U.2 / U.3 to USB 3.2 for Ext. U.2 access (not cheap, but may be useful)

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

TrumanHW

Active Member
Sep 16, 2018
253
34
28
USB 3.x - to - U.2 interface (for SSD | NVMe)
  • MFR: ICY Dock (always expensive)
  • Model: MB931U-1VB

  • Interface: USB-C 3.2
  • USB 3.0 Gen. 2 (USB-C interface by default)

  • www.ebay.com/itm/224622912015
    (just the cheapest eBay listing: I am NOT recommending the seller)



My Cheap Skate Method:
(which I've used since before I knew of any products)
  • SFF-8643 - to - U.2 SFF-8639 cable
  • SFF-8643 to M.2 (to the USB interface)
  • SFF-8639 to U.2 SSD

Providing Power:
  • USB3 - to - SATA's 15p SATA Power (failed)
  • AC Power - to - 4p Molex, then, 4p Molex to U.2's 15p SATA (worked)

Data Connection-Tests for NVMe:
(initial attempts were for TB3 to get max performance):

Failed* Attempts:
  • Thunderbolt 3 to M.2 (NVMe) via Thunderbolt 3 to M.2 (NVMe) "Fusion PCIe"
  • Thunderbolt 3 to M.2 (NVMe) via Thunderbolt 3 to M.2 (NVMe) "OWC Envoy Express"
  • Thunderbolt 3 to M.2 (NVMe) via Thunderbolt 3 to M.2 (NVMe) "Samsung X5"
Note: Many of the TB3 interfaces worked but only briefly; often just long enough to test it before unmounting, hence the designation "failed."

Successful Attempt(s):

  • USB 3.1 Gen. 2 - to - M.2 (NVMe) -- "QNINE" M.2 (NVME) via USB-C
  • Drive Tested with: Samsung PM983 3.72TB SSD (not the DCT version).
  • Drive remained mounted consistently unless any of the points of failure came loose (pretty easy to do).

Please feel free to let me know if there's anything you'd like me to test / try.
I have other interfaces & am happy to try better ideas with products I have.
Did I screw up? See a better way to keep the TB3 interface connected ..? Pls LMK.
 

RageBone

Active Member
Jul 11, 2017
617
159
43
i fear not, based on the assumption that the Controllers in such Adapters talk USB to NVME specifically.
The PCIe side is usually not visible / transparent to the host.

It will likely be different with Thunderbolt since that is PCIe through and through.
Hey, there were those thunderbolt GPU enclosures.
Those might work with a HBA
 
  • Like
Reactions: TrumanHW and Samir

sparx

Active Member
Jul 16, 2015
320
118
43
Sweden
Yeah that makes sense. I guess as thunderbolt keeps evolving the older GPU enclosures drops in price. Its not like I need one. But it would be pretty cool to buy a Supermicro 4U JBOD and put an Intel Nuc with a thunderbolt HBA inside of it with truenas :)
 

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
Yeah that makes sense. I guess as thunderbolt keeps evolving the older GPU enclosures drops in price. Its not like I need one. But it would be pretty cool to buy a Supermicro 4U JBOD and put an Intel Nuc with a thunderbolt HBA inside of it with truenas :)
Or on a smaller scale it really is viable I think to use and external thunderbolt array with a NUC or similar, Mac mini and so on.
Just difficult to find the right kind of unit, the OWC ones could be ok…. They also sell some that can support SAS cards for SAS disks. Hmmm
 
  • Like
Reactions: Samir

Cruzader

Well-Known Member
Jan 1, 2021
540
544
93
Yeah that makes sense. I guess as thunderbolt keeps evolving the older GPU enclosures drops in price. Its not like I need one. But it would be pretty cool to buy a Supermicro 4U JBOD and put an Intel Nuc with a thunderbolt HBA inside of it with truenas :)
m.2 slots on gen5 and up are usualy dual mode so you can use them for pcie.
mini pcie egpu pcbs are like 30-50$ (example aliexpress one), popular way to get 10gig sfp+ cards on nucs for vmware labs.

So its very doable fairly cheap.
 
  • Like
Reactions: TrumanHW and Samir

TrumanHW

Active Member
Sep 16, 2018
253
34
28
There are those thunderbolt GPU enclosures that might work with an HBA
Exactly -- and of COURSE it will (obviously) ...
The right way to look at Thunderbolt ..? It's NOTHING special, it's just a small, convenient External PCIe cable & interface.

There's nothing special about a "Thunderbolt 2 or TB3 eGPU"
It's just a "PCIe to Thunderbolt 3 enclosure" ... with adequate wattage.
You can use an "eGPU enclosure" for NICs or HBAs ... or, use a "non eGPU enclosure" for a GPU if you have adequate external power to the GPU.

I've thought about marketing a TB3 enclosure with a 400w AC adapter that has the 6-pin and 8-pin GPU power-connectors.


Now, just to ensure I'm not a total turd who gave out misinformation yesterday when saying that TB3 to the U.2 NVMe drives didn't work ...


THO I WAS ABLE TO TEST LONG ENOUGH TO GET PERFORMANCE NUMBERS

The connectivity was absolutely NOT reliable. And cannot (for some reason) be counted on to transfer data for more than a few minutes.
And I tested both with the PM983 ... and ... with an Optane P5800x
(which of course wasn't particularly impressive as it's also limited to PCIe 3.0 protocol on my MacBook Pro)

For whatever reason however (as TB3 still provides more power to my knowledge and I'm still using ext. power), USB 3.1 Gen 2 was MORE reliable.

That said, yes, TB3 produces faster (tho unreliable connectivity-duration) MB/sec:


USB 3.1 vs TB3 - PM983.png



TB3 (X5) + 980 Pro 16MBP - MBs.png
Above: Samsung 980 Pro (2TB) NVMe
Note: The 980 Pro is a PCIe 4.0 x4 device.
16in i9 MBP is a PCIe 3.0 x4 device.







Below, Testing the OPTANE p5800x via:

• Thunderbolt 3 the p5800x, only the price is remarkable.
• USB3.1 Gen 2 the p5800x [looked] less impressive than the 980 Pro.

(obviously I'm not sugg. the 980 Pro is superior but an anomaly of the testing method)

UPDATED - as I posted the wrong image as astutely noted by a member.

TB3 to Optane p5800x - Benchmarks.pngTB3 to Optane p5800x - Recognized.png


Although the Optane p5800x is also a PCIe 4.0 device, it too was limited by the PCIe gen 3.0 interface as well as being on a gerry rigged TB3.
As poor as the results for the Optane are, they're still just about the best results I've seen by any PCIe 3.0 dev - Thunderbolt 3 which wasn't also:

- in RAID-0 of a pair of NVMe devices (as tested using the NetStor dual NVMe device (but only be a modicum and not in benchmarking).
- is running (at least with the M.2 devices) off the power supplied using the Thunderbolt 3 port.

I've done TONS of tests using Thunderbolt caddies (Seagate FreeAgent -- Powered (3.5") and unpowered (2.5")
A powered Thunderbolt SATA SSD will usually be a little slower (370MB/s even if the drive can hit 500MB/s using internal SATA) ..
Than using a USB 3.0 to SATA cable (without external power).

Sometimes, a powered TB2 to SATA dock (HighPoint -- usually based on the Model SSD, like an evo 860) can hit 520 MB/s (synthetic).

But as a rule ... powered devices will generally provide better results.
Even if it's a USB 3.0 or USB 3.2 (3.x Gen 2) in a PCIe slot ... as in, without supplying the 15p SATA power to the card it will be slower.

No additional power cable (you'd think the 75w would be enough) generally yields ~75% the 'powered' performance ... (irrespective protocol).


If there's anything in there you'd like me to provide proof (tests) for ... provided I have the components I'll be happy to test.

I'm tempted to load MacOS on to the Optane P5800x to see how High Sierra or Mojave runs on a 12c MacPro 5,1 while I have them.
 

Attachments

Last edited:

Brian Puccio

Member
Jul 26, 2014
70
33
18
41
Thanks for all this info!

One minor nitpick: at the end you have the Samsung 983 screenshots and then you say P5800x below, but your screenshot is a repeat of a Samsung one above.
 
  • Like
Reactions: TrumanHW and Samir

RageBone

Active Member
Jul 11, 2017
617
159
43
@TrumanHW
Have i understood you right?
You were adapting to the U.2 drives with a u.2 cable to a pcie slot / m.2 on a Adapter with a Controllter that can do USB and TB?
And then over a Type C Cable to the machines TB / USB port?

To add a maybe explanation, i have heard that Thunderbolt is very picky about cables and that basically no cables above 30cm in length can do TB3 at 40Gbits with connections issues to be expected.
I would expect that the Type C cable + the u.2 cable together also exceed PCIe capabilities and therefore cause the issues.
Redrivers might fix the issues?

I mean, USB at 10 or 20 Gbits is still way more lenient in such matters and i'd expect less issues with USB compared to TB, like you described i think.
 
  • Like
Reactions: Samir

TrumanHW

Active Member
Sep 16, 2018
253
34
28
M.2 slots on gen5 and up are usually dual mode (can be used for pcie).
(Popular way to get 10GbE SFP+ cards on nucs for vmware labs)
Holy CRAP. It's a POPULAR WAY !?
I swear. I thought I was a genius for thinking this up having not seen anyone else do it. !!! LOL!!!

BUT, I'm actually going to be using an M.2 to TB3 to add an SFP28 to a MacBook Pro ...
I already have the TB3 to M.2 (NVMe) adapter (many) ...
Just gotta figure out which SFP28 dev is gonna be easiest to make "plug-n-play' with a wide range of ver. MacOS

If I'm not mistaken -- seems like the ATTO cards use a custom driver with Mellanox cards.

Do you have any insight to help ? I swear! I thought my idea was novell, genius and original (dammit!!) lol.
(I shoulda known everyone else saw this before me - just wish I didn't have to reinvent a wheel, lol).
 
  • Like
Reactions: Samir

TrumanHW

Active Member
Sep 16, 2018
253
34
28
Link for those units that have sas support?

I think SAS (unless you mean adding a controller? Which'd also be pretty easy) would always be easier than adding a U.2 NVMe device ... but, I've not tested that as those drives just weren't ever on my RADAR. Sorry.
 
  • Like
Reactions: Samir

TrumanHW

Active Member
Sep 16, 2018
253
34
28
Can't be Novell--they are just a faint memory now after being one of the biggest NOSes in history:
No no, I don't mean the proprietary Novell (OS) ... I meant, novel as in an original idea. :)

Going from a PCIe female x4 slot to an SFP28 card ...
Which'd be connected via an M.2 to to Thunderbolt 3 adapter to connect to that PCIe x4 slot ...

And by doing so -- provide SFP28 hardware to a MacBook Pro ... as, the ONLY device that exists..?

Are either SFP+ at 10Gb -- (Sonnet and many other mfrs (even QNAP) make such devices)
But, for SFP28 or QSFP+ to TB3 adapter ..? ONLY ATTO creates this and they are ALL well over $1,000.

SFP28 to TB3 is about $1,500.

My expected cost to use one of my existing TB3 to SFP28 (by using an M.2 (NVMe) to TB3) costs:
- Price of any M.2 NVMe device (the OWC EnvoyExpress is probably the least expensive at $90)
- Otherwise you can buy a dedicated TB3 to PCIe x4 for $200 ... as the above may be x2
- But I have many, so that's no cost ...
- The M.2 to PCIe x4 is under $30 (the one I've found that looks nice)
The least expensive unit I've found


But, obviously, I want to make sure I get one that MacOS either natively supports or for which I can add 'drivers' for.

I think Chelsio and Miricom often have MacOS drivers ...
ATTO is apparently a rebranded Melanox but uses a custom driver which I'm planning to test out also

Thanks for the interest and observations.
 
  • Like
Reactions: Samir

Cruzader

Well-Known Member
Jan 1, 2021
540
544
93
Holy CRAP. It's a POPULAR WAY !?
I swear. I thought I was a genius for thinking this up having not seen anyone else do it. !!! LOL!!!

BUT, I'm actually going to be using an M.2 to TB3 to add an SFP28 to a MacBook Pro ...
I already have the TB3 to M.2 (NVMe) adapter (many) ...
Just gotta figure out which SFP28 dev is gonna be easiest to make "plug-n-play' with a wide range of ver. MacOS

If I'm not mistaken -- seems like the ATTO cards use a custom driver with Mellanox cards.

Do you have any insight to help ? I swear! I thought my idea was novell, genius and original (dammit!!) lol.
(I shoulda known everyone else saw this before me - just wish I didn't have to reinvent a wheel, lol).
for NUCs a m.2-> x4 adapter and ribbon riser has been the road to 10g nics.
Ive done same to get 10g nic + sas hba on some itx boards with a single regular size slot also.

As for mac drivers im not a mac user so dunno, but same drivers as one normally would use for cards internaly.
So if the ATTO cards have that available there is not much of a issue beyond how to neatly solve the adapter spagetti.

The nucs with new printed top/bottom to put cards in almost look worthy of being official solutions :D
1643357353798.png
1643358837701.png
 

TrumanHW

Active Member
Sep 16, 2018
253
34
28
for NUCs a m.2-> x4 adapter and ribbon riser has been the road to 10g nics.
Ive done same to get 10g nic + sas hba on some itx boards with a single regular size slot also.

As for mac drivers im not a mac user so dunno, but same drivers as one normally would use for cards internaly.
So if the ATTO cards have that available there is not much of a issue beyond how to neatly solve the adapter spagetti.

The nucs with new printed top/bottom to put cards in almost look worthy of being official solutions :D
View attachment 21371
View attachment 21372
That's great confirmation that my planned method of adding a 28GbE NIC to my Mac will work. :)
Thanks buddy.