Amen. Too bad V-Nand UFS didn't make it to the GS7... Oh well, till my next upgrade. They'll have something more interesting by then, hopefully running Windows 10 Mobile.
I thought exactly the same, until they announced the TabPro S... Hopefully the "Pro" moniker carries on to their flagship smartphones where "pro" simply means Windows 10. Galaxy S8 Pro sounds good, and not too far fetched. Oh, and hopefully running on Exynos.
Nothing is stopping SoC manufacturers from slapping in support for future products. But if there *is* an issue, it would imagine it being power envelope. The 950 Pro lags behind AHCI in idle. It might be more efficient for tablets, but possibly not yet so for smartphones.
UFS (current and next gen) is a more viable solution for the next couple of years until progress hits the sweetspot power envelope for BGA SSDs on flagship smartphones (I'd imagine these would be solely reserved for flagships). That's when UFS will be passed down to midrange and low end while EMMC is gradually phased out.
The 6s and Galaxy 6+ use the same physical layer called m-phy (a mipi standard). The difference is in the higher layers. Apple use the m-pcie standard, which only keeps the pcie upper layers (dll and TL). Over that they use the nvme protocol for their controller. Samsung uses the same m-phy layer but, I believe, uses unipro signaling in lieu of pcie, and ufs at the application layer. Imho, Apple made the better call as to the application layers.
I can barely understand anything of all this M.2 stuff . Trying , but I need to look at AHCI/NVMe , PCIe2/4/sata and all that is just branded M.2. So how would BGA effect me as a regular consumer?
Urgh I wish they would take more of a USB/sata approach , which isn't perfect but something like M.2(AHCI/sata) M.3(NVMe/PCIe2), M.4(NVMe,PCIe4). Just something so the devices says compatible with M.2,M.3 and M.4 , because so many times if I wanne buy something on a website the specification is wrong/missing or it contains a Asteriks on their site stating depends on your region.
AHCI supports one queue with a depth of 32 requests. This is hardly enough to keep a modern SSD busy. NVMe, which M.2 implements, supports 65,535 queues each with a depth of 65,535 requests.
There is no point in making these devices backwards compatible with AHCI/SATA. The whole issues is AHCI/SATA was too slow. It's like saying we need to make jet engines compatible with prop airplane design. In order to go faster, we need to break the mold.
M.2 has a dozen possible keyings (A-M) and 11 possible slotted form factors (see diagram in article), and 8 possible thicknesses (urgh); it's intended primarily as an OEM interface not a consumer one and has way too many moving parts to get everything marked out like that even if was just an SSD spec. As a simplifying factor a lot of that seems to be spelling out every possible option now instead of trying to squeeze them in after the fact.
Only 5 of the keyings are defined and 3 actually used for current mainstreamish products (A - wifi, B - Sata/2x PCIe/sata SSDs, and M - 4x PCIe/sata SSDs). The SSD situation is settled a bit more by most systems using B/M slots. Long term x4 PCIe means that M's going to replace B for SSDs. PCIe is forward/backward compatible; you potentially leave performance on the table but if they fit they should work.
Sizes are defined by dimensions in mm and again only use a fraction of the available options. The 30mm wide ones are no-shows (I assume this was a sop to previous larger mobile card spec vendors so they could pretend not to need to change up as much all at once), and I've not seen anything using the ultra narrow 16mm form factor yet. Everything uses the 22mm wide versions: Wifi the smallest 22x30 size, SSDs can use the 22x42/60/80/110 sizes but AFAIK most are either 2242 or 2280 with 2/4 flash chips on them. 22110 would allow for 8 flash chip modules like in 2.5" form factor drives; but even desktop mobos tend not to free up enough space to fit one. At this point I suspect it's dead in the water; and hope U.2 (a 4xPCIe cable) will actually ship allowing for high performance 2.5" SSDs. PCIe m.2 ones currently have most of the speed titles due to larger bandwidth; but the smaller number of flash chips limits their capacity and in some cases bottlenecks them.
Great info! I'd happily read an article about this sort of thing, since these sort-of-internal interfaces in smaller devices aren't always obvious (nor is what is compatible with what).
I also wish they'd find a ZIF type socket that gives the smallness of BGA without being actually soldered down but I guess that's impossible or someone would have figured it out by now...
Probably not going to happen. You can space balls more tightly than holes in a socket. Intel's LGA socket probably comes closest in connection density; but the locking lever needed to hold the chip down tightly enough for good contact is hardly zero force.
@DanNeely , thx for the short summary ! Hope i can remmember it all if i ever upgrade a laptop or something and hope U.2 gets standard on my desktops :)
To be precise, M.2 and U.2 are physical connector and board space specifications. M.2 is the specification for mostly mobile, U.2 is a superset of a SAS port.
Another very similar spec is USB type C. Similarly, it defines the physical connector, but *many* different signals and/or power can transit on it, depending on the specifications supported. Signals for USB 3.1, Thunderbolt, DisplayPort can be carried and the port can use USB power specs, allowing up to 100W (20V x 5A).
Similarly, M.2 can carry PCIe signals (2x lanes or 4x lanes), SATA and more. For the desktop, the latest spec is U.2, which is a superset of SAS and can carry SATA, SAS and PCIe gen3 x4 lanes.
Basically, we are moving more and more towards universal plugs and cables, with the controller defining what transits on top.
Hmmmm... I see this, with 512GB and a SSD controller in a single 28mm x 28mm chip, and I'm wondering to myself... where are the AFFORDABLE 4TB SSDs? It would take 8 of these chips minus the SSD controllers of course plus one other controller. Or an 8TB which would be 8 chips on each side of the PCB.
Competition will help pricing once the other fabs get their VNand equivalents out...
Gosh I hope so. I could take three or four of those 4TB SSDs and toss them into a new 1U NAS build and get rid of the 4U one I have right now (4u, 5x2TB in IcyDock, G3220, 8GB DDR3) and it would use wayyyyyyy less power.
A smaller package deos not mean it's cheaper to produce. At some points things become more expensive the smaller you try to make them. There's no reason to believe this thing costs less than a M.2 SSD of the same capacity.
So when can I put more than 512G in my single-m.2 X99 mobo? No way I'm doing SATA/SATAe. I'm resigned to use one of the PCIe card slots if I ever want to grow storage.
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.
38 Comments
Back to Article
hojnikb - Tuesday, March 22, 2016 - link
Those bad boys need to find its way into the next galaxy s phone >:)lilmoe - Tuesday, March 22, 2016 - link
Amen. Too bad V-Nand UFS didn't make it to the GS7... Oh well, till my next upgrade. They'll have something more interesting by then, hopefully running Windows 10 Mobile.Shadow7037932 - Tuesday, March 22, 2016 - link
>hopefully running Windows 10 Mobile.I wouldn't hold my breath for that.
lilmoe - Tuesday, March 22, 2016 - link
I thought exactly the same, until they announced the TabPro S... Hopefully the "Pro" moniker carries on to their flagship smartphones where "pro" simply means Windows 10. Galaxy S8 Pro sounds good, and not too far fetched. Oh, and hopefully running on Exynos.Dansolo - Wednesday, March 23, 2016 - link
Microsoft employee spotted!checksinthemail - Monday, March 28, 2016 - link
Calm down - people who don't work for Microsoft can like their products too.obligatory BGA-SSD comment: I'm looking forward to these because they'll weigh significantly less.
jjj - Tuesday, March 22, 2016 - link
The article ignores the phone market but that doesn't mean that phones won't go there.Arnulf - Tuesday, March 22, 2016 - link
This is useless for phone SoCs which lack PCIe interface.lilmoe - Tuesday, March 22, 2016 - link
Nothing is stopping SoC manufacturers from slapping in support for future products. But if there *is* an issue, it would imagine it being power envelope. The 950 Pro lags behind AHCI in idle. It might be more efficient for tablets, but possibly not yet so for smartphones.UFS (current and next gen) is a more viable solution for the next couple of years until progress hits the sweetspot power envelope for BGA SSDs on flagship smartphones (I'd imagine these would be solely reserved for flagships). That's when UFS will be passed down to midrange and low end while EMMC is gradually phased out.
lilmoe - Tuesday, March 22, 2016 - link
*** I would imagine it being the power envelope...Solandri - Tuesday, March 22, 2016 - link
Power and temperature.https://www.youtube.com/watch?v=3hhdWwvh5kI
DanNeely - Tuesday, March 22, 2016 - link
Apple went with a PCIe SSD for their last generation of iOS flagships, so it is something that can be fit into a mobile SoC's power envelope.lilmoe - Tuesday, March 22, 2016 - link
Apple's solution isn't the "real thing" by any means, eMMC gets the same random read/write speeds, and UFS 2.0 (1 lane) far exceeds it.We're talking fully fledged Samsung V-NAND SSDs here. You know, like the best desktop/laptop SSDs on the market.
nandex - Tuesday, March 22, 2016 - link
Mobile SoC already has PCIe ports. this is interface for WIFI chips. There is no technology barrier to enable it for storage.Stan11003 - Tuesday, March 22, 2016 - link
The iPhone 6s has a pcie interface with NVME for it's storage. They no longer use SDIO/emmc and can reach 74K IOPS and 1,840 Read and 240 Write MB/s.tuxRoller - Tuesday, March 22, 2016 - link
The 6s and Galaxy 6+ use the same physical layer called m-phy (a mipi standard).The difference is in the higher layers.
Apple use the m-pcie standard, which only keeps the pcie upper layers (dll and TL). Over that they use the nvme protocol for their controller.
Samsung uses the same m-phy layer but, I believe, uses unipro signaling in lieu of pcie, and ufs at the application layer.
Imho, Apple made the better call as to the application layers.
barn25 - Tuesday, March 22, 2016 - link
Incorrect all modern chipsets from the snapdragon 805 and the exynos 5430 as well as intel z2500 series have pci-e suuport. common width is X2.mars2k - Tuesday, March 22, 2016 - link
My first thought tooBurns101 - Tuesday, March 22, 2016 - link
Yes, that would be beautiful my friend! The iphone 6S already have something similiar a believe...Spartus - Tuesday, March 22, 2016 - link
The PM971 SSD can perform 190K random write IOPS as well as up to 150K random write IOPSNeeds fixing
Ryan Smith - Tuesday, March 22, 2016 - link
Thanks!Vorl - Tuesday, March 22, 2016 - link
just a heads up, you have a typo. 4th paragraph: can perform 190K random write IOPS as well as up to 150K random write IOPSThe first one should be read.
plopke - Tuesday, March 22, 2016 - link
I can barely understand anything of all this M.2 stuff . Trying , but I need to look at AHCI/NVMe , PCIe2/4/sata and all that is just branded M.2. So how would BGA effect me as a regular consumer?Urgh I wish they would take more of a USB/sata approach , which isn't perfect but something like M.2(AHCI/sata) M.3(NVMe/PCIe2), M.4(NVMe,PCIe4). Just something so the devices says compatible with M.2,M.3 and M.4 , because so many times if I wanne buy something on a website the specification is wrong/missing or it contains a Asteriks on their site stating depends on your region.
bcronce - Tuesday, March 22, 2016 - link
AHCI supports one queue with a depth of 32 requests. This is hardly enough to keep a modern SSD busy. NVMe, which M.2 implements, supports 65,535 queues each with a depth of 65,535 requests.There is no point in making these devices backwards compatible with AHCI/SATA. The whole issues is AHCI/SATA was too slow. It's like saying we need to make jet engines compatible with prop airplane design. In order to go faster, we need to break the mold.
DanNeely - Tuesday, March 22, 2016 - link
M.2 has a dozen possible keyings (A-M) and 11 possible slotted form factors (see diagram in article), and 8 possible thicknesses (urgh); it's intended primarily as an OEM interface not a consumer one and has way too many moving parts to get everything marked out like that even if was just an SSD spec. As a simplifying factor a lot of that seems to be spelling out every possible option now instead of trying to squeeze them in after the fact.Only 5 of the keyings are defined and 3 actually used for current mainstreamish products (A - wifi, B - Sata/2x PCIe/sata SSDs, and M - 4x PCIe/sata SSDs). The SSD situation is settled a bit more by most systems using B/M slots. Long term x4 PCIe means that M's going to replace B for SSDs. PCIe is forward/backward compatible; you potentially leave performance on the table but if they fit they should work.
Sizes are defined by dimensions in mm and again only use a fraction of the available options. The 30mm wide ones are no-shows (I assume this was a sop to previous larger mobile card spec vendors so they could pretend not to need to change up as much all at once), and I've not seen anything using the ultra narrow 16mm form factor yet. Everything uses the 22mm wide versions: Wifi the smallest 22x30 size, SSDs can use the 22x42/60/80/110 sizes but AFAIK most are either 2242 or 2280 with 2/4 flash chips on them. 22110 would allow for 8 flash chip modules like in 2.5" form factor drives; but even desktop mobos tend not to free up enough space to fit one. At this point I suspect it's dead in the water; and hope U.2 (a 4xPCIe cable) will actually ship allowing for high performance 2.5" SSDs. PCIe m.2 ones currently have most of the speed titles due to larger bandwidth; but the smaller number of flash chips limits their capacity and in some cases bottlenecks them.
stephenbrooks - Tuesday, March 22, 2016 - link
Great info! I'd happily read an article about this sort of thing, since these sort-of-internal interfaces in smaller devices aren't always obvious (nor is what is compatible with what).I also wish they'd find a ZIF type socket that gives the smallness of BGA without being actually soldered down but I guess that's impossible or someone would have figured it out by now...
DanNeely - Tuesday, March 22, 2016 - link
Probably not going to happen. You can space balls more tightly than holes in a socket. Intel's LGA socket probably comes closest in connection density; but the locking lever needed to hold the chip down tightly enough for good contact is hardly zero force.plopke - Tuesday, March 22, 2016 - link
@DanNeely , thx for the short summary ! Hope i can remmember it all if i ever upgrade a laptop or something and hope U.2 gets standard on my desktops :)frenchy_2001 - Tuesday, March 22, 2016 - link
To be precise, M.2 and U.2 are physical connector and board space specifications.M.2 is the specification for mostly mobile, U.2 is a superset of a SAS port.
Another very similar spec is USB type C. Similarly, it defines the physical connector, but *many* different signals and/or power can transit on it, depending on the specifications supported. Signals for USB 3.1, Thunderbolt, DisplayPort can be carried and the port can use USB power specs, allowing up to 100W (20V x 5A).
Similarly, M.2 can carry PCIe signals (2x lanes or 4x lanes), SATA and more.
For the desktop, the latest spec is U.2, which is a superset of SAS and can carry SATA, SAS and PCIe gen3 x4 lanes.
Basically, we are moving more and more towards universal plugs and cables, with the controller defining what transits on top.
bill.rookard - Tuesday, March 22, 2016 - link
Hmmmm... I see this, with 512GB and a SSD controller in a single 28mm x 28mm chip, and I'm wondering to myself... where are the AFFORDABLE 4TB SSDs? It would take 8 of these chips minus the SSD controllers of course plus one other controller. Or an 8TB which would be 8 chips on each side of the PCB.Competition will help pricing once the other fabs get their VNand equivalents out...
extide - Tuesday, March 22, 2016 - link
Samsung's 48 layer V-NAND that just came out should soon allow a reasonably priced 4TB SSD, so .. soon!bill.rookard - Tuesday, March 22, 2016 - link
Gosh I hope so. I could take three or four of those 4TB SSDs and toss them into a new 1U NAS build and get rid of the 4U one I have right now (4u, 5x2TB in IcyDock, G3220, 8GB DDR3) and it would use wayyyyyyy less power.MrSpadge - Tuesday, March 22, 2016 - link
A smaller package deos not mean it's cheaper to produce. At some points things become more expensive the smaller you try to make them. There's no reason to believe this thing costs less than a M.2 SSD of the same capacity.Pinn - Tuesday, March 22, 2016 - link
So when can I put more than 512G in my single-m.2 X99 mobo? No way I'm doing SATA/SATAe. I'm resigned to use one of the PCIe card slots if I ever want to grow storage.asmian - Tuesday, March 22, 2016 - link
The grammar howlers in this piece just make you look amateur. You guys badly need to get a professional editor."Despite of tiny form-factor" -> despite having a...?
"whom in turn needed" -> who, as it's the subject not object.
http://www.dummies.com/how-to/content/choosing-to-...
Zertzable - Wednesday, March 23, 2016 - link
This. The content is great, but in many articles the grammar is extremely distracting.SaolDan - Friday, March 25, 2016 - link
I come to this site for the content. I dont care about the grammar.Michael Bay - Wednesday, March 23, 2016 - link
When do I get a 2Tb SSD for a reasonable price, damn it.