Happens more and more , because of cache structures and aimed for regular consumers the worse case scenario would show underwhelming numbers. See SSD like sandisk Ultra II .
I don't agree with it but sometimes you can only get review estimates.
I did when an old OCZ refused to take anymore writes and just outputted a bunch of errors. I was able to recover most of the data, but it's the main reason I've stuck with Samsung for the last 3-4 years.
My point is that SSD endurance isn't a problem for modern SSDs, in this case, we're talking about new 3D NAND SSD offerings branded by Intel. By your omission, the SSD which "refused to take anymore writes" is an OCZ SSD (widely reputed to be an awful distributor of SSDs ~5 years ago, declared bankruptcy in 2013, and they then got bought out by Toshiba), and as a result you've stuck with Samsung for the last 3-4 years.
Therefore, you bought an OCZ SSD back when they had lots of quality issues prior to their bankruptcy (older than the last 3-4 years), experienced one of these quality issues, and have stuck with Samsung drives since then, with presumably no issues.
Secondly, SSD endurance is typically measured in full drive writes per day (DWPD), and typical consumer usage of SSDs would be hard pressed to even have a single full drive full of write operations over the span of an entire month.
Thirdly, even beyond the normal spec of DWPD, most drives have specs that far underestimate their current capabilities of endurance. NAND flash on even TLC drives lasted over a PetaByte each (hammered by mixed I/O workloads 24/7 for over year) for some drives in TechPowerUp's SSD Endurance Test articles, far beyond any warranty and kept working until the SSDs eventually bricked themselves.
To be blunt, you could be dumb about wasting endurance by scheduling 3 completely unnecessary (and stupid) defragmentations to your SSD every day, and I'd be willing to bet that the SSD would still be fine after 5 years.
These days, typical consumers with typical consumer workloads shouldn't give a squat about SSD endurance metrics, as the SSD is far more likely to outlast its useful lifespan. 5 years from now, you won't care about a 128GB SATA 3 SSD bootdrive, as 512GB SSDs that are faster will be available for cheap, and who knows if manufacturers will still be bothering to include SATA 3 ports on motherboards 10 years from now. SATA is going the way of the IDE port and will disappear soon enough, but chances are, SATA 3 SSDs will keep ticking along, regardless of any "endurance" metrics you may be concerned about.
By no coincidence, the only SSD's I've ever seen "lock" themselves were Sandforce-based from OCZ (Vertex 2/3?) and ADATA (S510) which either lend to it being a bad firmware bug (most SF2281 firmware wasn't really stable until it was antiquated) or just two manufactures notorious for using shoddy binning practices. Most of the OCZ NAND should have ended up in USB flash drives, not performance SSD's.
I wouldn't say it's *never* a problem - my two-year-old, 120GB SanDisk Ultra II is almost halfway to its rated write lifetime without any stupid workloads, and I don't expect to replace it for any other reason before it reaches that.
Afterwards, well, I'll have to see how it behaves. Hopefully the spec is an underestimate, as most tests seem to find.
Endurance does matter. In the datacenter and enterprise world, only certain SSD's qualify for hard compatiabilty list's if they have certain DWPD's. Database applications and other applications that are write intensive can burn through a consumer rated SSD.
I find endurance interesting, because I support environments that need hundreds of MB/s of constant write speeds combined with multi terabyte storage requirement and read speeds must remain consistently in excess of multiple GBs/second.
We currently use ramdisks to satisfy the write endurance and performance requirements, but I'm seeing that modern pci-e SSD's should hit our performance requirements easily. However, endurance department doesn't seem so promising. I'm certainly disappointed with P3520 write endurance.
1500TB rated endurance is only 180days * 24h/day * 60min/h * 60sec/min * 100MB/sec with constant.
Even at very low 10MB/s write rate it is only supposed to last 5 years before hitting the endurance limit. Sure, for home users this is quite hard core, but for anything professional it is not!
I'm quite excited about the costs savings that might be realised with xpoint. Not compared to SSD, but to compared to ramdisks, especially combined with the non-volatility. Many things might become financially viable, because sustained heavy writes and having them reliably on disk could become much cheaper / easier.
Because if they did, they would have trouble selling the enterprise line. I suspect that plenty of startups or other cost-dutting companies build some pretty big RAID systems out of consumer gear.
RAID originally meant "inexpensive". It might again with consumer SSD drives, but make sure you understand your workload before trying this.
Unfortunately the 512GB is 189$. It's still ok but seq write and random perf are a bit weak for to make it great. Perf does decrease with capacity. at 512 GB you get 1,775MB/s read, 560MB/s write; 128,500 IOPS random read, 128,000 IOPS random write. 256 and 128GB become rather slow , especially the 128GB.
Nope. 3D XPoint isn't flash memory, and isn't anywhere close to being ready for use in such a broad range of products. These drives are using IMFT's 3D NAND flash memory, roughly comparable to the 3D NAND used by Samsung in the 850 and 950 SSDs and in development by the other NAND manufacturers, although there are some unique things about the approach Intel and Micron have taken, such as by sticking with a floating gate memory cell instead of switching to a charge trap design.
It would be nice to see the DRAM replaced with 3d xpoint, but that will be awhile (and hopefully irrelevant in 3d flash. Supposedly using a 3d TLC trench as a single SLC might as well have the endurance and speed of "real" SLC so lets you get the best of both worlds as a cache/buffer).
Wake me up when a memory load instruction returns a value from flash/3dxpoint without software assistance.
I've already made it a habit to use the full terms "3D NAND flash memory" and "3D XPoint memory" where appropriate. In future coverage that is related to 3D XPoint I will try to more clearly emphasize that 3D XPoint memory is not 3D NAND flash memory because it is not any kind of flash memory, but I don't want to bring 3D XPoint in to every discussion of 3D anything.
FYI, IMFT does not make 3D NAND. ALL of this NAND is from Micron and purchased by Intel. Products produced by IMFT come from Lehi, Utah only which are planar NAND and 3D XPoint. As mentioned in the very 1st paragraph, Intel opted out of the IMFT agreement in Singapore several years ago where all of the 3D NAND is made for Micron.
"... we now will utilize Intel’s facility in Dalian, China to help expand our manufacturing capacity in non-volatile memory. The expansion is part of our global multi-source supply strategy and will allow us to best serve our customers. We expect initial production of the 3D NAND technology in Dalian in the second half of 2016."
I don't think that is correct. If the 3D NAND isn't from IMFT, then why did Intel do a presentation about the 3D NAND (2 years ago at Investor Meeting, http://intelstudios.edgesuite.net/im/2014/pdf/2014... from page 44)?
“Intel expects China to consume nearly a third of the NAND chip market in 2016, and as a result is working to play a key role in helping China to build out this industry,” Technology Business Research (TBR) analyst Krista Macomber said in a report on Friday. The company jointly developed its 3D NAND and 3D XPoint technologies with another major US chipmaker, Micron Technology.
This in the first paragraph refers to final planar nand node, not 3D.
"Intel's relatively expensive 20nm flash or on 16nm flash that Intel had to buy on the open market due to their decision to not participate in the 16nm node at IMFT."
So, as a practical matter to most of us here, what is the approximate pricing? The truth is that, these days, most all SSDs are good enough for what most of us need. A difference in performance of 10-20% really isn't that serious.
600p pricing looks interesting, it's closer to that of the better SATA drives but performance should be at least halfway between those and the better M.2 drives... So basically it'll help take M.2 more mainstream. Shame they aren't pushing U.2 harder in the consumer space... Seems far more ideal for enthusiasts and workstations (even tho I'm running an SM951 myself).
"The 600p will be going after the client PCIe storage market from the opposite end: as one of the first TLC PCIe SSDs, its performance specifications don't set any records but it will be a much more value-oriented product than any of the M.2 PCIe SSDs currently on the market. Intel has confirmed that the 600p and 6000p are using a third-party controller, and it is most likely Marvell's 88SS1093."
Intel going for the value proposition while not using their own controller? Is it opposites day?
Good. To get wider adoption of PCI-E SSDs we need more volume. To get volume we'll need competition. To get that we'll need some non-premium options. Samsung's offerings are great. Excellent performers. Excellent value even, for the SM961. But some need lower price points, and for now that the only options right now are SATA which is a dead end for performance.
At least in some metrics these new 3D TLC PCI-E drives should outperform even the best SATA drives. I'm excited.
Thanks for the tip. Intel's pics didn't look like the metal packaging we've been seeing lately for SM2260, but that forum picture looks convincing even without the usual SMI logo.
Ummm the endurance of these devices is impressive, five times Samsung Pro SSDs (ten times in some models). There is a 10/20% lack of performance but in enterprise the endurance (and price) matters much. I think they will be very popular in upcoming months.
You gotta love how Intel marketing sprinkles their IoT piss over everything they come close to. More interestingly they're entirely missing the point of what the word embedded means... Why can't they simply stay with the term "Industrial", as they used to do...
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
48 Comments
Back to Article
zepi - Thursday, August 25, 2016 - link
How come endurance is not mentioned for many of the products?plopke - Thursday, August 25, 2016 - link
Happens more and more , because of cache structures and aimed for regular consumers the worse case scenario would show underwhelming numbers. See SSD like sandisk Ultra II .I don't agree with it but sometimes you can only get review estimates.
Billy Tallis - Thursday, August 25, 2016 - link
Intel only provided the product briefs prior to the announcement. The full spec sheets and ark.intel.com data aren't online yet.JoeyJoJo123 - Thursday, August 25, 2016 - link
And why is it that you care?Have you personally _ever_ lost any data due to an SSD's endurance being too poor? SSD endurance "mattering" is a meme.
Paul Tarnowski - Thursday, August 25, 2016 - link
I did when an old OCZ refused to take anymore writes and just outputted a bunch of errors. I was able to recover most of the data, but it's the main reason I've stuck with Samsung for the last 3-4 years.JoeyJoJo123 - Thursday, August 25, 2016 - link
You essentially just proved my point:My point is that SSD endurance isn't a problem for modern SSDs, in this case, we're talking about new 3D NAND SSD offerings branded by Intel. By your omission, the SSD which "refused to take anymore writes" is an OCZ SSD (widely reputed to be an awful distributor of SSDs ~5 years ago, declared bankruptcy in 2013, and they then got bought out by Toshiba), and as a result you've stuck with Samsung for the last 3-4 years.
Therefore, you bought an OCZ SSD back when they had lots of quality issues prior to their bankruptcy (older than the last 3-4 years), experienced one of these quality issues, and have stuck with Samsung drives since then, with presumably no issues.
Secondly, SSD endurance is typically measured in full drive writes per day (DWPD), and typical consumer usage of SSDs would be hard pressed to even have a single full drive full of write operations over the span of an entire month.
Thirdly, even beyond the normal spec of DWPD, most drives have specs that far underestimate their current capabilities of endurance. NAND flash on even TLC drives lasted over a PetaByte each (hammered by mixed I/O workloads 24/7 for over year) for some drives in TechPowerUp's SSD Endurance Test articles, far beyond any warranty and kept working until the SSDs eventually bricked themselves.
http://techreport.com/review/27909/the-ssd-enduran...
To be blunt, you could be dumb about wasting endurance by scheduling 3 completely unnecessary (and stupid) defragmentations to your SSD every day, and I'd be willing to bet that the SSD would still be fine after 5 years.
These days, typical consumers with typical consumer workloads shouldn't give a squat about SSD endurance metrics, as the SSD is far more likely to outlast its useful lifespan. 5 years from now, you won't care about a 128GB SATA 3 SSD bootdrive, as 512GB SSDs that are faster will be available for cheap, and who knows if manufacturers will still be bothering to include SATA 3 ports on motherboards 10 years from now. SATA is going the way of the IDE port and will disappear soon enough, but chances are, SATA 3 SSDs will keep ticking along, regardless of any "endurance" metrics you may be concerned about.
Samus - Thursday, August 25, 2016 - link
By no coincidence, the only SSD's I've ever seen "lock" themselves were Sandforce-based from OCZ (Vertex 2/3?) and ADATA (S510) which either lend to it being a bad firmware bug (most SF2281 firmware wasn't really stable until it was antiquated) or just two manufactures notorious for using shoddy binning practices. Most of the OCZ NAND should have ended up in USB flash drives, not performance SSD's.FLHerne - Thursday, August 25, 2016 - link
I wouldn't say it's *never* a problem - my two-year-old, 120GB SanDisk Ultra II is almost halfway to its rated write lifetime without any stupid workloads, and I don't expect to replace it for any other reason before it reaches that.Afterwards, well, I'll have to see how it behaves. Hopefully the spec is an underestimate, as most tests seem to find.
mdw9604 - Tuesday, August 30, 2016 - link
Endurance does matter. In the datacenter and enterprise world, only certain SSD's qualify for hard compatiabilty list's if they have certain DWPD's. Database applications and other applications that are write intensive can burn through a consumer rated SSD.bug77 - Friday, August 26, 2016 - link
Because endurance gets worse with every node shrink. 3D NAND mitigates some of that, but it's useful to have some numbers available.MrSpadge - Friday, August 26, 2016 - link
Because those are partly Enterprise drives, where endurance can matter a lot. One has to know it to judge for which workload the drives can be used.zepi - Tuesday, August 30, 2016 - link
600p and 6000p have endurance ratings of 72TBW in intel specs:http://ark.intel.com/products/94924/Intel-SSD-600p...
All drives regardless of size share this spec.
DC P3520 seems to have about 0.8TDWP or 1480TBW for 1.2TB model:
http://ark.intel.com/products/88722/Intel-SSD-DC-P...
zepi - Tuesday, August 30, 2016 - link
I find endurance interesting, because I support environments that need hundreds of MB/s of constant write speeds combined with multi terabyte storage requirement and read speeds must remain consistently in excess of multiple GBs/second.We currently use ramdisks to satisfy the write endurance and performance requirements, but I'm seeing that modern pci-e SSD's should hit our performance requirements easily. However, endurance department doesn't seem so promising. I'm certainly disappointed with P3520 write endurance.
1500TB rated endurance is only 180days * 24h/day * 60min/h * 60sec/min * 100MB/sec with constant.
Even at very low 10MB/s write rate it is only supposed to last 5 years before hitting the endurance limit. Sure, for home users this is quite hard core, but for anything professional it is not!
I'm quite excited about the costs savings that might be realised with xpoint. Not compared to SSD, but to compared to ramdisks, especially combined with the non-volatility. Many things might become financially viable, because sustained heavy writes and having them reliably on disk could become much cheaper / easier.
wumpus - Wednesday, August 31, 2016 - link
Because if they did, they would have trouble selling the enterprise line. I suspect that plenty of startups or other cost-dutting companies build some pretty big RAID systems out of consumer gear.RAID originally meant "inexpensive". It might again with consumer SSD drives, but make sure you understand your workload before trying this.
jjj - Thursday, August 25, 2016 - link
The 600p is apparently listed din Taiwan at only 5199 NT (164$) for the 512GB model. At that price it would be a very nice offer.Samus - Thursday, August 25, 2016 - link
Wow, that would be an amazing price for an Intel drive.jjj - Friday, August 26, 2016 - link
Unfortunately the 512GB is 189$. It's still ok but seq write and random perf are a bit weak for to make it great. Perf does decrease with capacity. at 512 GB you get 1,775MB/s read, 560MB/s write; 128,500 IOPS random read, 128,000 IOPS random write. 256 and 128GB become rather slow , especially the 128GB.Mr Perfect - Thursday, August 25, 2016 - link
Just to be clear, when you say Intel's 3D flash, you're talking about XPoint?Billy Tallis - Thursday, August 25, 2016 - link
Nope. 3D XPoint isn't flash memory, and isn't anywhere close to being ready for use in such a broad range of products. These drives are using IMFT's 3D NAND flash memory, roughly comparable to the 3D NAND used by Samsung in the 850 and 950 SSDs and in development by the other NAND manufacturers, although there are some unique things about the approach Intel and Micron have taken, such as by sticking with a floating gate memory cell instead of switching to a charge trap design.wumpus - Thursday, August 25, 2016 - link
It would be nice to see the DRAM replaced with 3d xpoint, but that will be awhile (and hopefully irrelevant in 3d flash. Supposedly using a 3d TLC trench as a single SLC might as well have the endurance and speed of "real" SLC so lets you get the best of both worlds as a cache/buffer).Wake me up when a memory load instruction returns a value from flash/3dxpoint without software assistance.
Meteor2 - Thursday, August 25, 2016 - link
You should put that info in the article.Billy Tallis - Thursday, August 25, 2016 - link
I've already made it a habit to use the full terms "3D NAND flash memory" and "3D XPoint memory" where appropriate. In future coverage that is related to 3D XPoint I will try to more clearly emphasize that 3D XPoint memory is not 3D NAND flash memory because it is not any kind of flash memory, but I don't want to bring 3D XPoint in to every discussion of 3D anything.mkozakewich - Thursday, August 25, 2016 - link
"This thing's a 3D printer, huh? That means it prints 3D XPoint Memory?"Mr Perfect - Thursday, August 25, 2016 - link
Okay, thanks for clarifying. For some reason I though Intel had given traditional 3D Flash a miss and gone straight to XPoint.Arnulf - Thursday, August 25, 2016 - link
No, they are talking about 3D flash, hence "3d flash".mmonnin03 - Thursday, August 25, 2016 - link
FYI, IMFT does not make 3D NAND. ALL of this NAND is from Micron and purchased by Intel. Products produced by IMFT come from Lehi, Utah only which are planar NAND and 3D XPoint. As mentioned in the very 1st paragraph, Intel opted out of the IMFT agreement in Singapore several years ago where all of the 3D NAND is made for Micron.woggs - Thursday, August 25, 2016 - link
http://www.theregister.co.uk/2015/10/22/intel_5bn_..."... we now will utilize Intel’s facility in Dalian, China to help expand our manufacturing capacity in non-volatile memory. The expansion is part of our global multi-source supply strategy and will allow us to best serve our customers. We expect initial production of the 3D NAND technology in Dalian in the second half of 2016."
mmonnin03 - Thursday, August 25, 2016 - link
Intel, Micron and IMFT are 3 separate companies. Dalian, Fab 68, is an Intel fab.witeken - Thursday, August 25, 2016 - link
I don't think that is correct. If the 3D NAND isn't from IMFT, then why did Intel do a presentation about the 3D NAND (2 years ago at Investor Meeting, http://intelstudios.edgesuite.net/im/2014/pdf/2014... from page 44)?woggs - Thursday, August 25, 2016 - link
3D nand is absolutely IMFT developed and Intel is sourcing it via IMFT and directly from Dalian fab 68..woggs - Thursday, August 25, 2016 - link
http://www.scmp.com/tech/enterprises/article/19017...“Intel expects China to consume nearly a third of the NAND chip market in 2016, and as a result is working to play a key role in helping China to build out this industry,” Technology Business Research (TBR) analyst Krista Macomber said in a report on Friday.
The company jointly developed its 3D NAND and 3D XPoint technologies with another major US chipmaker, Micron Technology.
woggs - Thursday, August 25, 2016 - link
This in the first paragraph refers to final planar nand node, not 3D."Intel's relatively expensive 20nm flash or on 16nm flash that Intel had to buy on the open market due to their decision to not participate in the 16nm node at IMFT."
melgross - Thursday, August 25, 2016 - link
So, as a practical matter to most of us here, what is the approximate pricing? The truth is that, these days, most all SSDs are good enough for what most of us need. A difference in performance of 10-20% really isn't that serious.melgross - Thursday, August 25, 2016 - link
Wow! Right after I posted my pricing question, I went to Computerworld, the next bookmark in my folder for computing, and found this:http://www.computerworld.com/article/3112635/data-...
Impulses - Thursday, August 25, 2016 - link
600p pricing looks interesting, it's closer to that of the better SATA drives but performance should be at least halfway between those and the better M.2 drives... So basically it'll help take M.2 more mainstream. Shame they aren't pushing U.2 harder in the consumer space... Seems far more ideal for enthusiasts and workstations (even tho I'm running an SM951 myself).ZeDestructor - Thursday, August 25, 2016 - link
It's all about the laptopsThe_Assimilator - Thursday, August 25, 2016 - link
"The 600p will be going after the client PCIe storage market from the opposite end: as one of the first TLC PCIe SSDs, its performance specifications don't set any records but it will be a much more value-oriented product than any of the M.2 PCIe SSDs currently on the market. Intel has confirmed that the 600p and 6000p are using a third-party controller, and it is most likely Marvell's 88SS1093."Intel going for the value proposition while not using their own controller? Is it opposites day?
MrCommunistGen - Thursday, August 25, 2016 - link
Good. To get wider adoption of PCI-E SSDs we need more volume. To get volume we'll need competition. To get that we'll need some non-premium options.Samsung's offerings are great. Excellent performers. Excellent value even, for the SM961. But some need lower price points, and for now that the only options right now are SATA which is a dead end for performance.
At least in some metrics these new 3D TLC PCI-E drives should outperform even the best SATA drives. I'm excited.
malventano - Thursday, August 25, 2016 - link
Billy - 600p is a Silicon Motion controller. More info here: http://www.pcper.com/news/Storage/Intel-Revises-Al...Billy Tallis - Thursday, August 25, 2016 - link
Thanks for the tip. Intel's pics didn't look like the metal packaging we've been seeing lately for SM2260, but that forum picture looks convincing even without the usual SMI logo.Gondalf - Friday, August 26, 2016 - link
Ummm the endurance of these devices is impressive, five times Samsung Pro SSDs (ten times in some models). There is a 10/20% lack of performance but in enterprise the endurance (and price) matters much.I think they will be very popular in upcoming months.
Daniel Egger - Friday, August 26, 2016 - link
You gotta love how Intel marketing sprinkles their IoT piss over everything they come close to. More interestingly they're entirely missing the point of what the word embedded means... Why can't they simply stay with the term "Industrial", as they used to do...TemjinGold - Friday, August 26, 2016 - link
600p drives up to 512 gb are now on sale at Newegg it seems. Too bad they don't have the 1TB.rhysiam - Monday, August 29, 2016 - link
That's strange the 1TB models aren't on Newegg. They're listed on Intel ARK (http://ark.intel.com/products/94926/Intel-SSD-600p... Plus as I write this they're available and in stock in Australia: https://www.pccasegear.com/products/36615/intel-60...Gasaraki88 - Friday, August 26, 2016 - link
Not impressed. Samsung 950Pro has been out for awhile now and still has better performance.Rayb - Saturday, August 27, 2016 - link
And you're still paying 85% < 95% premium over the cost of this drive.http://www.newegg.com/Product/Product.aspx?Item=9S...