprit here is initialization process. PCI-Ex devices do that in different way which takes much more time.

NVMe in general is useless to a user which will never utilize raw power of that architecture (gaming is prime example). Especially in low-level access mode during POST before Windows Protected Mode kicks-in. Here NVMe behaves no different (speed-wise) than first SSD on the market from 20 years ago or so.

 

[Technosense]

My 950 Pro M.2 was a surprise in everyday use...coming from 850 pro and evo's, it' performance on my machine was 3-4x and fast things move along noticeably quicker.

I did notice my old 850 evo that was running a system for 4 years was at about 1/2 the speed of the other in sequential reads...I use it for games now, this using the typical 10% OP.

So yeah, this drive is probably something you probably won't regret upgrading...not for a minute.

 

[Raoul Duke]

If my memory serves me right, I remember that user's who installed the Samsung 950 Pro as Windows boot drive saw similar & at times, slower performance than using a typical SATA SSD. I wonder if this 960 Evo & the 960 Pro is any different. ??
What I really want is a system drive that makes Windows load in seconds, shut down in seconds and makes everything Windows does snappier than sata SSD's.

my 950 Pro 512 is as fast as the day it was installed. I think the difference you will see from a SATA SSD might vary according to how good your SATA SSD is.