Supermicro AOC-SLG3-8E2P Octaport NVMe HBA

[Ouroboros]

I recently discovered that Supermicro has a PCIe PLX switch HBA card that supports 8 downstream devices via OCuLink, and that it's been out for maybe a year now.

AOC-SLG3-8E2P | Add-on Cards | Accessories | Products - Super Micro Computer, Inc.

Does anybody have any experience with this beast? Apparently the only thing google pulls up is a user on a Mac Pro forum claiming their tower got fried when trying to use it.

MP All Models - Blade SSDs - NVMe & AHCI

With 8 ports, it beats out Amfeltec's newer hexa Squid, but you can't buy it normally (Supermicro seems to be not selling this openly).

 

[Ouroboros]

More googling pulls up comparable intel product that is obtainable...

AXXP3SWX08080
8-Port PCIe Gen3 x8 Switch AIC AXXP3SWX08080 Product Specifications

Which is an PCIe x8 to 8 OCulink board, in a FHHL form factor. Though the sticking point is 4 of those OCulink ports are at 90 degrees, and I can't seem to find any right angle OCulink to U.2 cables, and nobody seems to have a female-female OCulink gender changer. My google-fu is weak apparently.

 

[csp-guy]

More googling pulls up comparable intel product that is obtainable...

AXXP3SWX08080
8-Port PCIe Gen3 x8 Switch AIC AXXP3SWX08080 Product Specifications

Which is an PCIe x8 to 8 OCulink board, in a FHHL form factor. Though the sticking point is 4 of those OCulink ports are at 90 degrees, and I can't seem to find any right angle OCulink to U.2 cables, and nobody seems to have a female-female OCulink gender changer. My google-fu is weak apparently.

Supermicro has oculink to U.2 cable for 15 bucks, using molex power connection.

https://store.supermicro.com/supermicro-55cm-oculink-to-u-2-pcie-with-power-cable-cbl-sast-0956.html

 

[zack$]

See here for the Intel part:

https://forums.servethehome.com/ind...ink-interface-supermicro-aoc-slg3-4e2p.21081/

 

[Rand__]

Wonder whether SM will do a PCIe4 part for that now that those are (soon to be) available with H12 boards...

 

[ullbeking]

I'm going to install the quad-port version (AOC-SLG3-4E2P) into an SC747 chassis ( SC747TG-R1400B-SQ | 4U | Chassis | Products | Super Micro Computer, Inc. ). My research leads me to the understanding that this is going to require swapping my current backplane, which supplies the 8x hot-swap front-loading SATA/SAS drive bays (BPN-SAS-747TQ), with the BPN-SAS-747TQ-N4 backplane, which has 4x NVMe/SATA/SAS drive bays plus 4x SATA/SAS drive bays ( https://www.supermicro.com/manuals/other/BPN-SAS3-747TQ-N4.pdf ).

Has anybody tried a solution like this or used this backplane or a similar one?

I'm also looking at solutions involving mobile racks, which, I understand, already come with backplanes that allow 4x NVMe connections per 1x 5.25" slot.. I've already got 1x DVD+-RW ODD installed and I have 2x 5.25" slots spare.

 

[Ouroboros]

There has been some discussion in another thread regarding using IcyDock 5.25 bay quad NVMe docks, as they use a "legacy" SFF-8643 pinout that may not play well with some cards. There was an intel AOC octoport card that apparently was not happy, though that used a PMC Sierra PCIe switch, compared to the Supermicro AOC which might be using a PLX switch. Though the IcyDock reports have not tested one of the SuperMicro OCulink to SFF-8643 cables with the OCulink end marked with a green band (which Supermicro takes pains to point out for use with their backplanes...)

 

[csp-guy]

I would like to now the max. Performance of these cards. For example with 8x optane 900p drive 2500:2000 read:write / drive performance. I think PCI-e 3.0 bus bandwidth too thin to get all available bandwidth of drives.

 

[Ouroboros]

I got one of these via ebay (they've started showing up there recently), but I'm unable to boot at all with it plugged into the x16 slot in a test system (Fujitsu Primergy TX100 S3), which only has PCIe 2.0 slots. Just get a single beep every 2 seconds with no post codes. Tried using the cut open x4 slot, as well as a riser adapter in the remaining x8 slot, but no joy. Some sort of compatibility issue (PCIe 3.0 should run in a 2.0 slot, but maybe SMBus conflict?) or DOA (looks like it came in the original packaging so I don't think so...)? Can't tell really. Need to scrounge another machine to see if it's just that system, but a full on blocking boot fail by adding a PCIe card is kinda harsh in this day and age.

Update:
Was able to stuff it into a TX1310 in its PCIe 3.0 x16 slot and it booted alright, but it appears this does NOT use a PLX chip, but a chip by PMC-Sierra. Don't have a U.2 drive and cable yet to test data functionality, but I am detecting 9 PMC-Sierra management endpoints (but that's also coming up as PCI bridges). This may have similar consequences for the Intel AXP3SWX08080 since that also seems to use a PMC-Sierra PCIe switch chip

 

[JoshDi]

Does anyone use this board (AOC-SLG3-8E2P) with Oculink to U.2 cables (no backplane)?

I am looking to use this with my X11-SPM-TF Supermicro board, Intel VROC, and Intel DC P3520 drives.

 

[JoshDi]

Does anyone use this board (AOC-SLG3-8E2P) with Oculink to U.2 cables (no backplane)?

I am looking to use this with my X11-SPM-TF Supermicro board, Intel VROC, and Intel DC P3520 drives.

I received the board (AOC-SLG3-8E2P) and it works fine with the supermicro Oculink to U.2 cables without a backplane. I am also using it with an Intel VROC Key. Supermicro informed me that it will only work in slot 4 on my board - not slot 6.

 

[Ouroboros]

A sorta update

MicroSATAcables.com is now making an OcuLink to SFF-8643 "legacy" type cable for use between that Intel card and the IcyDock cage.

OCulink SFF-8611 Straight Male to SFF-8643 Cable

When I get a chance, I will attempt to use that with the Supermicro card after testing with the Intel card.

 

[cyrus104]

@Ouroboros , any word on if the legacy type cables work with the SM card?

 

[bekax]

A sorta update

MicroSATAcables.com is now making an OcuLink to SFF-8643 "legacy" type cable for use between that Intel card and the IcyDock cage.

OCulink SFF-8611 Straight Male to SFF-8643 Cable

When I get a chance, I will attempt to use that with the Supermicro card after testing with the Intel card.

@Ouroboros

I'm also curious as to whether it works with those "legacy" cables ?

I'm planning to use the M.2 IcyDock MB720M2K-B with either an AOC-SLG3-4E4T or AOC-SLG3-8E2P, depending on whether HPE fixes the ML30 Gen10 server lack of quadfurcation in the BIOS (only bifurcation appears, not dual bifurcation).

 

[ullbeking]

I've still got my Supermicro AOC-SLG3-4E2P and a couple days ago an IcyDock MB699VP-B was delivered.

However it slipped my mind that I have no cables ready, and not only, but I haven't decided on which one of two options to jump for!! Not only this, but the store I bought the IcyDock MB699VP-B from has a 14-day refund policy, and I don't think I'll have decided on cables and have them arrive well in time!!

I think I'd better send the IcyDock cage back for a refund now, take care of the cables, and then re-order the cage just in case the nothing works.

Here are two options for "legacy" wired OCuLink cables:

• Change the Supermicro AOC-SLG3-4E2P HBA controller to a Highpoint SSD7120 HBA and use Highpoint 8643-8643-060 cables. It has been confirmed by Highpoint that their SSD7210 HBA is compatible with the IcyDock MB699VP-B via their 8643-8643-060 cables.
• Experiment with specially/legacy wired cables from another company (they are sending me a sample) and check whether what I have works when using those.

I'll let you know how things work out.

I'm even considering changing the whole lot to an Intel kit, but I don't have confirmation that the Intel HBA controller works with the mainboard.

 

[Rand__]

Just in case you can wait, there is an OCulink based version of that IcyDock enclosure in the works. ETA is still months out though.

 

[ullbeking]

Just in case you can wait, there is an OCulink based version of that IcyDock enclosure in the works. ETA is still months out though.

This is great news. How do you know?! They told me a year ago they would have a new version out in several months, but I think Covid probably messed up their plans.

I can actually wait. I'm going to configure my system in the most flexible way for future repurposing. Hopefully it is only a few months though.

 

[Bjorn Smith]

I have the Icy dock MB699VP-B, Supermicro AOC-SLG3-2E4 and they work like a charm together.

The cables I use are as far as I know it simple SFF-8643-SFF-8643 cables - nothing fancy. (CBL-SAST-0590) is what it says, so a supermicro cable

But I might be wrong - I got them together with two cards I bought from a guy on this site - and they might be "special"

 

[ullbeking]

I have the Icy dock MB699VP-B, Supermicro AOC-SLG3-2E4 and they work like a charm together.

Great to know!!

The cables I use are as far as I know it simple SFF-8643-SFF-8643 cables - nothing fancy. (CBL-SAST-0590) is what it says, so a supermicro cable

But I might be wrong - I got them together with two cards I bought from a guy on this site - and they might be "special"

I have also heard "something-like-rumors" that Supermicro's pin-to-wiring assignment for OCuLink cables is a bit different to other companies', i.e., "legacy" wiring. Aside from that I never heard anybody confirm or deny exactly what "version" of OCuLink wiring SM in an authoritative way.

Moreover, I have just been sent a sample cable that was created especially by the folks at a specialist cabling company. They have confirmed that these cables were created precisely for the problem we are currently discussing. As soon as this workstation is operational again I will report on whether it works with these cables too. This company apparently went to a lot of trouble to try to solve the very problem in this thread.

 

[Bert]

Reviving the old thread. I noticed the cable from SM to convert occulink to U2 but IIUC, each occulink port is 2 U2 channels but the cable only maps to single U2. Is a break out cable required here?