NVMe and an interesting technology change

April 3, 2019

Back in the middle of 2015, I wrote an entry on sorting out NVMe as the next way to connect SSDs to your system. Someone I know online was recently reading it, and he mentioned that he'd never heard of the 'U.2' connector that I talked about in that entry and that in fact I described as the way that future NVMe SSDs would be connected to your machine. In an aside in that entry, I wrote:

(PCIe and thus NVMe can also be connected up with a less popular connector standard called M.2. [...]

In 2015 and even 2016, U.2 was a reasonably big thing; you can read one example of it (and see some pictures of a U.2 connector and U.2 SSD) in places like this April 2016 article. Since then it has quietly vanished away, swept away by that 'less popular' M.2 standard that I mentioned in my aside. In fact the Wikipedia page on U.2 is pretty amusing (at least right now) due to its 'U.2 compared to M.2' section, which comes across pretty much as the last cry of people who cannot stand to see their beautiful thing crushed by the marketplace. U.2's fade out evidently didn't take all that long either, since when I did a late 2017 entry on sorting out M.2 and NVMe, it wasn't even on my radar and certainly didn't show up on any of the motherboards I looked at at the time.

(U.2 is apparently still sort of out there, especially in datacenter applications, and it seems to show up every so often on semi-consumer motherboards. See eg here.)

Technology change and failed standards are not exactly new to the PC world, but for me this is still an interesting and impressive example of it in action. U.2 was the obvious thing in the middle of 2015, and then two years later it had just disappeared completely.

(Possibly even in 2015 U.2 was more hype than anything else and I was taken in by it when I wrote my entry.)

Sidebar: Some speculations on why M.2 won out over U.2

I can think of a number of plausible contributing factors:

  • M.2 was easier to put into laptops, because it required fewer parts (eg no cables). That gave it volume and we all know volume drives down price.

  • M.2 was more flexible, since an M.2 connector can be used for either SATA or NVMe. Both manufacturers and consumers could trade off cost versus performance without having to change anything other than the M.2 card in use.

  • Most people don't use more than one or two drives in their computers.
  • Most people prefer the simplicity of plugging a M.2 card into a motherboard connector rather than mounting a separate drive and running cables to it.

  • Intel didn't make enough PCIe lanes available in consumer chipsets to run enough U.2 drives to be attractive. Whether U.2 or M.2, Intel was always only going to give you one or two.

  • M.2 SSDs (whether NVMe or SATA) are cheaper to make than U.2 SSDs.

While U.2 theoretically makes it easier to have larger NVMe SSDs, my impression is that in the consumer market the largest limiting factor on SSD sizes is how much people have been willing to pay for them. This certainly is the case for me.

(In the enterprise market I've read stories saying that the limit is how much data loss people want to be exposed to from one device failing.)


Comments on this page:

By AdamK at 2019-04-03 02:22:49:

U.2 supports dual port drives, I'm not sure if M.2 does that.

By rchard2scout at 2019-04-03 02:44:52:

For future reference: I've started editing the Wikipedia section you mentioned, and it'll probably be edited further soon. The original version is available here

So basically U.2 is more like a disk drive connector and M.2 is more like RAM sticks? It seems to me that all of the listed drawbacks of M.2 (including the now-deleted ones) apply exactly the same way to RAM sticks… but nobody would ask for RAM on a cable and/or in an enclosure, and it’s never been attached that way since at least the home computer days. SSDs have been becoming progressively less like disk drives and more like RAM from the very moment they caught on… so in retrospect, while it may be too bad for U.2 fans, it seems like the writing was on the wall?

(So extrapolating forward, my guess would be that the M.2 drawbacks will be addressed to at least the degree the same issues are addressed in RAM sticks, whereafter the subject will cease to be of any more interest than it is today when pertaining to RAM. We do have RAM sticks with heat sinks on them today and gamers buy them, for example.)

By cks at 2019-04-12 14:56:38:

Belatedly: I believe that M.2 drives do sometimes come with heat sinks and motherboards sometimes have heatsinks for them as well; apparently there are also aftermarket heat sinks as well, just to round out the options. However, there seems to be a bunch of people saying that heatsinks aren't worth it on M.2 drives because it doesn't seem to make a difference in drive performance under load.

(Heatsinks do lower the drive temperatures, but bare M.2 drives apparently don't throttle very much under high temperatures from high load.)

By Miksa at 2019-04-16 09:32:06:

The heatsink isn't practically necessary in normal use outside of benchmarking. But I have also read that the controller wants to run cool to avoid throttling, but the flash chips on the other hand work best when warm.

I think some M.2 drives actually have a thin metallic heat spreader to transfer heat from the controller to the flash chips.

Written on 03 April 2019.
« Our plan for handling TRIM'ing our ZFS fileserver SSDs
A sign of people's fading belief in RSS syndication »

Page tools: View Source, View Normal.
Search:
Login: Password:

Last modified: Wed Apr 3 01:12:39 2019
This dinky wiki is brought to you by the Insane Hackers Guild, Python sub-branch.