Intel getting into the TLC SSD game??? Say it ain't so...

[VirtualLarry]

http://www.newegg.com/Product/Product.aspx?Item=N82E16820167374&ignorebbr=1

I thought I had read that Intel wanted to get out of the consumer SSD market, as it was in a "race to the bottom", which means tanking Gross Profit Margins.

Now they release a TLC SSD? How does that even remotely make sense? (Maybe to compete with Samsung's 750 EVO in the OEM space? The listing at Newegg doesn't seem to be for a retail-boxed drive.)

http://ark.intel.com/products/94420/Intel-SSD-540s-Series-120GB-2_5in-SATA-6Gbs-16nm-TLC

Edit: No mention of what controller it uses. I'm really curious if Intel developed their own controller, or if they are just re-branding some 3rd-party controller.

Anyways, according to ARK, there's no temperature-monitoring technology on the drive, which is slightly concerning. Unless what they are referencing in ARK is something other than the usual SMART temp monitoring.

Still has a 5-year warranty, which seems to me to be Intel's ringing endorsement of TLC NAND flash technology for the consumer space.

Might get one to play around with, can't really be worse than the BX200 I picked up recently.

Edit: This one might be more interesting. A 1TB TLC M.2 SATA6G SSD, for under $350, with an Intel 5-year warranty.
http://ark.intel.com/products/94421/Intel-SSD-540s-Series-1_0TB-M_2-80mm-SATA-6Gbs-16nm-TLC

Here's a chart that breaks out the specs for each capacity. Surprise, surprise, the specs listed by Newegg for the 120GB sized drive, are a bit... optimistic. Intel specs random 4K write for the 120GB size at 13,500 IOPS. (Newegg lists 80K?)

http://www.intel.com/content/www/us/en/solid-state-drives/solid-state-drives-540s-series.html

 

[BSim500]

Still has a 5-year warranty, which seems to me to be Intel's ringing endorsement of TLC NAND flash technology for the consumer space.

After seeing the Samsung 840 and BX200, no amount of endorsing will convince me to get a 16nm TLC drive.

Edit: The whole low-end SSD market is like a cliff-drop. MX100/MX200 (16nm MLC) or 850 EVO (40nm TLC) are quite a bit cheaper than the 850 PRO (40nm MLC) with few reliability compromises. But the compromises needed for a further slight price reduction for bottom rung 16nm TLC complete with questionable long-term unpowered data retention and measurable slowdowns over time are just too far for me. And that's just for main drive usage. For backup drives of critical data likely to be powered off 99% of the time, I'll take 2x 4TB / 3x 3TB / 4x 2TB HDD's over 1x flaky 1TB SSD any day (even if it does have Intel stamped on it). The only time I'd ever use one is as a "disposable" scratch drive / SRT cache.

 

[cbn]

http://www.guru3d.com/news-story/intel-to-launch-540s-ssd-with-tlc-nand-flash-memory.html

According to the above link this drive will use SM2256 and SK Hynix 16nm planar TLC NAND.

This is the same controller and NAND combination used by the ADATA Premier SP550.

 

[VirtualLarry]

Thanks CBN. I'm not sure if I want to pay the premium for an Intel-labelled SP550 drive, even if it does have two years more of warranty.

I'm also fairly disappointed to find out that they aren't even using their "own" NAND. (The Intel-Micron joint-venture NAND fab(s).)

 

[cbn]

I'm also fairly disappointed to find out that they aren't even using their "own" NAND. (The Intel-Micron joint-venture NAND fab(s).)

According to the following post Intel didn't split the 16nm node with Micron:

http://www.tomshardware.com/news/intel-ssd-540s-5400-tlc,31525.html

Intel did not leap with Micron to build 16nm flash, a move that would have guaranteed the company in-house 16nm TLC. Instead, Intel chose to skip the node entirely, leaving Micron to foot the bill. This also left Intel, a company with NAND flash manufacturing capabilities, as a customer of SK Hynix for this product cycle. We suspect Intel expected new 256Gbit die 3D flash memory to come sooner than it has from its joint venture with Micron, IMFT.

 

[Hellhammer]

According to the following post Intel didn't split the 16nm node with Micron:

http://www.tomshardware.com/news/intel-ssd-540s-5400-tlc,31525.html

That is correct. Intel didn't invest in 16nm at all and they only own a share in the Utah fab, which is now producing 3D XPoint (perhaps some 20nm NAND as well). Hence Intel is sourcing NAND from SK Hynix on top of their own supply and supply agreements with Micron.

 

[R0H1T]

That is correct. Intel didn't invest in 16nm at all and they only own a share in the Utah fab, which is now producing 3D XPoint (perhaps some 20nm NAND as well). Hence Intel is sourcing NAND from SK Hynix on top of their own supply and supply agreements with Micron.

Seems like they don't even want to compete in the consumer NAND space

 

[Johnny Lucky]

Larry - Intel isn't the only company. News reports indicate other companies will be switching to TLC flash memory this year.

 

[Elixer]

Edit: No mention of what controller it uses. I'm really curious if Intel developed their own controller, or if they are just re-branding some 3rd-party controller.

That is what I am curious about. In order to be using TLC, then, they must have amped up the error detection, among other reliability routines to be able to offer a longer warranty.

 

[VirtualLarry]

Larry - Intel isn't the only company. News reports indicate other companies will be switching to TLC flash memory this year.

I really don't see why. For the miniscule price / cost savings, compared to the almost order-of-magnitude (well, OK, technically not, but I'm going to exaggerate for good purpose here) difference in reliability and data longevity?

I'm surprised that Intel, with their vaunted reputation for reliability above all in SSDs, is even touching TLC with a 10foot pole. Then again, they did sell SSDs with 2nd-gen SandForce controllers too...

Edit: The thing is, there's not even any price savings, at the consumer level, for these new Intel SSDs! They're currently selling for MORE than their prior MLC models. Which, to me, makes no sense.

 

[coercitiv]

Edit: The thing is, there's not even any price savings, at the consumer level, for these new Intel SSDs! They're currently selling for MORE than their prior MLC models. Which, to me, makes no sense.

Their prior MLC model is currently heavily discounted, at least in my country.

 

[bononos]

Seems like they don't even want to compete in the consumer NAND space

Intel has chosen to move production to China by retooling one of their fabs for vnand production and not teaming up with Micron. So Intel is going to be pumping out alot of vnand in future.

 

[myocardia]

After seeing the Samsung 840 and BX200, no amount of endorsing will convince me to get a 16nm TLC drive.

The Samsung 840 non-Evo used 21nm TLC, and the 840 Evo uses 19nm TLC.

Most of the Samsung silicon inside the 840 EVO is second-generation TLC NAND. These chips are fabbed on a 19-nm process, just a small step down from the 21-nm node used for the 840 Series' flash.

 

[cbn]

Might get one to play around with, can't really be worse than the BX200 I picked up recently.

I haven't had a chance to use a planar TLC drive yet.

For a basic office/browsing machine do you think the average person would notice the BX200 being slower than a faster SSD?

Also is your BX200 120GB 240GB? or a higher capacity?

 

[cbn]

After seeing the Samsung 840 and BX200, no amount of endorsing will convince me to get a 16nm TLC drive.

Edit: The whole low-end SSD market is like a cliff-drop. MX100/MX200 (16nm MLC) or 850 EVO (40nm TLC) are quite a bit cheaper than the 850 PRO (40nm MLC) with few reliability compromises. But the compromises needed for a further slight price reduction for bottom rung 16nm TLC complete with questionable long-term unpowered data retention and measurable slowdowns over time are just too far for me.

Those are some interesting results in your link "measurable slowdowns over time".

Apparently slow reads of old data was a problem on the Samsung 840 EVO (19nm TLC NAND), but they were able to fix that with firmware:

http://www.overclock.net/t/1512915/...es-benchmarks-needed-to-confirm-affected-ssds

 

[Glaring_Mistake]

Those are some interesting results in your link "measurable slowdowns over time".

Apparently slow reads of old data was a problem on the Samsung 840 EVO (19nm TLC NAND), but they were able to fix that with firmware:

http://www.overclock.net/t/1512915/...es-benchmarks-needed-to-confirm-affected-ssds

Yeah, those are the results of the latest test I performed on a BX200 and which was included about my thread on preliminary results regarding dropping read speeds for the BX200.

 

[JimmiG]

After seeing the Samsung 840 and BX200, no amount of endorsing will convince me to get a 16nm TLC drive.

Edit: The whole low-end SSD market is like a cliff-drop. MX100/MX200 (16nm MLC) or 850 EVO (40nm TLC) are quite a bit cheaper than the 850 PRO (40nm MLC) with few reliability compromises. But the compromises needed for a further slight price reduction for bottom rung 16nm TLC complete with questionable long-term unpowered data retention and measurable slowdowns over time are just too far for me. And that's just for main drive usage. For backup drives of critical data likely to be powered off 99% of the time, I'll take 2x 4TB / 3x 3TB / 4x 2TB HDD's over 1x flaky 1TB SSD any day (even if it does have Intel stamped on it). The only time I'd ever use one is as a "disposable" scratch drive / SRT cache.

Yeah, the recommendation for SSDs used to be, just buy the cheapest one, all of them are much faster than HDDs. This doesn't apply any more. Getting lower performance from a cheap drive doesn't bother me, but the data retention issues and performance degradation over time are huge issues that should have been ironed out at the testing/development stage long before they were released to consumers.
Maybe Intel have finally mastered 16nm TLC, but I'd rather not risk it.

 

[cbn]

Yeah, those are the results of the latest test I performed on a BX200 and which was included about my thread on preliminary results regarding dropping read speeds for the BX200.

Assuming this problem with planar TLC is able to be fixed in the same way it was with the 840 EVO (which involves re-writing the old data to make it fresh again) how much do we know about planar TLC that is left unpowered? (eg, laptop that hasn't been used for a few months or even a year).

Also another thing I am wondering about is how much these SM2256 drives slow down when the NAND's tunnel oxide wears out. According to the Anandtech SM2256 article LDPC soft decode (and RAID parity mechanism) kicks in when this happens......but it is slower than the fast decode that is used when the cells are new.


http://www.anandtech.com/show/9165/...-reference-design-ssd-review-tlc-for-everyone

Architecturally the SM2256 shares the same core design as its predecessor SM2246EN. The design is modular, which allows Silicon Motion to change parts of the controller without redoing the rest. It features the same single 32-bit Argonaut RISC processor core as the SM2246EN, which is quite unique because we have seen many SSD controller vendors moving towards multi-core ARM architectures. A single custom core obviously brings efficiency gains and we've witnessed those in the SM2246EN, but the downside of such limited CPU power is sustained performance when the controller has to perform garbage collection at the same time as processing host IOs.

The only dramatic change is in the error correction circuitry as the SM2256 supports Low Density Parity Check (LDPC) error correction codes instead of more common and less powerful BCH ECC. Silicon Motion calls its ECC technology as NANDXtend, and it's a combination of LDPC hard and soft decode along with RAID5-like data recovery. The benefit of having three levels of ECC is performance because LDPC soft decode and recovery from parity both have a relatively noticeable impact on performance and are typically only needed when the drive approaches its end of life (i.e. when the NAND has been cycled a lot). Uncycled NAND has much higher reliability because the tunnel oxide hasn't worn out due to P/E cycles, so only very little ECC is needed and LDPC hard decode is sufficient and also doesn't have a dramatic impact on performance.

The reason why hard decode is faster than soft decode lies in how the voltage of a cell is sensed. Hard sensing is binary based, so for an SLC cell like in the graph above the cell can be either 1 or 0. However, as you can see, the voltage threshold distributions overlap slightly and that's actually far worse with MLC and TLC since there are more voltage states. In soft sensing the voltage distributions are divided into several segments, which requires more precision and iterations. For example in segment 4 the bit value can be either 1 or 0 as the distributions overlap, so probability algorithms are used to figure out the correct value. To be honest, ECC codes and the way they work are way over my head, but in case you are familiar with ECC and want to learn more, I suggest you simply google LDPC as there are numerous publicly available academic papers that go into more depth about this topic.

Silicon Motion claims that its NANDXtend technology can extend the endurance of TLC NAND by up to three times, making TLC more robust for heavier workloads and also allowing the use of lower quality NAND that some OEMs may use anyway due to the lack of in-house binning equipment. Unfortunately I didn't have any time to do extended endurance testing with the SM2256 yet to validate Silicon Motion's claims, but I will be sure to test that once we have a retail drive on our hands,

 

[BSim500]

Assuming this problem with planar TLC is able to be fixed in the same way it was with the 840 EVO (which involves re-writing the old data to make it fresh again) how much do we know about planar TLC that is left unpowered? (eg, laptop that hasn't been used for a few months or even a year).

Not much, and for me that's the problem. Anything "CPU" or "GPU" gets dissected in minute detail by over a dozen of the usual tech sites. Yet the inability / unwillingness of literally any of the same tech sites to do even basic data retention or longer term performance tests (not just SSD's but USB sticks too) leaves a "deafening silence"...

I also don't see monthly data rewriting as a "fix" that actually solves the problem. At best it's a cheap workaround that hides it by massively reducing (already lower) real-world available P/E cycles (first by having to use an SLC cache to hide TLC's lower native performance whilst larger 500GB-1TB MLC drives write directly in one go), then having to constantly monitor the age of and then rewrite older TLC data due to having far lower overhead between 8 voltage states vs 4). This further widens the MLC vs TLC endurance gap far disproportionately to the barely 20% price difference. Even Anandtech's review technique of constantly writing over a period of days until life left in % drops by 1% then calculating "predicted endurance" hides the issue. And yet again, they don't even touch data retention, powered or unpowered.

Best case : Unpowered data retention isn't an issue. But I'll be more happy when I actually see someone take half a dozen TLC drives, image them with a load of CRC checksummed files, put them in a drawer for 6 months then test both speed & data accuracy. Then wipe clean and repeat with 12 months. SSD's are one of the reasons I've lost all faith in the common sense of modern tech review sites and their "storage experts" seeming unwillingness to do any serious testing beyond copy & pasting PR snippets from the manufacturer.

 

[Rebel_L]

Not much, and for me that's the problem. Anything "CPU" or "GPU" gets dissected in minute detail by over a dozen of the usual tech sites. Yet the inability / unwillingness of literally any of the same tech sites to do even basic data retention or longer term performance tests (not just SSD's but USB sticks too) leaves a "deafening silence"...

I also don't see monthly data rewriting as a "fix" that actually solves the problem. At best it's a cheap workaround that hides it by massively reducing (already lower) real-world available P/E cycles (first by having to use an SLC cache to hide TLC's lower native performance whilst larger 500GB-1TB MLC drives write directly in one go), then having to constantly monitor the age of and then rewrite older TLC data due to having far lower overhead between 8 voltage states vs 4). This further widens the MLC vs TLC endurance gap far disproportionately to the barely 20% price difference. Even Anandtech's review technique of constantly writing over a period of days until life left in % drops by 1% then calculating "predicted endurance" hides the issue. And yet again, they don't even touch data retention, powered or unpowered.

Best case : Unpowered data retention isn't an issue. But I'll be more happy when I actually see someone take half a dozen TLC drives, image them with a load of CRC checksummed files, put them in a drawer for 6 months then test both speed & data accuracy. Then wipe clean and repeat with 12 months. SSD's are one of the reasons I've lost all faith in the common sense of modern tech review sites and their "storage experts" seeming unwillingness to do any serious testing beyond copy & pasting PR snippets from the manufacturer.

While those tests would be really nice, how do you expect review sites to actually do tests like that. If they are not done along side the development of the product itself your talking about review sites pretty much doing reviews on obsolete products. If you can pull it apart and study it that's easy for a review. If you get high end tech product and want to run year + long tests your going to be publishing those reviews at a point when your readers may not even be able to purchase those products anymore. Its same reason you rarely see video card reviews of older cards with updated drivers towards the end of their life cycles; and that kind of review doesn't even take a year of setting up to get finished, just to re-do tests at a later date.

 

[BSim500]

While those tests would be really nice, how do you expect review sites to actually do tests like that.

I'm not asking for 5 year tests or even the largest capacity of each one, just a more useful 6-12 months test on small 256GB drives given that's the time-scale we saw on the 840's problems. After all, the 850 EVO came out in Dec 2014 and a 6 month test would have been completed 10 months ago so it's hardly as if you'll only know the results after the drive is obsolete. It doesn't have to be done for each SSD either, just give people a general idea of real-world 40nm vs 20nm vs 16nm process retention expectation beyond marketing fluff and an endless string of "probably's".

Its same reason you rarely see video card reviews of older cards with updated drivers towards the end of their life cycles; and that kind of review doesn't even take a year of setting up to get finished, just to re-do tests at a later date.

True though it's precisely those kinds of articles that receive wide praise in the comments section for doing unusual stuff (and are more likely for me to click on them).

 

[Rebel_L]

I'm not asking for 5 year tests or even the largest capacity of each one, just a more useful 6-12 months test on small 256GB drives given that's the time-scale we saw on the 840's problems. After all, the 850 EVO came out in Dec 2014 and a 6 month test would have been completed 10 months ago so it's hardly as if you'll only know the results after the drive is obsolete. It doesn't have to be done for each SSD either, just give people a general idea of real-world 40nm vs 20nm vs 16nm process retention expectation beyond marketing fluff and an endless string of "probably's".


True though it's precisely those kinds of articles that receive wide praise in the comments section for doing unusual stuff (and are more likely for me to click on them).

For HD's I suppose they do tend to stick around a little longer, although I still suspect tests of that nature don't really drive site traffic like review's of the latest and greatest. I agree it would be nice to see some more tests of the sort (usb sticks seem especially feasible since they are so cheap that it wouldn't cost them too much to actually buy the hardware in numbers to do some fun testing).

I also wonder if the interest for reliability testing becomes too small a subset of the enthusiast market. Speaking from my own habits during my enthusiast pc building days I would never have kept any hardware long enough to see most kinds of reliability of premature failure problems. The mem/hd forum is also far less active than say the gpu or cpu forum.

Has anand ever talked about what kind of hardware budgets they have vs how much they rely on review samples and such for their testwork? Would buying they extra components do something like this regularly adversely affect their overall budget or would it be easy to absorb?

 

[PliotronX]

I'm just wondering at this point when QLC will hit the scene with 5 total host write lifetimes of the cells...

 

[Glaring_Mistake]

Not much, and for me that's the problem. Anything "CPU" or "GPU" gets dissected in minute detail by over a dozen of the usual tech sites. Yet the inability / unwillingness of literally any of the same tech sites to do even basic data retention or longer term performance tests (not just SSD's but USB sticks too) leaves a "deafening silence"...

Best case : Unpowered data retention isn't an issue. But I'll be more happy when I actually see someone take half a dozen TLC drives, image them with a load of CRC checksummed files, put them in a drawer for 6 months then test both speed & data accuracy. Then wipe clean and repeat with 12 months.

NordicHardware is testing the 840/850 EVO and the SanDisk Ultra II for read speed issues (and they were at least considering adding the SP550 to the test too).
But it has already been a year and there hasn't been a progress update in a very long time.

Otherwise I don't know of any tech site doing that kind of testing.

And I may test a number of SSDs (and create hashfiles in order to test for unseen corruption of course) but I do not leave them unpowered for that long since I base that on the number of write cycles.
With one month per 1000 write cycles or thereabout at least.
Which means that the only one having to go six months unpowered would be the 850 Pro, not the most likely candidate to suffer from data corruption.

Though I am actually performing tests on exactly half a dozen SSDs using TLC NAND.

We'll see if I perhaps do some more long term testing later but right now that is how I conduct my tests.

 

[ShintaiDK]

When people buy TLC drives, you just cant compete with MLC drives.

Samsung knew that when they started with TLC.

Classic race to the bottom example.