Posted: 2026-07-03
Yesterday, I ran some tests comparing different fan configurations. Short depth, micro-ATX capable Supermicro 1U chassis come with either a blower or four counter rotating fans. I've also had some struggles sourcing heatsinks, but found that I could get 1U heatsinks with fans for less than a nice Dynatron 1U heatsink.
This made me wonder if just the CPU fan was sufficient. It is... kind of.
I have a shed/office that can get pretty toasty. Certainly warmer than Neutron. I've been testing the new 64GB memory + 2TB configuration. With CPU maxed out, the system was not overheating with just the CPU fan installed. However, there was no top on the case -- not very realistic for a rack installation. When I put the top cover on, the system would occasionally overheat, throttle, and get back to normal. It was enough warmer in the office that I assume it would be okay in the datacenter, but the margins were too small for me to be comfortable with it.
I reconfigured the system a few ways while running. It's actually pretty impressive how the motherboard will "re-learn" and adjust to different fans on the fly.
Here's some figures, very incomplete, with different configurations.
75-76W slowservers0# sysctl -a | grep degC hw.sensors.sdtemp0.temp0=38.25 degC hw.sensors.sdtemp1.temp0=38.50 degC hw.sensors.sdtemp2.temp0=38.25 degC hw.sensors.sdtemp3.temp0=38.75 degC hw.sensors.nvt0.temp1=36.75 degC hw.sensors.nvt0.temp4=62.25 degC hw.sensors.km0.temp0=62.50 degC
73W slowservers0# sysctl -a | grep degC hw.sensors.sdtemp0.temp0=43.00 degC hw.sensors.sdtemp1.temp0=40.25 degC hw.sensors.sdtemp2.temp0=39.50 degC hw.sensors.sdtemp3.temp0=40.00 degC hw.sensors.nvt0.temp1=37.25 degC hw.sensors.nvt0.temp4=62.75 degC hw.sensors.km0.temp0=55.50 degC
75-76W
76-77W
As the office was generally getting hotter as tests went on, I re-verified 72W draw with blower + CPU fan. Now I think just the CPU fan by itself uses the least power, but it can end up throttling, so it's not a viable option.
I think the range is kind of impressive, especially considering I'm using a 35W TDP CPU which is way lower than most configurations.
I also found that setperf=1 used 58W when maxed out. But waiting a few minutes for the system to re-learn the thermal balance, it ended up at 55W. So the fan different alone, with a blower + CPU fan, is at least 3W over idle. A notable amount in my opinion. We use a lot of power just to get rid of the power that we've used! I've seen the blower use about 9 watts at max speed, though it was not at max speed even with the CPU maxed out. It would use 9 watts in an "emergency mode" or at startup.
Interestingly, the small fans use more power than the single blower. I should've known that. I guess some earlier testing, even less scientific than this, made it seem that the blower used more power. The blower is way quieter, too.
While I like the redundancy with the counter-rotating fan setup, with the heatsink+fan there's still some redundancy as either one is mostly adequate. But together, they seem to benefit eachother.
I've also finished putting together "Half," a VPS host for the office rack. 1.5U chassis that fit my other cut down heatsink without needing the fins mashed. It has 32GB of memory and a 1TB Samsung 850 PRO. While the CPU is the same, it seemed to keep itself cool almost effortlessly. The two larger fans appear to be far more efficient and they're fairly quiet.
I realize now that 1U is pretty suboptimal as far as cooling goes, which ends up using more power for no good reason. The blower is decent, but it still seems less efficient getting heat out of a 1U case than a 1.5U case. While my power figures are pretty meager, a full rack, or perhaps a full circuit, of servers makes a pretty big difference.
The other thing is that a 1.5U (or 2U) short depth chassis is much more stout with railless mounting than a 1U chassis of the same depth. Now the 1.5U chassis I got from a tech thrift store was pretty perfect and may be hard to find again, with the motherboard just barely clearing the case. But avoiding rail cost is nice, especially while having a more efficient system.
If I could snap my fingers and have a fully loaded cabinet, I might choose 1.5U chassis over 1U. For the current 1/3 cabinet, probably not worth it since space is at more of a premium. But it seems like it'd be about the same cost, or maybe slightly less, to use taller hosts that move air more efficiently. And not paying for rails is nice! Which, I'm not now, but I probably should...
