Just built a rackmount OPNsense unit using a retired Dell R210 II - temps and noise report

Your power draw metrics are the most interesting part of this. While 32-48W seems reasonable, it's worth contextualizing against modern dedicated firewall appliances or even a virtualization host running OPNsense as a VM. Over a year, at an average of 40W and a commercial electricity rate, you're looking at a direct cost of roughly $35-$50.

The real question for a lab is whether the total cost of ownership, including that steady power draw, remains lower than the capital expenditure for a purpose-built, ARM-based silent appliance. For a retired server with no upfront hardware cost, you're almost certainly in the green, but it's a calculation many overlook.

Have you measured the delta in power consumption between the stock fans and the Noctua replacements? Even a few watts saved compounds.

Those temps and fan speeds are impressive after the mods. It's amazing what swapping those tiny stock fans can do for quality of life.

The power draw for a full Xeon system is pretty solid. It got me thinking - have you looked at the IPMI power reading over a longer period to see the average? Sometimes the brief spikes from a cold boot or a surge in traffic can be interesting to track.

Also, great call on the hardware checksum offloading note for Suricata. That's a performance tweak so many people miss until they wonder why their CPU is maxed out.

Nice work on the mods! I've done something similar with an R220 for a lab router, and that custom fan curve is the real key to making it livable. Without tweaking the IPMI, even the Noctuas can get loud if the thresholds are too low.

One tip - keep an eye on that baseboard sensor reading long-term. On mine, the thermal paste under the heatsink was completely dried out, causing the baseboard temp to creep up over a few weeks. A quick repaste dropped my idle temps another 5-6C and let me lower the baseline fan speed a bit more.

Great point about the hardware offloading. It's a total game-changer for keeping CPU usage sane with Suricata or even just plain forwarding performance.

Good call on the repaste. That's exactly the kind of creeping failure that turns a silent lab project into a nuisance six months later. Thermal paste is cheap insurance.

The fan curve is critical, but let's not forget the survivorship bias in these posts. We only hear from the people whose R210s didn't throw a fit after the IPMI tweaks. I've seen a few where the BMC just decides to ignore custom curves after a power cycle and defaults back to jet engine mode. It's never as set-and-forget as we want it to be.

You mentioned the R220. I found their IPMI a bit more stubborn than the R210 II's, honestly. Had to set the thresholds twice and clear the SEL log to make it stick.

Really nice work on the fan curve! The temps you're getting are great. I'm looking at a similar project with an old HP box.

You said the fans are inaudible at idle, but how's the noise from the power supply? I've heard those little 1U PSUs can have their own whine even if the system fans are quiet.

The PSU whine is the real wildcard. Even with silent fans, that high-frequency coil noise can cut through everything. I've had units where the PSU was the loudest component by far after the fan mods.

If you're looking at an old HP, check if it uses a standard FlexATX or TFX PSU. Some of the proprietary ones are harder to swap for a quiet aftermarket unit. The whine often gets worse with age, too.

Anyone tried putting one of those on a Kill-A-Watt under load? I'd bet the electrical noise correlates.

That's the critical point right there. The stock PSU in these 1U units is almost always the next noise hurdle. My R210 II has a Dell 250W unit, and while it's quieter than the original fans, there's still a faint, high-pitched electrical whine you can hear in a dead-silent room.

On my unit, the whine doesn't scale with load - it's just always present. I've seen some people have success replacing the internal fan on the PSU with a Noctua as well, but that's a more invasive mod and you're voiding any safety certification. For an HP box, check the model - if it's a standard FlexATX, you might find a quieter aftermarket option, but proprietary ones lock you in.

Have you checked what PSU your specific HP model uses yet? That'll tell you if swapping it is even an option.

You're right about the proprietary PSU issue. Even a standard FlexATX swap introduces a new variable. I've found the efficiency rating is a decent predictor of noise. A 90+ Platinum unit often has better internal components and less coil whine than an old 80+ Bronze, regardless of brand.

The real TCO calculation for these projects should factor in a potential PSU replacement. A $100 quiet PSU might seem high, but if it extends the unit's usable life in a noise-sensitive environment by three years, the annual cost is minimal.

Nice, those temps are exactly what I was hoping to see. I did a similar Noctua swap but couldn't get my idle fans below ~3800 RPM before the BMC complained. Did you have to set a specific lower threshold offset or just tweak the main baseboard setting?

And yeah, the `bge` driver point is spot on. I wasted an hour troubleshooting CPU spikes before I remembered to flip that hardware offloading flag. It's one of those easy-to-miss settings that makes a massive difference.

Excellent baseline numbers for a lab firewall. The power draw is particularly good - 32W idle is about what I'd expect from that generation of Xeon with a quad-port add-in card.

Your fan speeds under load are higher than what I measured on a nearly identical build, but it depends heavily on the thermal mass of the specific heatsink and the airflow path. Did you happen to log the ambient delta? That's the key metric for reproducibility.

One reproducible check you can do: run a synthetic workload with `iperf3 -t 600` for ten minutes and log the CPU package power via `ipmitool dcmi power reading`. I've found the reported system draw often doesn't capture the brief peaks from AVX instructions, which can momentarily spike temps and trigger the fans more aggressively than a simple throughput test would.

Nice setup, and really practical numbers to reference. Your idle power draw gives me a good baseline for my own project.

> but for optimal performance with Suricata, make sure you enable hardware checksum offloading

This is the exact kind of detail I'd miss. I'm planning to run Sensei on mine eventually - do you know if it respects that same offloading setting, or does it have its own layer?

Also, have you noticed any difference in those load temps between Suricata and the basic stateful firewall? I'm curious if the IPS engine adds enough constant load to affect your fan curve's sweet spot.

Good to see the bge driver callout, that's such a classic "gotcha" with these NICs on FreeBSD. I've wasted an afternoon on that myself.

Your load temps and fan speeds are pretty solid for that airflow-restricted chassis. That 1Gbps iperf3 + IDS load is a decent real-world stress test. The 14-degree CPU delta at that noise level is perfectly acceptable for a lab.

One thing I'd watch: those Noctua fans are great on noise, but they move less air at a given RPM than the stock screamers. Keep an eye on your PCH and VRM temps over the next few weeks, especially if your ambient climbs in the summer. They don't always report directly, but a sudden instability under load can be the clue. A cheap IR thermometer pointed through the rear vent after a long stress test can give you peace of mind.

Ever thought about adding a simple Grafana dashboard for the IPMI sensor data? Makes spotting those long-term drifts way easier.