MCIO 8i SFF-TA-1016 vs SFF-8654 8i - are they mechanically and electrically interchangeable function-wise? (with loss of bandwidth)

[sam55todd]

Hello,

I'm waiting for delivery of SFF-8654 8i adapter for dual Gen4 M.2 NVMe drives like this one: Cablecy Raid Card VROC Slimline SFF-8654 8X Raid0 Hyper Adapter Dual 2X NVME | eBay

And two MCIO cables (my motherboard has three MCIO 8i connectors for 6 drives total via those connectors, without touching other options {onboard M.2, SAS/SATA, etc.})
e.g. dual-cable assembly from dell labeled by seller as "07TPV2 FOR DELL Slimline SAS MCIO SFF-8654 8I 7TPV2 Data Cable" listed here 07TPV2 FOR DELL Slimline SAS MCIO SFF-8654 8I 7TPV2 Data Cable | eBay (got a good discount on it)

Motherboard: SuperMicro X13DEI X13DEI | Motherboards | Products | Supermicro (MCIO jacks wired directly into CPU PCIe Gen5 lines).

I practically have no doubts cables (despite being a bit angled) will fit on MB side but have concerns on cable matching socket on NVMe adapter side.

Does anyone has experience with this, to share insights and results?

Will cutting plastic on cable jack be sufficient to make signal chain work properly?
or there are more caveats to expect?

Target state to make this work at least on PCIe Gen3 level, or preferably Gen4.

Thank you.

 

[sam55todd]

Sadly after extra research into specifications and despite same width and number of contacts - it looks like this won't work since female end of SFF-8654 can receive 1.14mm thick plug (which are actually 1mm as for PCIe-mini PCBs) but MCIO plug has thickness of 1.57mm (exact match to PCIe PCB thickness).



In conclusion: MCIO is too thick (by 38%) for plugging into SFF-8654 receptacle.
(but plastic looks relatively flexible inside SFF-8654 receptable, I'm just worried it can push and fold metallic contacts inside due to excess pressure and too broad entry point)

not like this has much of relevance but NVMe PCB thickness is 0.8mm (width of all 3 is roughly 22mm)

 

[sam55todd]

All above is overridden by getting couple of right cables 85 Ohm Mini Cool Edge IO / MCIO / SFF-TA-1016 to SFF-8654 by 10Gtec on Amazon (74 pins / 8i on both ends), a bit pricey ($30 each) and too long (75cm) with 1+ month delivery time (theoretically China to US Amazon to UK Amazon but in reality route might be shorter, although Amazon mostly goes strictly to their distribution centre policies, therefore it might unnecessarily bounce a little around a world) but at least workable solution.
Extra length (at 16GHz{G4}/32GHz{G5} frequencies, therefore increased loss and very high sensitivity to connector and conductor quality) puts it below PCIe Gen5 standard (plus it most likely pairs are unshielded anyway) but my NVMe drives are Gen4 so I hope this shouldn't be a problem (for now, before next upgrade).
p.s. on pics above forgot to measure insertion depth, but it's irrelevant anyway.

p.p.s. electrically they are interchangeable apart from slight difference in sideband channels but those should detect automatically anyway.

 

[DaveLTX]

All above is overridden by getting couple of right cables 85 Ohm Mini Cool Edge IO / MCIO / SFF-TA-1016 to SFF-8654 by 10Gtec on Amazon (74 pins / 8i on both ends), a bit pricey ($30 each) and too long (75cm) with 1+ month delivery time (theoretically China to US Amazon to UK Amazon but in reality route might be shorter, although Amazon mostly goes strictly to their distribution centre policies, therefore it might unnecessarily bounce a little around a world) but at least workable solution.
Extra length (at 16GHz{G4}/32GHz{G5} frequencies, therefore increased loss and very high sensitivity to connector and conductor quality) puts it below PCIe Gen5 standard (plus it most likely pairs are unshielded anyway) but my NVMe drives are Gen4 so I hope this shouldn't be a problem (for now, before next upgrade).
p.s. on pics above forgot to measure insertion depth, but it's irrelevant anyway.

p.p.s. electrically they are interchangeable apart from slight difference in sideband channels but those should detect automatically anyway.

32GT/s not 32GHz. PCIe 5.0 runs at 4GHz. (Don't confuse clock speed with data rates requiring "16GHz nyquist"!) but yes it could be a little marginal

However, PCIe G5 retimers are pretty much required so 75cm is actually passable

 

[sam55todd]

32GT/s not 32GHz. PCIe 5.0 runs at 4GHz. (Don't confuse clock speed with data rates requiring "16GHz nyquist"!) but yes it could be a little marginal

However, PCIe G5 retimers are pretty much required so 75cm is actually passable

It's not like I've tested myself but rather was a result of quick-google for "PCIe gen frequency" landing me on a following page:

What Is PCIe? A Basic Definition

What is the meaning of PCIe? PCIe 4.0, PCIe 5.0, PCIe 6.0 and older generations explained.

www.tomshardware.com

another reason to take everything published on tomshardware with a pinch of salt.

 

[KaneTW]

32GT/s not 32GHz. PCIe 5.0 runs at 4GHz. (Don't confuse clock speed with data rates requiring "16GHz nyquist"!) but yes it could be a little marginal

However, PCIe G5 retimers are pretty much required so 75cm is actually passable

PCIe gen 5 is a 32GT/s NRZ protocol. The fundamental frequency is 16GHz.
(PCIe gen 6 is a 64GT/s PAM4 protocol, which also has a fundamental frequency of 16GHz.)

75cm is doable with good PCB material and good cables without retimer.

See https://blog.samtec.com/wp-content/...1_successful_PCIe_interconnect_guidelines.pdf for details.