Investigating the issue
** Edit ** (24 Sep)
We have updated this story with temperature effects on the read speed of old data. Additional info on page 3.
** End edit **
** Edit 2 ** (26 Sep)
New quote from Samsung:
"We acknowledge the recent issue associated with the Samsung 840 EVO SSDs and are qualifying a firmware update to address the issue. While this issue only affects a small subset of all 840 EVO users, we regret any inconvenience experienced by our customers. A firmware update that resolves the issue will be available on the Samsung SSD website soon. We appreciate our customer’s support and patience as we work diligently to resolve this issue."
** End edit 2 **
Over the past week or two, there have been growing rumblings from owners of Samsung 840 and 840 EVO SSDs. A few reports scattered across internet forums gradually snowballed into lengthy threads as more and more people took a longer look at their own TLC-based Samsung SSD's performance. I've spent the past week following these threads, and the past few days evaluating this issue on the 840 and 840 EVO samples we have here at PC Perspective. This post is meant to inform you of our current 'best guess' as to just what is happening with these drives, and just what you should do about it.
The issue at hand is an apparent slow down in the reading of 'stale' data on TLC-based Samsung SSDs. Allow me to demonstrate:
You might have seen what looks like similar issues before, but after much research and testing, I can say with some confidence that this is a completely different and unique issue. The old X25-M bug was the result of random writes to the drive over time, but the above result is from a drive that only ever saw a single large file write to a clean drive. The above drive was the very same 500GB 840 EVO sample used in our prior review. It did just fine in that review, and at afterwards I needed a quick temporary place to put a HDD image file and just happened to grab that EVO. The file was written to the drive in December of 2013, and if it wasn't already apparent from the above HDTach pass, it was 442GB in size. This brings on some questions:
- If random writes (i.e. flash fragmentation) are not causing the slow down, then what is?
- How long does it take for this slow down to manifest after a file is written?
Subject: General Tech | August 7, 2014 - 10:37 AM | Ken Addison
Tagged: podcast, video, Thecus, n2560, asus, strix, strix 780, flash media summit, Samsung, tlc, vnand, Marvell, gtx 880, x99s sli plus
PC Perspective Podcast #312 - 08/07/2014
Join us this week as we discuss the Thecus N2560 NAS, ASUS STRIX GTX 780, Flash Media Summit News and more!
The URL for the podcast is: http://pcper.com/podcast - Share with your friends!
- iTunes - Subscribe to the podcast directly through the Store
- RSS - Subscribe through your regular RSS reader
- MP3 - Direct download link to the MP3 file
Hosts: Jeremy Hellstrom, Josh Walrath, Scott Michaud, and Morry Tietelman
Week in Review:
EVGA Contest Winner
User: Lt Dan 521
News items of interest:
Flash Memory Summit 2014
Hardware/Software Picks of the Week:
Ryan: Soccer moms
Jeremy: Playing StarCraft On An ARM, natively!
Subject: Storage, Shows and Expos | August 5, 2014 - 01:19 PM | Allyn Malventano
Tagged: FMS, vnand, tlc, ssd, Samsung, FMS 2014, Flash Memory Summit
Just minutes ago at the Flash Memory Summit, Samsung announced the production of 32-layer TLC VNAND:
This is the key to production of a soon-to-be-released 850 EVO, which should bring the excellent performance of the 850 Pro, with the reduced cost benefit we saw with the previous generation 840 EVO. Here's what the progression to 3D VNAND looks like:
3D TLC VNAND will look identical to the right most image in the above slide, but the difference will be that the charge stored has more variability. Given that Samsung's VNAND tech has more volume to store electrons when compared to competing 2D planar flash technology, it's a safe bet that this new TLC will come with higher endurance ratings than those other technologies. There is much more information on Samsung's VNAND technology on page 1 of our 850 Pro review. Be sure to check that out if you haven't already!
Another announcement made was more of an initiative, but a very interesting one at that. SSDs are generally dumb when it comes to coordinating with the host - in that there is virtually no coordination. An SSD has no idea which pieces of files were meant to be grouped together, etc (top half of this slide):
Stuff comes into the SSD and it puts it where it can based on its best guess as to how it should optimize those writes. What you'd want to have, ideally, is a more intelligent method of coordination between the host system and the SSD (more like the bottom half of the above slide). Samsung has been dabbling in the possibilities here and has seen some demonstrable gains to be made. In a system where they made the host software aware of the SSD flash space, and vice versa, they were able to significantly reduce write latency during high IOPS activity.
The key is that if the host / host software has more control over where and how data is stored on the SSD, the end result is a much more optimized write pattern, which ultimately boosts overall throughput and IOPS. We are still in the experimentation stage on Storage Intelligence, with more to follow as standards are developed and the industry pushes forward.
It might be a while before we see Storage Intelligence go mainstream, but I'm definitely eager to see 3D TLC VNAND hit the market, and now we know it's coming! More to follow in the coming days as we continue our live coverage of the Flash Memory Summit!
Subject: General Tech, Storage | June 4, 2014 - 04:37 PM | Scott Michaud
Tagged: computex 2014, computex, tlc, ssd, Samsung, 845DC EVO
Well that was an alphabet soup of a title.
Samsung has just announced a new line of SSDs, based on three bit per cell (TLC) memory, for enterprise customers. The Samsung 845DC EVO is rated at 530MB/s reads with 87,000 IOPS. The company will also cover up to 600TB of writes under its warranty (no mention of length in years, though). The drive will be available "later this month" in 240GB, 480GB, and 960GB models. Samsung did not mention price in their press release, but Anandtech claims the 240GB will be $250, the 480GB will be $490, and the 960GB will be $969.
Samsung's SSDs will give you some TLC???
This is basically $1/GB scaling, plus $10. I must admit, this is getting pricy. In the consumer space, we have recently seen 512GB for $199. That said, SSDs are not known for sticking to their MSRP. Also, these are enterprise-rated drives. Being TLC-based, I wonder how much (if any) SLC-style write cache was included, as per the consumer 840 EVO.
Lastly, Samsung claims that these drives use around 4W under load. This is much lower than hard drives but a little high for SSDs, according to benchmarks that I have seen. That said, there are a few ways to parse that (for example, if they mean that its peak is typically 4W, which would be pretty good for a 960GB drive).
The Samsung 845DC EVO will be available later this month for a little over $1/GB.
Subject: General Tech | April 30, 2014 - 09:59 AM | Jeremy Hellstrom
Tagged: Samsung, ssd, tlc
Samsung has been working with TLC flash for a while now, both the original 840 and the 840 EVO utilize that type of flash, the increased yields offer lower pricing at the cost of a reduced number of writes before the flash begins to fail. The Register has posted their announcement of a new product line aimed at the data centre; the PM835T family will come in 240GB, 480GB and 960GB models and will also use TLC flash, with pricing predicted to be comparable to consumer level drives. With Samsung's 10nm-class TLC flash the experts at SMART suspect a 500 phase/erase cycle lifetime however depending on how Samsung has designed the drives the actual number could be much higher, they do offer a 3 year warranty on their current TLC drives. For now Samsung is not releasing an official expected lifetime for these drives which raises a question, will enterprise feel the short term cost savings are worth the long term replacement costs?
"Triple-level cell (TLC) flash chips mean fabs can extract more flash capacity from a silicon wafer, and so production costs are lower than for two-level cell MLC technology. Samsung says it gets "a 30 per cent increase in manufacturing efficiency compared to SSDs that use 2-bit NAND flash components."
Here is some more Tech News from around the web:
- Sandisk smashes glass ceiling with 4TB enterprise SSD @ The Inquirer
- Acer debuts new products @ DigiTimes
- Android accounts for whopping 99 percent of mobile malware @ The Inquirer
- StarTech USB 3.0 Hub with Ethernet Review @ TechwareLabs
Subject: Storage | July 26, 2013 - 03:08 PM | Jeremy Hellstrom
Tagged: TurboWrite, tlc, ssd, slc, Samsung, 840 evo, MEX controller
Along with Al's review of the new EVO line you can get a second opinion from The Tech Report about the performance of the new SSD with a fast cache. The majority of the storage is 19nm TLC NAND but there is an SLC cache sitting between the controller and that long term TLC storage to help with the overall responsiveness of the drive, aka TurboWrite. In the 120 and 250GB models that cache is 3GB while in the larger models you get a 6GB cache. In their real world testing the new EVO drive is incredible at large file copying though Sandforce drives can beat it in small file copy speeds, likely thanks to the compressed write trickery that controller family is so good at. Check out the review here and keep your fingers crossed that MSRP is the acual price these drives sell at.
"Samsung's entry-level 840 EVO SSD combines affordable TLC NAND with a server-style SLC cache. We explain the drive's unique buffering tech and explore how it affects performance."
Here are some more Storage reviews from around the web:
- Samsung 840 EVO SSD @ The SSD Review
- Samsung SSD 840 EVO Review: 120GB, 250GB, 500GB, 750GB & 1TB Models Tested @ AnandTech
- Samsung 840 EVO 250GB, 750GB SSD Review @ Custom PC Review
- Samsung 840 Evo SSD @ Hardware.info
- Samsung unveils 840 EVO solid-state drive family @ The Tech Report
- 240GB OCZ Vertex 450 Solid State Drive @ Benchmark Reviews
- Plextor M5M 128GB mSATA SSD Review @ Legit Reviews
- OCZ Vertex 3.20 240GB SSD @ eTeknix
- OCZ Vector 512GB SSD @ Kitguru
- RunCore Pro IV 1.8 Inch ZIF SSD @ LanOC Reviews
- Silicon-Power Velox V55 240GB @ Legion Hardware
- Seagate 600 Pro SSD 400GB @ Bjorn3D
- Securely Erasing Your SSD with Linux: A How-To @ Techgage
- Seagate Central 3TB review: User-friendly? @ Hardware.info
- Silicon Power Blaze B20 32GB USB 3.0 Flash Drive @ NikKTech
- USB 3.0 Flash Drive Roundup July 2013 @ Legion Hardware
- Icy Dock FlexCage 2 Bay and 3 Bay Hard Drive Enclosure Review @ HiTech Legion
- Zalman ZM-VE400 USB 3.0 HDD/SSD Enclosure @ Funky Kit
- QNAP TS-421 & QTS 4.0 @ techPowerUp
- Thecus N2520 review: first NAS with Intel Atom CE5315 @ Hardware.info
Introduction and Specifications
Last week, Samsung flew a select group of press out to Seoul, Korea. The event was the 2013 Samsung Global SSD Summit. Here we saw the launch of a new consumer SSD, the 840 EVO:
This new SSD aims to replace the older 840 (non-Pro) model with one that is considerably more competitive. Let's just right into the specs:
Subject: Storage | July 17, 2013 - 10:12 PM | Allyn Malventano
Tagged: tlc, ssd, slc, sata, Samsung, cache, 840 evo
Samsung's release of the 840 EVO earlier today likely prompted some questions, such as what type of flash does it employ and how does it achieve such high write speeds. Here is the short answer, with many slides in-between, starting off with the main differences between the 840 and the 840 EVO:
So, slightly increased specs to help boost drive performance, and an important tidbit in that the new SSD does in fact keep TLC flash. Now a closer look at the increased write specs:
Ok, the speeds are much quicker, even though the flash is still TLC and even on a smaller process. How does it pull off this trick? Tech that Samsung calls TurboWrite.
A segment of the TLC flash is accessed by the controller as if it were SLC flash. This section of flash can be accessed (especially written) much faster. Writes are initially dumped to this area and that data is later moved over to the TLC area. This happenes as it would in a normal write-back cache - either during idle states or once the cache becomes full, which is what would happen during a sustained maximum speed write operation that is larger than the cache capacity. Here is the net effect with the cache in use and also when the cache becomes full:
For most users, even the smallest cache capacity will be sufficient for the vast majority of typical use. Larger caches appear in larger capacities, further improving performance under periods of large write demand. Here's the full spread of cache sizes per capacity point:
So there you have it, Samsung's new TurboWrite technology in a nutshell. More to follow (along with a performance review coming in the next few days). Stay tuned!
Subject: General Tech | April 11, 2013 - 11:38 AM | Jeremy Hellstrom
Tagged: tlc, Samsung, flash memory, 10nm
A process shrink to 10nm wasn't enough for Samsung, they also doubled the density of their MLC flash storage to 128Gbits at a physical size of 170.6mm2. They claim write speeds of up to 18MB/sec and 400Mbit/s bandwidth through their toggle interface. Even better news for consumers is that this should be even cheaper than current MLC flash which will help continue to drive the price of SSDs down. Samsung didn't tell The Inquirer when we can expect to see these in devices but it should not be too long from now that we are doing longevity testing on this new Flash.
"MEMORY MAKER Samsung has announced that it is producing 128Gbit 3-bit multi-level cell (MLC) NAND flash chips for solid-state disk (SSD) drives"
Here is some more Tech News from around the web:
- SSH an ill-managed mess says SSH author Tatu Ylonen @ The Register
- 10 Hot New Linux-Ready Embedded ARM Modules @ Linux.com
- Microsoft: Here's some cash, channel. PLEASE sell Office 365 @ The Register
- TRENDnet AC1750 Dual Band Wireless Router (TEW-812DRU) Review @ Madshrimps
- ACTi D32 review: 3MP outdoor IP security camera @ Hardware.info
- Win a speedy Plextor M5 Pro Xtreme 256GB SSD @ Tweaktown
- Win a Cyberpower X7-100 Fangbook @ Kitguru
Taking an Accurate Look at SSD Write Endurance
Last year, I posted a rebuttal to a paper describing the future of flash memory as ‘bleak’. The paper went through great (and convoluted) lengths to paint a tragic picture of flash memory endurance moving forward. Yesterday a newer paper hit Slashdot – this one doing just the opposite, and going as far as to assume production flash memory handling up to 1 Million erase cycles. You’d think that since I’m constantly pushing flash memory as a viable, reliable, and super-fast successor to Hard Disks (aka 'Spinning Rust'), that I’d just sit back on this one and let it fly. After all, it helps make my argument! Well, I can’t, because if there are errors published on a topic so important to me, it’s in the interest of journalistic integrity that I must now post an equal and opposite rebuttal to this one – even if it works against my case.
First I’m going to invite you to read through the paper in question. After doing so, I’m now going to pick it apart. Unfortunately I’m crunched for time today, so I’m going to reduce my dissertation into the form of some simple bulleted points:
- Max data write speed did not take into account 8/10 encoding, meaning 6Gb/sec = 600MB/sec, not 750MB/sec.
- The flash *page* size (8KB) and block sizes (2MB) chosen more closely resemble that of MLC parts (not SLC – see below for why this is important).
- The paper makes no reference to Write Amplification.
Perhaps the most glaring and significant is that all of the formulas, while correct, fail to consider the most important factor when dealing with flash memory writes – Write Amplification.
Before geting into it, I'll reference the excellent graphic that Anand put in his SSD Relapse piece:
SSD controllers combine smaller writes into larger ones in an attempt to speed up the effective write speed. This falls flat once all flash blocks have been written to at least once. From that point forward, the SSD must play musical chairs with the data on each and every small write. In a bad case, a single 4KB write turns into a 2MB write. For that example, Write Amplification would be a factor of 500, meaning the flash memory is cycled at 500x the rate calculated in the paper. Sure that’s an extreme example, but the point is that without referencing amplification at all, it is assumed to be a factor of 1, which would only be the case if you were only writing 2MB blocks of data to the SSD. This is almost never the case, regardless of Operating System.
After posters on Slashdot called out the author on his assumptions of rated P/E cycles, he went back and added two links to justify his figures. The problem is that the first links to a 2005 data sheet for 90nm SLC flash. Samsung’s 90nm flash was 1Gb per die (128MB). The packages were available with up to 4 dies each, and scaling up to a typical 16-chip SSD, that only gives you an 8GB SSD. Not very practical. That’s not to say 100k is an inaccurate figure for SLC endurance. It’s just a really bad reference to use is all. Here's a better one from the Flash Memory Summit a couple of years back:
The second link was a 2008 PR blast from Micron, based on their proposed pushing of the 34nm process to its limits. “One Million Write Cycles” was nothing more than a tag line for an achievement accomplished in a lab under ideal conditions. That figure was never reached in anything you could actually buy in a SATA SSD. A better reference would be from that same presentation at the Summit:
This shows larger process nodes hitting even beyond 1 million cycles (given sufficient additional error bits used for error correction), but remember it has to be something that is available and in a usable capacity to be practical for real world use, and that’s just not the case for the flash in the above chart.
At the end of the day, manufacturers must balance cost, capacity, and longevity. This forces a push towards smaller processes (for more capacity per cost), with the limit being how much endurance they are willing to give up in the process. In the end they choose based on what the customer needs. Enterprise use leans towards SLC or eMLC, as they are willing to spend more for the gain in endurance. Typical PC users get standard MLC and now even TLC, which are *good enough* for that application. It's worth noting that most SSD failures are not due to burning out all of the available flash P/E cycles. The vast majority are due to infant mortality failures of the controller or even due to buggy firmware. I've never written enough to any single consumer SSD (in normal operation) to wear out all of the flash. The closest I've come to a flash-related failure was when I had an ioDrive fail during testing by excessive heat causing a solder pad to lift on one of the flash chips.
All of this said, I’d love to see a revisit to the author’s well-structured paper – only based on the corrected assumptions I’ve outlined above. *That* is the type of paper I would reference when attempting to make *accurate* arguments for SSD endurance.
Get notified when we go live!