Samsung ramps up 64-layer V-NAND flash production and prepares new memory products for the second half of 2017.
Samsung Increases 64-Layer V-NAND Production : Read more
Samsung Increases 64-Layer V-NAND Production : Read more
- Status
dudmont
Reputable
- Feb 23, 2015
- 1,404
- 0
- 5,660
I'm wondering if you could add some more info/thoughts, in regards to your article/post lastweek about spinning disks and mlc nand SSDs, as it pertains to Samsung and it's flash. More specifically, whether you think Samsung SSDs, don't fall into the same performance concerns that you discussed.
PC-Cobbler
Distinguished
- Jun 12, 2015
- 47
- 0
- 18,530
You're not the only one worried about the future of high-performance SSDs. I won't buy TLC, let alone QLC. Samsung will garner all of the business, both consumer and enterprise, for improved performance and reliability, while Intel, Micron, SK hynix, Toshiba, WDC, and the rest can compete for the bottom feeders.
I'm not too concerned with Samsung. 3rd Gen V-NAND sits with the other fabs MLC. The new 4th Gen is even faster (as we talked about in this piece). Samsung knows how to handle their business and understands the performance market.
In contrast look at what we've received from Crucial, Intel, Sk hynix and Toshiba/OCZ over the last couple of years. Other than the RD400 and SSD 750 Series everything else was mainstream at best. Both of those products used MLC and for the most part that's gone going forward.
Other high-performance MLC products like the MyDigitalSSD BPX (and other Phison E7-based products) used Toshiba planar MLC. That, too, will go away fairly quickly. The E7 isn't compatible with Toshiba BiCS FLASH. E8 is but only uses a PCIe 3.0 x2 connection to the host but is one of the very few quad-core flash processors left.
Marvell has the 88SS1094 controller but is only proven with Toshiba planar MLC. I've tested it with planar TLC but I wouldn't call it high performance (review coming soon).
What we're really left with is Samsung's current and next gen products and Optane SSD. I expect both of those to remain expensive even when the other companies start to drop retail prices as we come out of the shortage. The other option is Optane Memory with a HDD. That article will be up shortly. We've had a lot more time to experiment with Optane Memory and it's much better than I first thought.
In contrast look at what we've received from Crucial, Intel, Sk hynix and Toshiba/OCZ over the last couple of years. Other than the RD400 and SSD 750 Series everything else was mainstream at best. Both of those products used MLC and for the most part that's gone going forward.
Other high-performance MLC products like the MyDigitalSSD BPX (and other Phison E7-based products) used Toshiba planar MLC. That, too, will go away fairly quickly. The E7 isn't compatible with Toshiba BiCS FLASH. E8 is but only uses a PCIe 3.0 x2 connection to the host but is one of the very few quad-core flash processors left.
Marvell has the 88SS1094 controller but is only proven with Toshiba planar MLC. I've tested it with planar TLC but I wouldn't call it high performance (review coming soon).
What we're really left with is Samsung's current and next gen products and Optane SSD. I expect both of those to remain expensive even when the other companies start to drop retail prices as we come out of the shortage. The other option is Optane Memory with a HDD. That article will be up shortly. We've had a lot more time to experiment with Optane Memory and it's much better than I first thought.
dudmont
Reputable
- Feb 23, 2015
- 1,404
- 0
- 5,660
That's what it looks like. Would be nice if one of the controller designer/manufacturers would produce a really topflight performance oriented piece to give some competition. I use almost exclusively Samsung SSDs, but without someone pushing, incentive to improve doesn't exist(look at Intel v AMD for the last 7-8 years, until Ryzen).
Ram, thanks for replying.
Looks like we can have PCIE M.2s all we want, but that doesn't guarantee performance, even with PCIE 4 and 5 coming out in the not too distant future.
Rookie_MIB
Splendid
- Jul 27, 2013
- 1,965
- 10
- 21,865
I'm just wanting to see some of the pricing come down to reasonable levels. For a vast majority of the users, they want better reliability than spinny disks (already a given due to fewer moving parts), better speed (again, easy even for the slower SSD), and better capacity (which is still needing work).
Just give me a 2.5" drive, 3-4TB capacity, decent controller and speeds (300-400MB R & W) at a $200-250 price range and they'll fly off the shelves. That would be ideal for rebuilding large NAS systems with more reliable flash.
Just give me a 2.5" drive, 3-4TB capacity, decent controller and speeds (300-400MB R & W) at a $200-250 price range and they'll fly off the shelves. That would be ideal for rebuilding large NAS systems with more reliable flash.
I'm not sure I got all the pieces together...
The 48 layer is 3rd gen and this new 64 layer is 4th gen?
And they are all MLC?
I got an 850 NVMe PRO last fall just before the 860 announcement (I was searching Internet for 860 news but found none).
Shortly after I bought the 850, the 860 was announces, but I was happy still happy believing the 850 PRO was the last MLC.
But that was not the case then? Samsung rock on with MLC!? (I won't be unhappy if so! :-D)
The 48 layer is 3rd gen and this new 64 layer is 4th gen?
And they are all MLC?
I got an 850 NVMe PRO last fall just before the 860 announcement (I was searching Internet for 860 news but found none).
Shortly after I bought the 850, the 860 was announces, but I was happy still happy believing the 850 PRO was the last MLC.
But that was not the case then? Samsung rock on with MLC!? (I won't be unhappy if so! :-D)
IndignantSkeptic
Distinguished
- Apr 19, 2011
- 507
- 0
- 18,980
valeman2012
Distinguished
- Apr 10, 2012
- 1,272
- 11
- 19,315
A Safety Question - When they decided to release these, are they have some protection to prevent the SSD from exploding.
It's 32GB per die. There are a number of due (up to 16) per package and a handful of packages per SSD. The performance comes from reading and writing to several at a time.
What the big die from Micron (384Gbit) taught us is that without high capacity drives the performance is awful. Micron's 2nd gen TLC goes back to 256Gbit per die. I suspect they will back off from delivering 10TB consumer SSDs in the "near future". It was something that was said during the 1st gen 3D announcement.
We expected 2nd Gen IMFT 3D to double die density, 512Gbit for MLC and 768Gbit for TLC. It was more than an expectation, we were told as much from someone inside at the start of Gen 1. The block and page sizes would have increased and that could nearly double the latency on the erase cycle. You can see where that would be an issue after the reviews of products with 384Gbit die.
I'd have to say there is a reason why every other company went with charge trap technology rather than floating gates. It's been fun listening to engineers tell us about the problem every since IMFT announced floating gate 3D. Sk hynix started out developing floating gate 3D and scrapped it for charge trap tech.
BiCS 3 is the odd man out that we don't know too much about. We tested some late BiCS 2 flash some time ago and the performance wasn't that great. I've heard endurance was a an issue, too. I had a conversation about BiCS 3 the other day and was told to expect quite a bit more from it.
Sadly WD is going to keep Toshiba in court of the sale. It's kind of like a child throwing a temper tantrum in the store for not getting a new toy a week before Xmas. There was supposed to be an announcement in Japan yesterday and I stayed up all morning for it. The time came and went. I figured it would be right after the Nikkei closed, around 5AM my time. That was a waste. WD files an injunction in a CA court.
Most Japanese analysts expected Broadcom to win that bid. That is an interesting play because Broadcom is the front name now for Avago. Avago bought up nearly every tech company with a price tag over the last couple of years. In that group is LSI, SandForce (part of which was sold to Seagate), PLX, Brocade, the list goes on and on. Avago still makes controllers behind the scenes, see the recent Intel Broadcom SSD announcement that says nothing while saying everything. That relationship dates back to the LSI and early Intel SSD days. Imagine Broadcom with some of the best controllers and a flash fab to boot.
My enthusiasm goes away if WD gets full control of the fabs. What a lovely TLC-based Black Series SSD you have there. It's kind of like a Fiero. It looks fast but in the end you will just want to burn it down.
I really need to make a separate account to say some of this stuff.
What the big die from Micron (384Gbit) taught us is that without high capacity drives the performance is awful. Micron's 2nd gen TLC goes back to 256Gbit per die. I suspect they will back off from delivering 10TB consumer SSDs in the "near future". It was something that was said during the 1st gen 3D announcement.
We expected 2nd Gen IMFT 3D to double die density, 512Gbit for MLC and 768Gbit for TLC. It was more than an expectation, we were told as much from someone inside at the start of Gen 1. The block and page sizes would have increased and that could nearly double the latency on the erase cycle. You can see where that would be an issue after the reviews of products with 384Gbit die.
I'd have to say there is a reason why every other company went with charge trap technology rather than floating gates. It's been fun listening to engineers tell us about the problem every since IMFT announced floating gate 3D. Sk hynix started out developing floating gate 3D and scrapped it for charge trap tech.
BiCS 3 is the odd man out that we don't know too much about. We tested some late BiCS 2 flash some time ago and the performance wasn't that great. I've heard endurance was a an issue, too. I had a conversation about BiCS 3 the other day and was told to expect quite a bit more from it.
Sadly WD is going to keep Toshiba in court of the sale. It's kind of like a child throwing a temper tantrum in the store for not getting a new toy a week before Xmas. There was supposed to be an announcement in Japan yesterday and I stayed up all morning for it. The time came and went. I figured it would be right after the Nikkei closed, around 5AM my time. That was a waste. WD files an injunction in a CA court.
Most Japanese analysts expected Broadcom to win that bid. That is an interesting play because Broadcom is the front name now for Avago. Avago bought up nearly every tech company with a price tag over the last couple of years. In that group is LSI, SandForce (part of which was sold to Seagate), PLX, Brocade, the list goes on and on. Avago still makes controllers behind the scenes, see the recent Intel Broadcom SSD announcement that says nothing while saying everything. That relationship dates back to the LSI and early Intel SSD days. Imagine Broadcom with some of the best controllers and a flash fab to boot.
My enthusiasm goes away if WD gets full control of the fabs. What a lovely TLC-based Black Series SSD you have there. It's kind of like a Fiero. It looks fast but in the end you will just want to burn it down.
I really need to make a separate account to say some of this stuff.
Michael_687
Prominent
- Jun 16, 2017
- 1
- 0
- 510
I wonder how this is going to affect flash prices, and how soon can we expect flash prices to start going down again.
dudmont
Reputable
- Feb 23, 2015
- 1,404
- 0
- 5,660
Ram, you're seeing the divide right here, between users looking for a large cheap drive, and some of us who are more concerned that the performance side of the market will take a dive due to lack of players. You've given us a lot of info to chew on, thanks much.
My concern with optane is solely the lack of interface to really take advantage of it's potential. Until we get PCIE 4 and 5 drives going, PCIE 3 just appears to be saturated by even nand based drives(although I'm beginning to rethink this thought based on recent news). Latency advantages are nice, but the DMI pipe is only so big and thus seems to me to be the big bottleneck.
My concern with optane is solely the lack of interface to really take advantage of it's potential. Until we get PCIE 4 and 5 drives going, PCIE 3 just appears to be saturated by even nand based drives(although I'm beginning to rethink this thought based on recent news). Latency advantages are nice, but the DMI pipe is only so big and thus seems to me to be the big bottleneck.
I've seen two price models. One says supply will stay tight until Q1 2018 and then the bottom drops out in Q2. The other model shows supply easing up as soon as Toshiba gets BiCS 3 out and then the model drops out in Q1 2018. Either way, by this time next year SSDs will be "cheap" again. We just need the companies to get us high capacity drives at a low price. They will take what we will give, though.
I don't see PCIe 4.0 or 5.0 doing much for consumer workloads unless flash gets better. By the time we're using PCIe 5.0 the SSD industry will be beyond flash. Pick your flavor as far as what will win out. Optane is just a form of phase change memory and all of the NAND companies have worked on it as well as other new storage types. The technology exists already but they just need to make it hold more data. It's the same thing that happened to DRAM and NAND.
I got off track a bit. PCIe 4.0 will only increase sequential performance. We need lower latency NAND (like Samsung's 4th Gen) to get better low queue depth performance. 4th Gen is exciting. When was the last time a NAND fab announced the program latency in a press release? The other part that excites me about the new 4th Gen is Samsung's history of increasing QD1 random read performance. We saw it go from 10K to 12K to 14K and then 16K (16,616 so knocking on the 17K door). 970 EVO or whatever they announce in a few months should give us 18 to 19K if we use the trend line.
To put Optane into perspective, the Optane Memory module delivers 71K at QD1. As a cache device paired with a HDD it's still 47K.
I don't see PCIe 4.0 or 5.0 doing much for consumer workloads unless flash gets better. By the time we're using PCIe 5.0 the SSD industry will be beyond flash. Pick your flavor as far as what will win out. Optane is just a form of phase change memory and all of the NAND companies have worked on it as well as other new storage types. The technology exists already but they just need to make it hold more data. It's the same thing that happened to DRAM and NAND.
I got off track a bit. PCIe 4.0 will only increase sequential performance. We need lower latency NAND (like Samsung's 4th Gen) to get better low queue depth performance. 4th Gen is exciting. When was the last time a NAND fab announced the program latency in a press release? The other part that excites me about the new 4th Gen is Samsung's history of increasing QD1 random read performance. We saw it go from 10K to 12K to 14K and then 16K (16,616 so knocking on the 17K door). 970 EVO or whatever they announce in a few months should give us 18 to 19K if we use the trend line.
To put Optane into perspective, the Optane Memory module delivers 71K at QD1. As a cache device paired with a HDD it's still 47K.
- Status
TRENDING THREADS
-
-
Question I'm looking for a temporary new board after mine died ?
- Started by chucky9
- Replies: 3
-
News Intel CEO Pat Gelsinger retires, effective immediately — also steps down from BOD, two co-CEOs step in- Started by Admin
- Replies: 124
-
Discussion What's your favourite video game you've been playing?- Started by amdfangirl
- Replies: 4K
-
Facebook
Twitter
Instagram