When do you think we will finally see U.2 for SSD become popular for client?

[cbn]

I've been thinking this would most likely happen when NAND gets relatively cheap for at least two reasons:

1. Current third party SATA 6 Gbps SSD controllers AFAIK can only handle up to 2TB (Samsung 850 EVO 2.5" is available in 4TB capacity though)

2. M.2 form factor can't hold as many packages as the 2.5" form factor (though I wonder how much true impact this has when each package can hold up to eight chips or in some cases sixteen chips)

 

[Elixer]

Yeah, once NAND is down in price...

 

[cbn]

I got to thinking after writing this post perhaps having some kind of 3.5" form factor enclosure with U.2 connector and internal RAID would also help? (In a small way)

This way a person could take four or more small capacity SATA SSDs (edit: or HDDs) and make one useful large volume SSD (edit: or SSHD) out of it? (particularly useful for PCs with chipsets not supporting RAID).

 

[cbn]

Or maybe a 3.5" form factor enclosure with U.2 connector and internal RAID forM.2 NVMe drives?

EDIT: This enclosure (connected to a PCIe 4.0 x 4 or PCIe 3.0 x 4 U.2 port) filled with either PCIe 4.0 x 4, PCIe 3.0 x 4 or PCIe 3.0 x 2 M.2 NVMe SSDs and compared to large capacity PCIe 4.0 x 4 or PCIe 3.0 x 4 M.2 or 2.5" form factor NVMe SSDs.

EDIT2: Here is ASUS Hyper Express 3.5" enclosure (with picture below) which can fit two M.2 and two mSATA SSDs. (So four or more M.2 2280 NVMe in a similar 3.5" enclosure should be do-able)

 

[cbn]

According to the following article U.2 and M.2 would both stand to gain a large increase in bandwidth sooner than expected:

http://www.networkworld.com/article...ress-4-0-is-done-5-0-spec-nears-approval.html

PCI Express 3.0, or PCIe, was finished in 2010, and motherboards began to appear in 2011. The 4.0 spec should have been done within three years but only now is being finished because if there’s one way to screw up development, it’s to have it done by committee, and 700 cooks can really spoil the broth.

Now the PCI-SIG has announced that PCIe 4.0 specification is finished, and it plans to have PCIe 5.0 ready to go by 2019, which is liable to cripple the 4.0 spec before it even gets out of the gate.

Two things I wonder about:

1. How soon PCIe 5.0 will show up on Intel and AMD motherboards?

2. How much NAND will it take to saturate Sequential Read for PCIe 5.0 x 4? (For good/fast NAND it shouldn't take that much, but perhaps with the large die NAND (which has less parallelism per GB) would the SSD need to be 2.5" form factor in some cases?)

EDIT: Looking at the published specs of the Intel 600p NVMe SSD in the Anandtech review here I extrapolated out that it would take 2TB of the Intel-Micron Gen1 3D TLC to saturate PCIe 5.0 x 4 for Sequential Read. Of course, by the time PCIe 5.0 is out Intel-Micron will be using different NAND (at the very minimum Gen 2, but much more likely Gen 3....but how much parallelism per GB will this have compared to Gen1? Same or less?

 

[cbn]

Maybe another factor helping U.2 would be if the heat from a PCIe 4.0 or PCIe 5.0 controller needs the 2.5" form factor (acting as a heatsink) to reach its full potential?

 

[Billy Tallis]

but how much parallelism per GB will this have compared to Gen1? Same or less?

More, actually. The 600p uses 384Gb 32L TLC parts while the 545s uses 256Gb 64L parts. There will also be a large-die 512Gb 64L part from Intel/Micron, but that won't get used in the 545s series and will hopefully only be used on drives with a high enough total capacity to fully populate every channel on the controller.

 

[cbn]

More, actually. The 600p uses 384Gb 32L TLC parts while the 545s uses 256Gb 64L parts. There will also be a large-die 512Gb 64L part from Intel/Micron, but that won't get used in the 545s series and will hopefully only be used on drives with a high enough total capacity to fully populate every channel on the controller.

I know Intel-Micron Gen1 (32 layer) 256Gb MLC/384 Gb TLC has four planes, Gen 2 (64 layer) has eight planes?

 

[cbn]

I've been thinking this would most likely happen when NAND gets relatively cheap for at least two reasons:

1. Current third party SATA 6 Gbps SSD controllers AFAIK can only handle up to 2TB (Samsung 850 EVO 2.5" is available in 4TB capacity though)

2. M.2 form factor can't hold as many packages as the 2.5" form factor (though I wonder how much true impact this has when each package can hold up to eight chips or in some cases sixteen chips)

Regarding #1, recently Phison announced their S12 SATA 6 Gbps controller. With the S8 and S10 able to handle 1TB and 2TB respectively I'm thinking the S12 will probably be able to handle 4TB (like the Samsung 850 EVO).

 

[cbn]

Regarding #1, recently Phison announced their S12 SATA 6 Gbps controller. With the S8 and S10 able to handle 1TB and 2TB respectively I'm thinking the S12 will probably be able to handle 4TB (like the Samsung 850 EVO).

The following article claims S12 will be able to handle 8TB:

https://www.tweaktown.com/news/59075/phisons-new-ssd-line-up-featured-fms-2017/index.html

Phison isn't just doing PCIe storage; they are also updating their SATA line-up with their new S12 controller. Like the E12, the 8-channel S12 is designed to support 3D flash. The S12 will support 3D MLC/TLC/QLC NAND flash and LDPC (Low-Density Parity Check) error correction technology. The S12 will support capacities of up to 8TB.