Does this U.2 cable exist?

[Ericloewe]

I've looked around a bunch with very little success, so I'm going to crowdsource this question:

Does a cable exist that can provide SATA (and/or SAS) plus PCIe to a single SFF-8639 (U.2 backplane side connector), so that a compact backplane could provide full SATA (with optional SAS) and PCIe connectivity?

This is a weird one at first glance, so here are the details:

Background
U.2 (SFF-8639) extends the legacy SAS connector (SFF-8482) by adding all the extra pins necessary for PCIe x4 signalling, independently of the existing SAS/SATA pins. (As a note, SFF-8482 was itself an extension of the SATA connector that added a second channel, to allow for multipath connections to disks.) This effectively means that U.2 backplanes can be analyzed as a legacy SATA/SAS backplane plus a PCIe backplane squished together into one PCB and connector, with independent uplinks for PCIe and for SATA/SAS. This allows for a backplane that is wired for both SATA/SAS and PCIe to take SATA/SAS drives or U.2 (and U.3) drives.

U.3 is a scam heavily pushed by Microchip and Broadcom. It allows for simpler backplanes by reusing some of the connections used for SAS as PCIe. This reduces the number of differential pairs that need to be routed across the backplane PCB and across cabling back to the respective hosts. However, it requires a tri-mode controller and (for most practical scenarios) a tri-mode expander. These are absurdly expensive, even by the standards of legacy SAS devices. To make things worse, implementations are often less than optimal - Broadcom, for instance, presents NVMe drives as SCSI devices, negating much of the advantage enjoyed by NVMe drives through lower protocol overhead. This is in addition to the - typically - limited bandwidth available to the tri-mode SAS controller (Both the LSI/Broadcom SAS 9500 and SAS 9600 have a PCIe 4.0 x8 connection to the host, meaning that any setup over two PCIe 4.0 SSDs or four PCIe 3.0 SSDs is immediately oversubscribed. This is a worse oversubscription scenario than was typically seen in plain SAS setups, despite the fact that many NVMe devices do make use of most of their available bandwidth.) In fairness, HPE is trying something interesting, with U.3 backplanes wired only for PCIe x1. This is interesting more than useful, but it allows PCIe connectivity without needing additional differential pairs when compared to SAS/SATA, at the cost of much reduced bandwidth (The promise of higher-speed, lower lane count SSDs has not really turned into a popular option for anybody. Maybe PCIe 6.0 will bring single-lane SSDs with performance equivalent to x4 PCIe 4.0 drives.).

Although most U.2 server chassis support SATA/SAS through dedicated uplink connectors, separate from the PCIe connector, this is not a popular configuration outside rackmount servers, as far as I can tell. However, there are some retrofit hot-swap bays for ATX-style chassis that present a U.2 drive-side connector (example unit from Icy Dock) and can be used with either SATA/SAS or PCIe. Thing is, they could actually operate be simultaneously wired up for both with an appropriate cable.

What am I looking for:
My goal is to have one or more hot swap bays in a standard ATX-style chassis (in 5.25" or 3.5" external bays), which will support SATA, SAS (single-lane) and PCIe without needing to be rewired, as is typical in a server chassis (see for instance the Dell R6515 10-bay backplane or the Supermicro BPN-SAS3-826EL1-N4 backplane). To make this work in a backplane that presents a disk-side SFF-8639, I would need a cable like the following crude diagram:

Code:

SFF-8643 (PCIe 3.0/4.0 x4)------\
SATA data connector (SATA/SAS)-------U.2/SFF-8639
SATA power connector (power)----/

Similar cables that omit the SATA/SAS part are plentiful, if absurdly expensive, but I have not been able to find a "fully-featured" cable.

Are you completely insane? Do you not have anything better to do?
I mean, yes I am, and yes I do. I'd like to satisfy my OCD on this one without sticking a Dell R630 under my desk.

 

[cobrakiller58]

I was unaware of such a cable existing, I thought because the signaling is different for NVME and SAS they MUST be routed to different controllers unless you have tri-mode controller.

 

[Ericloewe]

Well, yes, that's exactly what I need/want. It's a weird cable, for sure, if anyone makes one, but it's no uglier than a QSFP breakout, and those are mainstream. Sure, it confuses clueless users, but that ship sailed when they decided to re-use SFF-8643 for PCIe.

 

[cobrakiller58]

It certainly looks like something that could be made, if the dock does not have a switch(supermicro using jumpers) to dictate the drive type in it then the cable itself would need the switch.

 

[Ericloewe]

It certainly looks like something that could be made, if the dock does not have a switch(supermicro using jumpers) to dictate the drive type in it then the cable itself would need the switch.

What makes you say that? SAS/SATA and NVMe are, in essence, completely independent in U.2. The only reason you can't typically use both at once is because disks are one or the other. Of course, I also want to get a dual-M.2 to U.2 adapter that provides one M.2 SATA and one M.2 PCIe slot. Not because I really need to double-stack drive bays, but because I want to see IPMI/iDRAC freak out at having two disks in one bay.

See also the El-Cheapo PCIe slot to U.2 adapter I bought the other day, which has a SATA port on the side so you can use the card for SAS/SATA drives too, because reasons. No switches there, just a bunch of traces to the U.2 connector and power regulation to get the 5 V line to the SSD.

 

[cobrakiller58]

My limited understanding is U.2 is just an interface which is typically implemented in the consumer market for NVMe protocol. The linked media you shared implies all three require switching of some sort based on the type of drive that is connected either SATA/SAS or NVMe, if a tri-mode controller is not involved the backplane or the cable will have to handle the routing to either a PCIe controller, SAS controller or SATA controller.
I might be totally wrong, but that is my current understanding of how this works
Edit:
Is there a technical reason I'm not seeing for the cable you want being available? SFF-8087 for sas/sata and SFF-8643 for PCIe on mainboard side and SFF-8639 on drive side
or is it just about having redundant components?
Edit2:actually SFF-8087 would be wasteful, what would do the single sata/sas connection?

 

[Ericloewe]

The linked media you shared implies all three require switching of some sort based on the type of drive that is connected either SATA/SAS or NVMe

No, on the contrary, U.2 is implemented as a new set of pins crammed onto the SAS connector, for the PCIe lanes, PCIe clock, and SMBus. So both parts are independent.

Is there a technical reason I'm not seeing for the cable you want being available? SFF-8087 for sas/sata and SFF-8643 for PCIe on mainboard side and SFF-8639 on drive side
or is it just about having redundant components?

Cables for SATA/SAS or for NVMe are plentiful, if not necessarily cheap. I'm looking for a cable that does both at once.

Edit2:actually SFF-8087 would be wasteful, what would do the single sata/sas connection?

A traditional SATA-style data connector, for which I cannot find an SFF number. I'm not aware of any other formal designation, so I've long taken to calling them SATA-style connectors.