Qotom Denverton fanless system with 4 SFP+

[blunden]

My VyOS running the linked image is hooked up to a Juniper EX3300, running JunOS 15.1R7-S4 and works fine. Milage may definitely vary at least with the problematic commit.

Interesting. Well, the offending commit was added specifically to work around a bug seen with some Juniper MX5 "link partners", whatever that last part means.

When you say "linked image", do you mean the ones linked by @kryptonian above? Those use the out-of-tree Intel drivers that are known to work just fine.

 

[kryptonian]

Interesting. Well, the offending commit was added specifically to work around a bug seen with some Juniper MX5 "link partners", whatever that last part means.

When you say "linked image", do you mean the ones linked by @kryptonian above? Those use the out-of-tree Intel drivers that are known to work just fine.

Yes, I'm the same person who linked them above.

 

[blunden]

Yes, I'm the same person who linked them above.

Haha, I didn't look at the username.

Yes, I would imagine the Intel driver possibly not having that behavior change or they might have additional changes that make it work in both cases.

It's too bad that you didn't manage to build the in-kernel driver with the offending commit reverted. That would've been a very interesting test.

Unfortunately, unless this gets the attention of the Linux kernel developers (discussion on the mailing list seems to have stopped) I don't think we'll see this fixed in any upcoming LTS kernels, or possibly ever. Clearly, the regression wasn't noticed before it was merged or when the author was testing the change, and Intel seems to consider it solved on their end in their own driver.

 

[kryptonian]

Yes, I would imagine the Intel driver possibly not having that behavior change or they might have additional changes that make it work in both cases.

It's too bad that you didn't manage to build the in-kernel driver with the offending commit reverted. That would've been a very interesting test.

Unfortunately, unless this gets the attention of the Linux kernel developers (discussion on the mailing list seems to have stopped) I don't think we'll see this fixed in any upcoming LTS kernels, or possibly ever. Clearly, the regression wasn't noticed before it was merged or when the author was testing the change, and Intel seems to consider it solved on their end in their own driver.

The issue with the kernel sources is that the source tree is large, and just git revert didn't wanna work in CI, so I gave up on that idea. Other possibility is to use git patch to revert it but haven't wanted to spend the compute to test that out as I'm building the images on Buildjet instead of the free runners (as the free runners are way slower).

I also don't know why I never got really any answer as to what's the actual difference between the out-of-tree driver and built-in or BSD and Linux driver.

 

[TheGeekn°72]

Funny thing, similar thing is also available from TekLager.se. The one with C3558R: TLSense C3558R: 2x 10Gbps SFP+, 7x 2.5Gbps LAN, Atom C3558R CPU, and there used to be one with C3758, and I have the latter version.

I'm guessing this is just a rebrand or something because Qotom seems to be the OEM as they ship from factory and have the BIOS images and Windows drivers downloads available on their website

Something to keep in mind, is that it seems the SFP+ ports might have some uh, driver issues in the Linux kernel, and as a result most of the times it will not link up even if the SFP+ module is regonized. If you use the Intel's out-of-tree driver for it however, it does link-up and work perfectly with 10G but your milage may vary. I'm currently building VyOS with the out-of-tree driver as a result, and releases and source is available: Releases · samipsolutions/vyos-build. (Related VyOS task: T5619 Update the Intel ixgbe driver due to issues with Intel X533)

This has been mentioned a number of times in the last 2~3 pages, apparently it has to do with a specific but recent version of the kernel ? Not well versed enough about it to get into the details but I recommend rewinding the thread over said few pages, maybe you can notify us of something we haven't mentioned/find mention of something you didn't know ? Everybody wins

EDIT : whoops, didn't refresh my page, I didn't see all the replies made after the one I replied to, my bad !

 

[blunden]

I'm guessing this is just a rebrand or something because Qotom seems to be the OEM as they ship from factory and have the BIOS images and Windows drivers downloads available on their website

Yes, Qotom is an OEM/ODM that develops and sells designs for other companies, as well as directly to consumers. TekLager sell devices from a number of different OEM/ODMs after testing them. They then offer fast shipping, warranty, etc. on top, as well as extra services and validation. Before AliExpress and Amazon, I imagine that sourcing these devices was a lot more of a hassle, especially if you didn't plan to buy in bulk.

EDIT: To clarify, I would assume that Qotom gives the customers they design these type pf devices for some period of exclusivity before they start selling them directly on the open market.

 

[blunden]

The negotiation behavior implemented by the potentially problematic commit mentioned in your links and also mentioned in here previously seems to be required on devices on both ends for that change to work reliably.

Since Patrick tested with two devices that both had that driver change, it makes sense that it worked in their tests. The problem is that most other devices and OS:es don't.

@Patrick Any chance you could test again, but with something other than an X553 (using the same driver code) on the other end? See the reasoning above.

 

[blunden]

@Vesalius Not sure why you deleted your post since the last part was interesting.

To answer your first part, yes, I think the tests he performed falls under this scenario that the Amazon employee found:

Specifically, our 10G link works when both sides of the fiber are
running the in-tree 6.1 ixgbe driver with this autonegotiation change.

My understanding is that the same change remains in kernel 6.5 (and later) so that should still apply. It's interesting that you don't see the same behavior (working link) with matching kernel 6.x drivers though.

 

[kingneutron]

Please bear with me as I'm brand-new to 10Gbe networking, but would these inexpensive 10G transceivers work with the Qotom Q20332G9?

https://www.amazon.com/gp/product/B01N1H1Z2F/ref=ewc_pr_img_1?smid=AE2OZG2NN3099&th=1

Would also like to know if I need to buy dual-mode or single-mode fiber. TIA

I also posted a question on Amazon to the seller asking for a user manual as I'm interested in more info on the Console RJ45 port. FYI I bought and just received mine, just waiting on VGA cable to arrive (hopefully tomorrow) before I boot it.

 

[blunden]

Please bear with me as I'm brand-new to 10Gbe networking, but would these inexpensive 10G transceivers work with the Qotom Q20332G9?

https://www.amazon.com/gp/product/B01N1H1Z2F/ref=ewc_pr_img_1?smid=AE2OZG2NN3099&th=1

Would also like to know if I need to buy dual-mode or single-mode fiber. TIA

I also posted a question on Amazon to the seller asking for a user manual as I'm interested in more info on the Console RJ45 port. FYI I bought and just received mine, just waiting on VGA cable to arrive (hopefully tomorrow) before I boot it.

Yes, those transceivers should work. According to others, the Intel SFP+ slots are not vendor locked.

The fiber cables you need to buy depend on the transceivers you choose. Those transceivers you linked are for multi-mode fiber so you'd buy either OM3 or OM4 cable with LC-LC connectors.

How long is the run? If it's just a relatively short run to a switch, you can opt for a passive DAC cable instead. They are cheaper, lower power and ever so slightly lower latency (since it doesn't need to do any conversion).

I was unable to get output from my console port with my Cisco-style console cable (while I waited for my VGA to HDMI adapter to arrive). I haven't checked if I need to enable console redirection in the BIOS though.

Note that the port numbering of the RJ45 ports can be a bit confusing. In VyOS, the ports were numbered like this for me:

SFP+:

eth0 eth2
eth1 eth3

RJ45:

eth7 eth6 eth8
eth5 eth4

 

[kingneutron]

Thanks. FYI I just wasted an evening configuring the preflashed Win10 install.

DO NOT try to update the Intel network drivers for Win10 Pro -- I tried with the latest downloads from Intel website AND Snappy Driver Update, and ALL of the 2.5Gbit Ethernet ports disappeared.

Win10 won't even boot to desktop now after trying a rollback from the last restore point. (There's a reason I'm not a Windoze guy anymore. Fkg stupid OS. Can't rollback the driver from Configure either...) Under Win10 Pro I was only getting ~31MiB/sec on Winscp file transfers talking to my Macpro 2013. Bullshit performance, I have no idea how Patrick got decent 2.5Gbit transfer speeds if he was using the default install and non-tweaked settings.

Tomorrow will try Suse Tumbleweed and/or Proxmox 7.4 based on other posts I've seen. I don't think this qotom unit is really designed for win10.

 

[blunden]

Thanks. FYI I just wasted an evening configuring the preflashed Win10 install.

DO NOT try to update the Intel network drivers for Win10 Pro -- I tried with the latest downloads from Intel website AND Snappy Driver Update, and ALL of the 2.5Gbit Ethernet ports disappeared.

Win10 won't even boot to desktop now after trying a rollback from the last restore point. (There's a reason I'm not a Windoze guy anymore. Fkg stupid OS. Can't rollback the driver from Configure either...) Under Win10 Pro I was only getting ~31MiB/sec on Winscp file transfers talking to my Macpro 2013. Bullshit performance, I have no idea how Patrick got decent 2.5Gbit transfer speeds if he was using the default install and non-tweaked settings.

Tomorrow will try Suse Tumbleweed and/or Proxmox 7.4 based on other posts I've seen. I don't think this qotom unit is really designed for win10.

I wouldn't use the preinstalled Windows install on any machine. Also, I don't think anyone really expects you to run Windows on these. If that's what people want, there are better mini-PCs to buy.

It's clearly meant to be used as a software router or a virtualization host, which usually means either Linux or BSD. I honestly don't know why they bother preinstalling Windows on it, but I suppose they might do so more for hardware testing at the factory.

I hope you have a better experience with the other OS:es you try.

 

[TheGeekn°72]

Note that the port numbering of the RJ45 ports can be a bit confusing. In VyOS, the ports were numbered like this for me:

SFP+:
eth0 eth2
eth1 eth3

RJ45:
eth7 eth6 eth8
eth5 eth4

in OPNsense they also have peculiar designations :

SFP+:
ix0 to ix3, I don't have any SFP equipment to slot in so I couldn't localize for the physical placement just yet but it's not a far fetched guess to think it'd be in the same order, right ?

RJ45:
eth3 eth2 eth4
eth1 eth0

 

[blunden]

in OPNsense they also have peculiar designations :

SFP+:
ix0 to ix3, I don't have any SFP equipment to slot in so I couldn't localize for the physical placement just yet but it's not a far fetched guess to think it'd be in the same order, right ?

RJ45:
eth3 eth2 eth4
eth1 eth0

Looks like it has a similar order in OPNsense too then, which makes sense. Good to know.

Yeah, the SFP+ slots are likely labeled as follows in OPNsense:

ix0 ix2
ix1 ix3

The difference seems to be that VyOS doesn't differentiate between physical connection type, so it names the SFP+ slots as ethX too. If you add an offset of 4 to your RJ45 interface names they line up with mine.

 

[TheGeekn°72]

The difference seems to be that VyOS doesn't differentiate between physical connection type, so it names the SFP+ slots as ethX too. If you add an offset of 4 to your RJ45 interface names they line up with mine.

the AliExpress product page has a slide with the proper interface order but that's still a bit disappointing, the box has no documentation or they could have got a cheap sticker to put on the panel... ah whatever, I'm nitpicking I guess

 

[blunden]

the AliExpress product page has a slide with the proper interface order but that's still a bit disappointing, the box has no documentation or they could have got a cheap sticker to put on the panel... ah whatever, I'm nitpicking I guess

I had not noticed that. Thanks!

I didn't necessarily expect any documentation. What they could've done though is print the correct numbering on the outside, not just numbering that looks logical but that has no relation to the order the hardware is enumerated. However, based on all the issues some people seem to have with the CWWK N100 and N305 units, this seems like a fairly insignificant issue in the grand scheme of things.

On an unrelated note, I saw that there are actually BIOS settings for controlling whether it should power on after a power loss. They are just hidden pretty deep down in the menus. I think it might've been a few menus in under South Bridge settings or something like that, which surprised me a bit.

 

[intelic76]

Does anyone know what's the name for the SATA power connector on the MB of this mini PC? From what I can see on pictures and searching on the internet, looks like it's PH 2. Can anyone confirm this?

Not sure I'd seen a reply yet on the connector. been trying to find another one myself for a second drive. a little bummed that it only came with one

 

[SlowmoDK]

Not sure I'd seen a reply yet on the connector. been trying to find another one myself for a second drive. a little bummed that it only came with one

Have you tried your asking seller, not for connector name, but for extra sata-adaptor ?
I can't imagine it costing more than a few $

And ya with two internal special sata ports, it's a cop out only to include one adaptor

 

[blunden]

Not sure I'd seen a reply yet on the connector. been trying to find another one myself for a second drive. a little bummed that it only came with one

Just be careful. While the port seems to be reused by other devices, the pinout isn't necessarily the same. I speak from experience.

Have you tried your asking seller, not for connector name, but for extra sata-adaptor ?
I can't imagine it costing more than a few $

And ya with two internal special sata ports, it's a cop out only to include one adaptor

I guess they figured that it only contains mounting for one SATA drive, so why include a second cable.

 

[SlowmoDK]

I guess they figured that it only contains mounting for one SATA drive, so why include a second cable.

True, but with a 3d printer or grinder/dremmel anything is possible

I would cut a exit hole for sata cable and just mount 2nd drive on the outside using 2 screws for each drive, and just realized u cant screw that in

but something like that hehe