It doesn't have an IPMI, but it uses an ASPEED device as a basic GPU to provide that VGA port.
Qotom Denverton fanless system with 4 SFP+
- Thread starter is39
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
Mine doesn't seem to run too hot, usually in the low 40s, including while it is working (not too hard, just OPNSense and Home Assistant). I have though about getting a second one of these and turning it into a 1@ or 2U NAS (make a custom front panel with hotswap bays), but I think a turnkey solution or a used server with the bays already there would be a lot easier.
One thing I thought about was to tune the fan speed for lower noise (it is in an open hallway near the living room) while managing decent temps.
Fan speed? It sounds like you must have the rack mounted model then, because the desktop one is fanless.
My fanless C3758 unit currently reports core temperatures between 44-47°C, most of 44-45°C. Load is pretty light at the moment though.
That thing has a Tjmax of 93°, so there's nothing wrong until the temps are nearing ~80° and cores start to throttle. The throttle log can be checked via the dev.cpu.N.coretemp.throttle_log sysctl on BSD (no idea how/if linux can read that log).
A passively cooled unit will *always* run "hot" in terms of "I burned my finger". Anything above ~40° feels hot and can cause burns if given enough time, so a case temp of 50-60° will feel *very* hot but the CPU is still perfectly fine.
I wouldn't put heat-sensitive SSDs in there though - let alone space heaters like Samsung NVMEs that already struggle to stay cool in a ventilated case... Some low-power drives are fine though. I never had issues with e.g. WD blue in other passively cooled systems like the quad-port N5105 appliances.
A passively cooled unit will *always* run "hot" in terms of "I burned my finger". Anything above ~40° feels hot and can cause burns if given enough time, so a case temp of 50-60° will feel *very* hot but the CPU is still perfectly fine.
I wouldn't put heat-sensitive SSDs in there though - let alone space heaters like Samsung NVMEs that already struggle to stay cool in a ventilated case... Some low-power drives are fine though. I never had issues with e.g. WD blue in other passively cooled systems like the quad-port N5105 appliances.
Yes, that was my impression too.
I bought a cheap IR thermometer to get a rough idea of the temperature of the case, and especially my transceivers. Unfortunately, the transceiver I have it in (besides my DACs) doesn't report its own temperature. It has been running perfectly fine so far though so it's hopefully all fine.
Yeah, I went for an old MLC SATA SSD as I don't need massive storage speed in VyOS and it's power consumption is supposedly very low.
I also removed one of my two RAM sticks to reduce power consumption since I don't need more than 16 GB in a router. Maybe if I start running a bunch of containers, but I'm currently not.
Code:
655 root@gw-host:~ # sysctl dev.cpu | grep temper
dev.cpu.7.temperature: 44.0C
dev.cpu.6.temperature: 46.0C
dev.cpu.5.temperature: 45.0C
dev.cpu.4.temperature: 42.0C
dev.cpu.3.temperature: 44.0C
dev.cpu.2.temperature: 43.0C
dev.cpu.1.temperature: 45.0C
dev.cpu.0.temperature: 46.0C
656 root@gw-host:~ # smartctl -iA /dev/nvme0 | grep -E 'Model|Temperature:'
Model Number: WDC PC SN530 SDBPNPZ-256G-1114
Temperature: 52 Celsius
657 root@gw-host:~ # smartctl -iA /dev/nvme1 | grep -E 'Model|Temperature:'
Model Number: WDC PC SN530 SDBPNPZ-256G-1114
Temperature: 53 Celsius
658 root@gw-host:~ # ifconfig -v ix0 | grep -E 'vendor|temp'
vendor: FS PN: SFP-10GSR-85 SN: F2210814302 DATE: 2022-07-29
module temperature: 46.98 C voltage: 3.25 VoltsThat unit is currently running 6 jails: gateway/firewall (PF), nameserver (nsd + unbound), dhcp, zabbix, PostgreSQL master node and angie reverse-proxy.
System load averages are hovering around 0.2-0.4, mostly because of the zabbix server and PostgreSQL.
The FS transceivers have a temperature range up to 70°C, so still way over 20° headroom.
I do have the rack mount version, I suppose I could have mentioned that. I have tempos similar to everyone else from what I can tell. I wonder what Antioch's temps are that he is concerned with this. To his point though. I would also be interested in some sort of tuning guide on these devices.Fan speed? It sounds like you must have the rack mounted model then, because the desktop one is fanless.
My fanless C3758 unit currently reports core temperatures between 44-47°C, most of 44-45°C. Load is pretty light at the moment though.
Mine are also in the mid-to-upper 40s. The case is hot to the touch, but I take sko's point about one's fingers being a poor temperature measuring device.
I am still not able to get it to go below C-state C2 which is what triggered me to wonder about the temperature to begin with. Has anyone gotten the CPU to hit lower C-states?
Edit: Am using the passively cooled case.
I am still not able to get it to go below C-state C2 which is what triggered me to wonder about the temperature to begin with. Has anyone gotten the CPU to hit lower C-states?
Edit: Am using the passively cooled case.
Additional question: Has there been any discussion about RJ45 SFP+ module chipset compatibility? I picked up a Marvell based device, and the downlink is completely unstable in the Qotom. However, it works fine in one of my switches. I wonder if a Realtek or Broadcom based RJ45 SFP+ adapter would fare better?
Edit: Am currently running proxmox.
Edit: Am currently running proxmox.
IIRC the low-power SoCs never supported more than C1/C2. At least I've never seen any other states on Atom or even Xeon-D...
As was already discussed multiple times: RJ45 SFP+ adapters are COMPLETELY out of spec. Especially on low-power appliances/SoCs they usually won't work, as they are 2-3x over the power budget for SFP+.
So like any non-standard device it should be considered purely coincidental if they work at all...
I really don't get why on this forum there is such a fierce opposition to just use (MUCH cheaper) fiber and finally abandon ancient copper for anything above 1/2.5G...
For residential setups, many ISPs are now providing multi-gig Internet... and the connection from their ONT or cable modem terminates to 2.5/10GBaseT.
Certainly, I bought my Qotom box on the expectation that in the long term, my WAN would be 10GBaseT and my LAN would be DAC to my switch. I have the LAN part set up that way at the moment, but I haven't upgraded to my ISP's >1 gigabit service so I'm using one of the copper ports.
I guess the 'better' way to do this would be to set up a VLAN on my (16-port) switch, plug the ISP 10GBaseT into a copper port, then run a DAC from an SFP+ port to the Qotom box, but that's... using up two scarce ports on an expensive switch.
I would also note that many computers with 10G ports use copper - all the Apple stuff (Mac studio, Mac Pro, some iMacs, etc) uses Aquantia 10G copper controllers, and the crazy expensive enthusiast PC motherboards with 10G onboard also use copper. And, one thing that I've noticed that bothers me - recent enthusiast motherboards are getting rid of PCI-E slots, so... less and less likely that you'll be able to just add a 10G SFP+ card.
Thank you for the information, I was quite unaware. I will go back and read the entire thread from the beginning.
Oh, I would love to ditch the copper. However, I'm in exactly the situation that Vivien describes:
I don't think my ISP provides an SFP-ONU for 10G connections (it seems they only do for 1G), which would be the ideal solution. I have just opened up the EPON ONU and discovered that it contains an SFP module. I am considering removing it and plugging it directly into the quotom unit and seeing what happens. However, tinkering with SFP and Fiber is far beyond my knowledge, and I assume it won't work.
The other thought I had was to see if I can grab a SFP-ONU with similar specs and a web management interface off of AliExpress, change the ONU ID to match the ISP provided ONU and see if that works. But again, I'm way out of my depth here. Any comments or suggestions are welcome.
In the end I may end up doing something similar to this:
Except my switch is all SFP, so I'd still have to use the RJ45 SFP adapter.
Is your ISP large enough to have people talking about stuff online?
Here, there's one very large ISP that used to do SFPs that then went into their all-in-one gateways, and it's well-understood on forums how to take those SFPs out, put them in something else, tag the proper VLANs, etc. (And then this ISP moved to a newer generation of hardware without the SFPs)
My rough sense as someone who's read plenty of forum threads but never dealt with this himself - this doesn't require any knowledge of SFPs/fiber/etc - it just requires knowing the right networking parameters. And obviously you can't ask the ISP...
You might try this solution if you are on XG-PON. Assuming you have AT&T or Frontier, the step-by-step tutorial is all there to config it. Another ISP, it will probably also work but will take experimenting to find the settings. They have a Discord server that can probably help with that.
Last edited:
I think a large percentage of us already know that, but there are situations where hey might be your only reasonable option. I'm for instance using one for my WAN. While running fiber all the way up to my apartment might be possible, it wouldn't necessarily be cheap or easy to do. Using one of the lower power ones rated for 100 meters allows me to get full speed over existing cabeling. So far so good. I might revisit running fiber at a later date though, we'll see.
For my LAN it's DACs and fiber transceivers all the way, apart from PoE powered access points.
Did you enable flow offloading? I didn't do too much tweaking on mine (C3758) and I got 9431/9128 on my latest speedtest.net test.
Come to think of it, I also increased the ring buffers on the interfaces to 4096 and enabled gro and gso offloading.
Ya I also prefer sfp+ cage then copper rj45port, but sadly most consumer base device are copper rj45, my country almost all telco provide ont with copper rj45 out. Only 1 Telco do have the ONU offering. End up I got to use sfp+ to copper rj45 sfp+ module.
We get our termination directly via (BiDi) SFP+ transceivers we can simply put in our switches; no intermediate hardware required. OTOH we also have proper active (ethernet) termination, not PON...
Regardless of that, you can simply connect the uplink (also from your media-converting ONT) to a switch first via an access port to a dedicated VLAN for the WAN uplink (disable STP, VTP etc on that VLAN to not spam your provider network with such broadcasts!). This way it's also trivial to configure redundant routers - simply connect both their uplinks to that WAN VLAN.
I've been doing that with all uplinks for years, regardless of the actual medium or possible intermediate hardware (e.g. modems in passtrough-mode on ancient copper lines...).
Switches are usually more tolerant towards out-of-spec power requirements as they have a much higher power budget - this might change with current-generation switches, however. E.g. A 12-port SFP+ switch that is rated at 18-20W will in no way allow 2-3W modules to work, simply because the power budget doesn't suffice. The only viable and "supported" solution is either a switch with RJ45 10G ports or a media converter with RJ45 and SFP+.
Those field-terminated connectors are perfectly fine for that last mile and usually have <0.3dB of insertion loss.
Regardless of that, you can simply connect the uplink (also from your media-converting ONT) to a switch first via an access port to a dedicated VLAN for the WAN uplink (disable STP, VTP etc on that VLAN to not spam your provider network with such broadcasts!). This way it's also trivial to configure redundant routers - simply connect both their uplinks to that WAN VLAN.
I've been doing that with all uplinks for years, regardless of the actual medium or possible intermediate hardware (e.g. modems in passtrough-mode on ancient copper lines...).
Switches are usually more tolerant towards out-of-spec power requirements as they have a much higher power budget - this might change with current-generation switches, however. E.g. A 12-port SFP+ switch that is rated at 18-20W will in no way allow 2-3W modules to work, simply because the power budget doesn't suffice. The only viable and "supported" solution is either a switch with RJ45 10G ports or a media converter with RJ45 and SFP+.
You can get drop-cable kits with one side pre-terminated and a field-spliceable (mechanical splicing; no polishing required) connector for the second end for much less than a copper cable would cost for that length. E.g. 50m are usually <20$; 100m <40$...
Those field-terminated connectors are perfectly fine for that last mile and usually have <0.3dB of insertion loss.
Yeah, I configure the ring buffer on both WAN and LAN. Definitely try flow offloading.Ok I go check on the flow offloading. , rx and tx buffer, I only configure on eth0 . Guess I got to configure that on my lan interface too.
No, definitely don't use jumbo frames on WAN.