No mfgs have even announced any shipping products for it.
Actually, there is one:
Asrock's USB 3.1 5.25" bay panel. Seems like
Asus made one too. Don't know if it shipped, though.
That seriously sounds like a complete guess. How many motherboards had SATA Express at launch? How many AM4 motherboards are out there now for a lunch platform? Do you believe every motherboard following this AM4 launch will be feature parity with existing boards throughout the service life? You're taking a bunch of observations (new platform, new CPUs, new chipset, new designs) and saying that it's diminishing, which it is, but you provide complete conjecture as to *why* its diminishing. That also includes the fact that you seem to think it has some sort of "permanence" about it. Samsung SSD shipments were down Q1 2016, but no one considered it permanent. And it wasn't.
Okay, first of all, you can't possibly be drawing parallels between sales trends of an entire product segment (SSDs) and adoption of a "new" interface. Where on earth is the commonality there? History has shown us time and time again that slow or non-existent adoption of new connector standards is a clear indication that the standard will go away quickly. See Firewire or eSATA for reference.
Second, SATAe has been plentifully available on motherboards since Haswell. Yet devices just don't show up. How do you explain that?
And sure, we haven't seen half of the AM4 motherboards supposed to be available (30-something of the 82+ AMD mentioned). Yet very few of them - including high-end models - have SATAe.
My numbers aren't wrong, but I could see the mis-interpretations. In my original post, I was discussing AM4, because AMD is implementing SATA Express on Chipset. In that area, reviewers have noted that the 2 SATA Express Ports can be divided into SATA, used as 2 x2 PCIe NVMe ports, or coalesced into an x4 PCIe port, hence why it can also be another M.2 slot.
Sorry, but you said this:
If they want, the end user can deploy 2 SATA ports. If they want, the end user can deploy 2 PCI-e NVM drives. If they want, the OEM can take the 4 PCI-e lanes and redirect that for any general purpose they want, or just make a slot.
I.e. you're explaining it as if 2 sata ports (which with an additional connector make up a single SATAe connector) could somehow transmit 4 lanes of PCIe, or connect 2 NVMe drives. Which is simply false. I'll cut you some slack, though, and say your wording was simply not clear. Still, the X370 chipset allows for either 6 SATA ports + 2 lanes of PCIe 2.0, or 4 SATA ports + 4 lanes of PCIe 2.0. For now, the main implementation of this seems to be to use the dedicated x4 PCIe 3.0 from the Ryzen CPU to run an m.2 slot, and use the remaining lanes for SATA ports and the remaining PCIe for onboard devices or PCIe slots (and in some cases a second m.2 slot with either SATA or SATA/PCIe 2.0 x2 speeds).
Again, as you noted, it's conjecture. You have provided no evidence, while I have provided evidence indicating the opposite:
http://www.anandtech.com/show/9369/...-sata-6gbps-jmf815-pcie-controllers-next-year
Ahem. You're quoting an article from 2015 stating that controllers will be coming "next year". If so, where are they? Did you miss 2016? Or might they also be one of the victims of manufacturers cutting their losses relative to a stillborn standard? Who here is guilty of conjecture now? "Oh, they said it was in development, so it'll for sure arrive at some point!"
Controllers stay on process nodes a very long time due to cost reasons. Even after the jump to 28nm, it still doesn't necessarily mean there aren't room for x2 controllers. You're welcome to bring conjecture if you want, but stating it as some sort of fact is silly.
EDIT: Noting that @cbn posted this information ahead of me. I made this post in my spare time over 3 hours, so I missed it.
🙂
Again, the post is from 2015. I can't find a single mention of the controller since. Either it's cancelled or it has just never been used in a single consumer-facing product.
To the first point, that's fine. Again, SATA Express does not require devices be made for the interface to live.
To the second, a single brand new platform launch does not a trend make. Since you said "facts", can you bring any evidence forward with any numbers that shows SATA Express is releasing on fewer motherboards within stable platforms? Or is this more of an observation?
Some examples:
Now, again, those are a few examples. Do I think they're indicative of a trend? Yes, considering that for the last two generations, pretty much every high-end board has had SATAe connectors.
To your first point: actually, I'd argue that it does. Why? Because storage devices represent the
vast majority of connected devices in PCs today. Bay devices, AICs outside of GPUs, and other expansions have largely gone extinct thanks to more and more features being integrated into motherboards. If you don't have SATAe storage devices, you're already limiting the use of that connector to a tiny handful of people.
To your second: It doesn't make a trend, but see above. SATAe is slowly but surely disappearing from motherboards. Some manufacturers (Asus, Asrock) are doing away with it faster than others (Gigabyte seems to love SATAe), but it's happening.
It adds board complexity? How?
By having to route PCIe lanes out to the far edges of the board, and requiring these to be right next to SATA ports.
AM4 leaves it entirely up to the OEM. They could not implement it at all, or they could make additional SATA ports. Or they could even make the block a x4 PCIe slot, or hell, make it an M.2 slot. The idea behind AM4's implementation is that it gives the OEM total freedom on the final implementation without the complexity of having to add an additional controller to the motherboard. How can M.2 help with that? How can U.2 help with that?
Have I argued for adding controllers to motherboards? Do either m.2 or u.2 require those? No. I'm arguing for u.2 (or something similar) with lane splitting support - a
standard PCIe feature that's sadly mostly disabled in consumer chipsets. Every single server chipset in the world supports it. With lane splitting, the chipset handles everything you need.
Why do you think that space constricted M.2 slots do not add to board complexity when you have to route a slot, and make sure no components interfere with the M.2 card that ends up mounted on he motherboard. Seriously, that doesn't make any sense. As for the connector, compared to what? M.2 only has a kludge of a connector. U.2 carries 2 channels as well but is a much more robust cable, SFF-8639. The cables are individually shielded! Have you ever bought Mini HD SAS cables? Are you aware of how much those things cost right now?
Please, pay attention: that I'm arguing that SATAe adds board complexity doesn't
in any way indicate that I'm arguint
against m.2 also doing the same. Of course it does! Saying anything else would be completely bonkers. The difference is that m.2
is a standard that actually has a use. As such, it's a requirement on a modern motherboard. SATAe really isn't. Which, again leads to the board complexity argument: when you
have to have m.2, adding SATAe
adds more complexity which increases costs without any tangible gain. It's that simple.
And yeah, u.2 cables are crazy expensive, as are most hihgly shielded enterprise cables. This would of course have to change for consumer adoption. The thing is, SATAe cables wouldn't really have been any cheaper. At all. After all, they're still sending rather sensitive PCIe signals over a wire, and are thus equally susceptible to interference and the like.
I would love that too. Lots of uses. But we don't have it. The rest is simply an opinion which I can't agree with. U.2 takes less board space as a connector, but gives up flexibility. It's entirely personal opinion on which you would rather have. I agree the SATA Express cable is therefore much larger, but that's a trade-off. Do you think people will want to spend $40 on what's essentially a SAS3 cable? We can't even get people to spend more than $70 on their power supplies a great deal of the time.[/quote]
No. Having separate connectors
does not give up flexibility. Say you have two identical boards, one with 2xSATAe, and one with 4 SATA ports and an x2 u.2 port. Which of those allow you to connect the most devices at the same time? The second, given enough lanes to allow use of all ports at once. And if you don't, then your point is moot. In addition, u.2 allows manufacturers to route SATA and PCIe lanes separately, requiring less protection against interference and thus less complex boards.
U.2 lane splitting already exists, because it's primarily a commercial product designed for Enterpise use. Dual Port U.2 drives like the Intel DC3600 have been on the market for a year now. But they go many to one, rather than the many to many you're wanting. That's because U.2 either sends 4x PCIe for 1 controller, or 2, 2x PCIe for dual controller mode. Unlike SATA Express, all of the lanes are sent down each U.2 connector to each device. How would you do what you wanted? Would you go back to the days of master / slave drives where you manually set each drive, or use a separate cable that only had half the lanes in each U.2 connector? That standard doesn't exist by the way. You'd have to create a whole new SFF cable standard. And even when you did achieve that, you'd end up with 2, 2x PCI-e connectors. That seems like a lot of work for something we already have though. It's called SATA Express 😉
So ... you're arguing that because enterprise devices use a feature (lane splitting) in one way, it could never be used in another. Sure, that makes sense. If the chipset is capable of communicating with separate controllers separately, it doesn't matter if these are on a single PCB or in different buildings (as long as the signal is intact and latency is acceptable). As long as the chipset supports lane splitting on the 4 PCIe lanes routed to the u.2 port, it shouldn't care whether you connect a single x4 device, four x1 devices, 2 x2 devices, or 2 x1 + 1 x2 on the other end.
Could dual-device cables be a solution to this? Sure, but that would be clunky. I'd argue for a connector standard that splits along the middle (like EPS connectors and the like, just better suited for its task, obviously. As such, something like 2 SATA cables
alone would work. Just without the extra doohickey, and placed side-by-side vertically to each other, not the ridiculous "How wide can we make this connector?" approach of SATAe. With either one capable of transmitting PCIe x2. And not requiring those horrible breakout power cables. Okay? U.2 is
far from perfect. But it's far better suited for modern computing than SATAe. That cable is straight out of the "practicality is a word I've never heard of" rulebook of '90s computing.
There isn't a doubt in my mind that PCIe over cables is a growing necessity in modern PCs, and it would greatly help a whole host of add-ons. The PCIe card standard is for a large part outdated in design, and we need more flexible, smaller form factor solutions for added features. SATAe, though, is
not the solution to this. It's incompatible with small devices due to its huge size, and it's wildly impractical - which has seriously limited adoption of standards previously.
