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.
Hi, "mount the FAN directly on the side between side cage and PSU " <-- did this, didn't seem to help... Used the included foam and mounting tape for baffles.
Anyone also crazy enough to do this? It took some mods but it fits:
That's a 18tb 3.5" drive, but it's not detected. That slot is meant for 2.5" drives, so I think only 5v power is provided from motherboard. I can give it the 12v from the PSU, but any ideas how to give it both 12v and 5v?
That's a 18tb 3.5" drive, but it's not detected. That slot is meant for 2.5" drives, so I think only 5v power is provided from motherboard. I can give it the 12v from the PSU, but any ideas how to give it both 12v and 5v?
I also am seeing the same behaviour running opnsense and 10G SFP's.
It seems there is some sort of weird receive issue.E.g it seems that the qotom is able to push 10G outbound from itself no probs with a single iperf stream.
However reverse significantly worse for single stream. with multi stream I am able to hit ~7Gbits
From my PC to qotom on 10g network (using bidir with iperf3)
Code:
[ 7][RX-C] 4.00-5.00 sec 1.09 GBytes 9.33 Gbits/sec
[ 5][TX-C] 5.00-6.00 sec 119 MBytes 1.00 Gbits/sec 0 1.12 MBytes
Code:
ix0: flags=1008843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST,LOWER_UP> metric 0 mtu 1500
description: LAN (lan)
options=4e03f2b<RXCSUM,TXCSUM,VLAN_MTU,JUMBO_MTU,TSO4,TSO6,LRO,WOL_UCAST,WOL_MCAST,WOL_MAGIC,RXCSUM_IPV6,TXCSUM_IPV6,HWSTATS,MEXTPG>
media: Ethernet autoselect (10Gbase-SR <full-duplex>)
status: active
nd6 options=21<PERFORMNUD,AUTO_LINKLOCAL>
Code:
[ 5] 7.00-8.00 sec 132 MBytes 1.10 Gbits/sec 0 1.30 MBytes
Code:
ast pid: 88544; load averages: 0.73, 0.55, 0.49 up 2+23:41:35 15:27:10
504 threads: 11 running, 418 sleeping, 75 waiting
CPU: 1.2% user, 0.0% nice, 10.8% system, 12.5% interrupt, 75.5% idle
Mem: 104M Active, 2933M Inact, 2490M Wired, 56K Buf, 26G Free
ARC: 1214M Total, 803M MFU, 226M MRU, 5865K Anon, 23M Header, 153M Other
915M Compressed, 2231M Uncompressed, 2.44:1 Ratio
Swap: 16G Total, 16G Free
PID USERNAME PRI NICE SIZE RES STATE C TIME WCPU COMMAND
12 root -56 - 0B 1088K CPU2 2 16:37 100.00% intr{swi1: netisr 2}
Code:
[ 5] 9.00-10.00 sec 1.09 GBytes 9.34 Gbits/sec
Code:
last pid: 92314; load averages: 0.63, 0.54, 0.49 up 2+23:42:11 15:27:46
504 threads: 9 running, 419 sleeping, 76 waiting
CPU: 0.1% user, 0.0% nice, 9.5% system, 6.3% interrupt, 84.0% idle
Mem: 104M Active, 2933M Inact, 2487M Wired, 56K Buf, 26G Free
ARC: 1215M Total, 803M MFU, 232M MRU, 425K Anon, 23M Header, 153M Other
916M Compressed, 2232M Uncompressed, 2.44:1 Ratio
Swap: 16G Total, 16G Free
PID USERNAME PRI NICE SIZE RES STATE C TIME WCPU COMMAND
11 root 187 ki31 0B 128K CPU1 1 70.5H 92.82% idle{idle: cpu1}
11 root 187 ki31 0B 128K CPU3 3 70.6H 92.04% idle{idle: cpu3}
11 root 187 ki31 0B 128K CPU2 2 70.5H 91.56% idle{idle: cpu2}
11 root 187 ki31 0B 128K CPU4 4 70.4H 89.28% idle{idle: cpu4}
11 root 187 ki31 0B 128K CPU7 7 70.5H 88.73% idle{idle: cpu7}
11 root 187 ki31 0B 128K RUN 6 70.5H 86.72% idle{idle: cpu6}
11 root 187 ki31 0B 128K RUN 5 70.5H 85.06% idle{idle: cpu5}
90746 root 68 0 19M 7856K sbwait 5 0:05 63.03% iperf3{iperf3}
11 root 187 ki31 0B 128K CPU0 0 69.7H 52.48% idle{idle: cpu0}
12 root -60 - 0B 1088K WAIT 0 15:40 47.11% intr{swi1: netisr 0}My 1U 3758R just got delivered. On a quick test with the SODIMM I have lying around on various equipment which I borrowed from. Here is what the result are.
Considering the low amount of sample pool, the conclusion I can narrow is that the Double sided RAM are more likely to work. The Crucial/Micron RAM that fail is working fine on my other equipment(retested after failed in Qotom).
With the test as a baseline, I'm more willing to test this pair of TimeTec ECC RAM which appear to be based on SK Hynix chips. I'll find out in a few days if this pair of 2x16GB sticks will work.
PS: if somebody feel like starting a Google spreadsheet on this, you are welcome to take this data. Considering all the equipment STH are running the test, maybe somebody from STH should start it. Just one file, and each tab is for a series of equipment that shared the same board design.
Considering the low amount of sample pool, the conclusion I can narrow is that the Double sided RAM are more likely to work. The Crucial/Micron RAM that fail is working fine on my other equipment(retested after failed in Qotom).
With the test as a baseline, I'm more willing to test this pair of TimeTec ECC RAM which appear to be based on SK Hynix chips. I'll find out in a few days if this pair of 2x16GB sticks will work.
PS: if somebody feel like starting a Google spreadsheet on this, you are welcome to take this data. Considering all the equipment STH are running the test, maybe somebody from STH should start it. Just one file, and each tab is for a series of equipment that shared the same board design.
No. That's a typo. All 4 SFP+ ports from C3758R SoC are 10GbE Intel X553.
On Qotom's website you can find a BIOS flashing utility, which allows you to config 2 of these ports to 2.5GbE.
Just wanna try and see what is the actual total bandwidth look like, so I set this up in proxmox.
I also setup another 8 bridges where each bridge only has one interface, excluding the one I login. Then I setup 8 LXC each is a debian 12 built with only one iperf3 installed. Then I make a directly for each port to another. eg: ix0 to ix1, ix2 to ix3, igc0 to igc1 and igc2 to igc3 (or equivalent of). And here is a result.
The result as you can see, the total throughput probably a little over 10gbit/s
I have tried setting auto-negotiation to manual, result is pretty much the same. But the CPU utilization is pretty high, about 80%+.
I also setup another 8 bridges where each bridge only has one interface, excluding the one I login. Then I setup 8 LXC each is a debian 12 built with only one iperf3 installed. Then I make a directly for each port to another. eg: ix0 to ix1, ix2 to ix3, igc0 to igc1 and igc2 to igc3 (or equivalent of). And here is a result.
The result as you can see, the total throughput probably a little over 10gbit/s
I have tried setting auto-negotiation to manual, result is pretty much the same. But the CPU utilization is pretty high, about 80%+.
The system is C3758R based with one 8GB stick of non ECC RAM and a stick of NVMe gen 3. Two of the SFP+ I have DAC cable and the other two has SFP+ LR.
Total power usage with idle CPU and all 9 ports occupied (no traffic): 28watts
Total power usage with all 9 ports occupied and full traffic (previous test): 38watts
Total power usage with max CPU (sysbench) and all 9 ports occupied (no traffic): 34watts
I suppose this could go above 50watts if you have dual ECC RAM sticks, NVMe gen 4, a spinning HDD and all SFP+ are optical. The PSU on this only go up to 50w, so this could be a problem.
Total power usage with idle CPU and all 9 ports occupied (no traffic): 28watts
Total power usage with all 9 ports occupied and full traffic (previous test): 38watts
Total power usage with max CPU (sysbench) and all 9 ports occupied (no traffic): 34watts
I suppose this could go above 50watts if you have dual ECC RAM sticks, NVMe gen 4, a spinning HDD and all SFP+ are optical. The PSU on this only go up to 50w, so this could be a problem.
If I read my iperf3 bidirectional results correctly, I get roughly 15 Gbit/s in a quick test to a speedtest server on the internet (with multiple streams) so it seems like there might be some overhead in your setup if the idea is to test total throughput.
This was with a 10 Gbit/s computer as the client to let the C3758 (non-R) based Qotom box act solely as a router. A local server would obviously also be a more accurate test, but I don't feel like messing with it too much now that it's my main router.
Good news to report, the TimeTec ECC RAM 32GB (2x16GB) I ordered arrived. For under $80 ECC RAM, I guess it's alright. Going through Memtest86 right now, and just passed the first round. The RAM training before BIOS show up is about 2 mins, well it's more than a minute, but not more than two. Didn't exactly measure the time. So, if anyone want two stick of 16GB ECC, this should be good. Price is just a bit more than a stick of Crucial 16GB ECC. Of course if you want to leave yourself an option to go 64GB and only want to use one stick of 32GB, then going Crucial/Micron maybe better.
I use OPNSense. I want to change my LAN interface from IGC0 to IX0. How can I do this without losing access to web interface? I have both ports connected to the same switch. If I just change the assignment in SSH shell and then disconnect the IGC0 connection will that work? Really worried about losing web access to interface.
Thank you
Thank you
Why not do it from the local console? There's a CLI option right when you log in to change the interface assignments...
Realistically, I suspect you could do it in the web interface and it would just work fine, but like you, I'd like to have a backup plan first.
I think you can do it in bridge. Make a bridge and assign ip to the bridge. You can have one ip on the physical onterface and diff ip on the bridge.
Infact it is better to use bridge instead of actual physical interface. It is more flexible in the future when making changes
NO, don't do that! That's not what bridges are for - you will create a loop and the switch will most likely disable both interfaces.
Just use a lagg interface configured as failover, use the ix interface as primary and igc as fallback.