News NVMe 2.0 Brings Support for PCIe-Connected Hard Disk Drives

[Admin]

Version 2.0 of the NVM Express protocol has just been announced, revealing that it will also add support for hard disk drives.

NVMe 2.0 Brings Support for PCIe-Connected Hard Disk Drives : Read more

 

[hotaru.hino]

Having everything on one protocol sounds nice, but then I ask how will this all be connected?

 

[JarredWaltonGPU]

Having everything on one protocol sounds nice, but then I ask how will this all be connected?

Yeah, no doubt. Will we use U.2 cables for all our devices? And I guess we'll have to have shared PCIe lanes if we're going to have more than a few U.2 / M.2 / whatever ports.

 

[kal326]

WD mentioned this last year about SMR hard drives support in ZNS. Makes sense given the sequential write performance in large block SMR drives versus random write operations.

 

[dehjomz]

A few questions:

• What connector will be used to attach hard drives to pcie via the nvme protocol?
• Does this mean we may finally see a samsung evo 870 successor that eschews sata and uses pcie instead, and runs much faster than 550MB/sec?
• What's the theoretical maximum speed of a pcie hard drive?
• Are we potentially talking pcie5.0 bandwidth in a 2.5 inch hard drive form factor?

 

[hotaru.hino]

• What connector will be used to attach hard drives to pcie via the nvme protocol?
• Does this mean we may finally see a samsung evo 870 successor that eschews sata and uses pcie instead, and runs much faster than 550MB/sec?
• What's the theoretical maximum speed of a pcie hard drive?
• Are we potentially talking pcie5.0 bandwidth in a 2.5 inch hard drive form factor?

• The only NVMe based connector that I'm aware of for use with traditional storage drives is U.2
• That would be the 970.
• No faster than what the fastest hard drive on the market is. Hard drives haven't even saturated SATA 3 yet.
• Maybe available to it, but no hard drive is going to reach those speeds.

 

[mikewinddale]

And now SATA Express feels even more rejected. (For those who don't remember, the SATA Express connector was backwards compatible with SATA and the AHCI protocol but it also included two lanes of PCIe and the NVMe protocol.)

 

[USAFRet]

Are we potentially talking pcie5.0 bandwidth in a 2.5 inch hard drive form factor?

Not a chance.
Mechanical drives are still mechanical.

They can't even saturate a SATA II connection.

 

[InvalidError]

Not a chance.
Mechanical drives are still mechanical.

They can't even saturate a SATA II connection.

As pointed out in the article, Seagate's fastest SATA HDDs can max SATA III out using dual actuators. Bandwidth could hypothetically increase further with triple and quad actuators in the future. My 7 years old WD Black peaks at 240MB/s sequential, so SATA II was already quite close to being a bottleneck for fast HDDs back then.

Even SATA3 is becoming inadequate for today's fastest HDDs.

 

[TommyTwoTone66]

Having everything on one protocol sounds nice, but then I ask how will this all be connected?

These will be for servers. There will be PCIe connectors inside NAS enclosures, large-scale SAN devices or storage arrays inside giant server racks.

Consumer grade stuff will continue to be solid state drives in NVMe slots.

 

[InvalidError]

Consumer grade stuff will continue to be solid state drives in NVMe slots.

I'd imagine that dual-actuator HDDs able to push SATA3 beyond its limits will get into the upper-end of mainstream for people who want 8+TB of storage or have a NAS. If you can get multi-actuator HDDs with 2-4X the throughput, you can rebuild a compromised RAID array 2-4X as fast.

 

[derekullo]

If dual actuators is 524 MB/s then 8 actuators is 2096 megabytes a second.

Add enough actuators and a hard drive becomes pseudo-random access.

 

[ptmmac]

I am pretty sure we could be seeing filesystems like ZFS that have all the newest goodies so that we will no longer have to worry about where we are saving data to to the right disk insure a fast connection. There is no doubt making storage simpler, and more resilient would be a boon to everyone. This is one step on the way to something better. Hopefully we will be seeing the first fiber optic interconnects between silicon in the next 3-5 years.

 

[littlechipsbigchips]

Great ! This would Allow Mechanical HDD to include NVME SSD caching into the drives connected directly to the PCIe lanes . Bravo ! it will be coming soon.
Expect HDD/NVSSD drives to pop up soon !

we will see HDD cached speeds that matches modern NVME SSD speeds . forget about 512MB/s , talk in thousands !

 

[dehjomz]

Not a chance.
Mechanical drives are still mechanical.

They can't even saturate a SATA II connection.

There are SSDs that use the 2.5 inch form factor like the Samsung evo 860 and 870. But because they use sata, they’re capped at 550 MB/sec.

Moving to pcie could spur innovation in this space and allow Samsung and others to release SSDs that use the 2.5 inch form factor, rather than the m.2 form factor. I would love to see that happen.

 

[InvalidError]

Moving to pcie could spur innovation in this space and allow Samsung and others to release SSDs that use the 2.5 inch form factor, rather than the m.2 form factor. I would love to see that happen.

NVMe in 2.5"/3.5" form factor wouldn't be much of an innovation, just a different form factor for the same stuff. As I have written in the 12VO stories, we're long overdue for new IO standards based on 12VO and PCIe. NVMe 2.0 would need to introduce some form of new cables and connectors to bring PCIe off-board, it would be nice if that included 12V power so NVMe 2.0 drives don't need separate legacy power connectors and cables or intermediate DC-DC power conversions on the motherboard.

 

[dehjomz]

NVMe in 2.5"/3.5" form factor wouldn't be much of an innovation, just a different form factor for the same stuff. As I have written in the 12VO stories, we're long overdue for new IO standards based on 12VO and PCIe. NVMe 2.0 would need to introduce some form of new cables and connectors to bring PCIe off-board, it would be nice if that included 12V power so NVMe 2.0 drives don't need separate legacy power connectors and cables or intermediate DC-DC power conversions on the motherboard.

The bandwidth limit of sata in my opinion is the biggest roadblock to innovation in this space. I never understood why a sata 6gbps successor wasn’t ever implemented. So I’m excited that pcie will now be used. Yes it’s a different form factor but you also have to consider the surface area. With m.2 the memory is compressed into a small area, which leads to high heat density especially on pcie4.0 m.2 drives. If they could spread the memory out over a larger surface area like in 2.5 inch or 3.5 inch drive, that should be much easier to cool, especially as pcie speeds increase to 5.0 and 6.0. Case in point: the pcie4.0 Samsung 980 pro and western digital sn850 run hot! I wonder how much cooler those drives would run if they were instead implemented in a 2.5 or 3.5 inch enclosure? I’d argue that a cool-running pcie4.0 nvme drive with speeds of 6000MB/sec or more in a 2.5 inch enclosure with gobs of terabytes is innovation (compared to the Samsung 870 evo with its 550MB/sec).

I wholeheartedly agree with you that I’m interested in seeing what cable they use to power these devices as well as transfer data to/from the pcie bus. They could usb-c similar to thunderbolt, since we know such cables can carry power delivery, pcie, and data. But of course thunderbolt is limited to only 4 lanes of pcie3.0. So I’m interested to see what they develop to take advantage of pcie4,5, and 6 when it arrives.

 

[InvalidError]

With m.2 the memory is compressed into a small area, which leads to high heat density especially on pcie4.0 m.2 drives. If they could spread the memory out over a larger surface area like in 2.5 inch or 3.5 inch drive, that should be much easier to cool

I wholeheartedly agree with you that I’m interested in seeing what cable they use to power these devices as well as transfer data to/from the pcie bus. They could usb-c similar to thunderbolt, since we know such cables can carry power delivery, pcie, and data. But of course thunderbolt is limited to only 4 lanes of pcie3.0.

Making NAND easier to cool isn't particularly important since NAND actually prefers running hot as it lowers the amount of energy needed to move charge in/out of cells which in turn reduces wear, albeit at the expense of increased leakage which reduces data retention time from 10+ years at 25C to one year at 85C. At one point, there were even papers about putting heaters in NAND chips to bring worn blocks to 500+C in order to restore their gate's insulation most of the way back to new condition. It is mostly the controller that needs cooling, NAND only requires enough cooling to keep it from erasing itself by overheating and get it back to ambient after writes for data retention.

As for what connector to use, I would also go along the line of something similar to the internal type-C header, stripped of all unnecessary extras like alt-modes and USB-PD. Have straight PCIe + 12V5A, anything else can be handled by adapter boards and backplanes. Imagine having only one single cable about the same size as SATA carrying power + 4.0x4 data for a 2-12 bays storage backplane. You can have cheap backplanes that do hard-wired bifurcation for up to four drives and fancier ones with PCIe switches to extend the maximum number of drives beyond four and share bandwidth across the whole lot. 12V5A would also be enough for four typical 7200RPM 3.5" HDDs, so at least one of those NVMe 2.0 motherboard ports could have HSIO/FlexIO lanes with SATA3 support for a SATA backplane, maybe add an optional 12V AUX connector in case someone wants to use some ultra-fast SATA HDDs.

 

[littlechipsbigchips]

I wonder when HDD will Evolve from spinning plates into fixed non spinning plates and one to one head for read writes !!!