The Silence of the Fans

The good thing about using a server-grade machine as your desktop is having raw computing power at your fingertips. The downside is living next to a machine that sounds like a fleet of quadcopters taking off. Luckily, loud server fans can be replaced with quieter units if you know what you’re doing.

Servers are a breed apart from desktop-grade machines, and are designed around the fact that they’ll be installed in some kind of controlled environment. [Juan] made his Dell PowerEdge T710 tower server a better neighbor by probing the PWM signals to and from the stock Dell fans; he found that the motherboard is happy to just receive a fixed PWM signal that indicates the fans are running at top speed. Knowing this, [Juan] was able to spoof the feedback signal with an ATtiny85 and a single line of code. The noisy fans could then be swapped for desktop-grade fans; even running full-tilt, the new fans are quieter by far and still keep things cool inside.

But what to do with all those extra fans? Why not team them up with some lasers for a musical light show?

50 thoughts on “The Silence of the Fans

  1. Part of me says “small 555 circuit coulda done this easy.” The other part of me is like “using the sound output to generate a square wave? I didn’t think of that. Cool.”

    Do the fans have equivalent CFM rating?

    1. When it just needed a non-changing PWM, I thought 555, too. But a tiny drop in board for $2? Eh, why not. It’s not like he needs a thousand of them.

      As for the airflow, I’m sure it’s less. Maybe not less with quiet fans at 100% than the loud fans at 27%, but they almost certainly can’t match the original fans at 100%. *If* he knows he won’t need that speed *and* doesn’t mind not being informed when a fan fails, it’s a good plan to reduce noise/save sanity. There are trade-offs, but he must have deemed them acceptable.

      1. If the replacement fans are tach fans, he could always use the ATiny to give him a heads up when the fan ceases reporting its speed. A loud piezo should be sufficient.

  2. Dell PC’s are notorious for noise – even many of the desktop PC’s. I had one that had a centripetal fan and I couldn’t code on the PC because the changing fan noise was too distracting.

    I now have a Dell desktop that I had to rebuild the cooling for because it was a small form factor PC case with a dual core CPU and I went and put a quad-core in it.

    Here’s what did / learnt.

    Fans aren’t Fans! The fans installed in this article have a high noise rating for a low CFM airflow. Fans that are low noise and high CFM are designed differently and have a longer tapering off of the outer trailing edge of the fan blades like the old genuine Intel fans.

    You need two fans. The CPU fan has to go just fast enough to distribute the heat away from the CPU and into the case cavity. Then you can use a slower and quieter fan to bring cooler air into the case. I also pot a dust filter on the case fan so it takes 1 minute to clean the dust filter rather than half an hour to clean out the case and heat sink.

    Fans are only half the equation. The heat sink efficiency is equally (or more) important.

    Copper and aluminum have vastly different properties.

    Copper is a very good thermal conductor to draw heat away from the CPU but it heats up quickly as it can’t store much heat. Aluminum on the other hand is a poorer thermal conductor BUT it can store much more heat. That why the old Intel radial heat-sinks had a copper slug in the middle of the aluminum heat sink. The copper slug draws the heat away from the smaller surface area of the CPU and then conducts it to a lager surface area of the aluminum.

    Next is the thermal conductivity to air (radiation). Surface area plays a big roll in this but it’s more then just how big the heat sink is. A polished surface has the lowest surface area and then better still is an abraded surface and for copper an oxidized surface is better again and for aluminum an anodized surface is best.
    I ended up buying a polished radial copper heat-sink and oxidized the copper in ammonium persulphate and it works wonderfully and low speed and low noise for a quad core in a small form factor case.

    1. Don’t forget that size and RPM are not the only things that matter. You also have to look at pressure at RPM, you may have a fan that is running quiet and think it has the same CFM but since it is running at a higher pressure the actual delivered CFM is lower.

    2. “Copper is a very good thermal conductor to draw heat away from the CPU but it heats up quickly as it can’t store much heat. Aluminum on the other hand is a poorer thermal conductor BUT it can store much more heat. That why the old Intel radial heat-sinks had a copper slug in the middle of the aluminum heat sink. The copper slug draws the heat away from the smaller surface area of the CPU and then conducts it to a lager surface area of the aluminum. ”
      That is actually a load of horse shit.

      1. Copper block of same volume as aluminum can store more energy (raising the same amount of Kelvins) even thou it has more then twice lower specific heat, it is more than 3 times denser.

      2. Heat sinks shouldn’t store energy, they should dissipate it. So your comment about storing energy is pointless?

      3. Intel used copper slug and aluminum radiator for cost/performance factor.
      Aluminum is a great material for fins, it is easily machinable, cheap, light, it forms a thin layer of oxides.

      There were pure copper radiators and they didn’t fail on performance side, Aluminum is just cheaper while combo of copper heat pipe aluminum fin can work nearly as good being much cheaper.

      1. You have this completely confused as you your are attempting a static analysis. The reality is a CPU heat sink is dynamic.

        1) Take two long rods in a vacuum, one copper and one aluminum. Heat them both equally from one end. Observe the copper rod is now hot at the other end and the aluminum is NOT! Given that the same energy was applied to both rods, how would you then account for the conservation of energy?

        2) See above.

        3) If what you say is correct then why have a copper slug at all?

        4) RE: Pure copper radiators – go google thermal emissivity copper oxide aluminum oxide – copper oxide is a magnitude better that aluminum

        Aluminum Highly Polished 0.039 – 0.057
        Aluminum Rough 0.07
        Aluminum Commercial Sheet 0.09
        Aluminum Heavily Oxidized 0.2 – 0.31
        Aluminum Anodized 0.77

        Copper Polished 0.023 – 0.052
        Copper electroplated 0.03
        Copper heated and covered with thick oxide layer 0.78

        In case you missed it –

        Aluminum Rough 0.07
        Copper heated and covered with thick oxide layer 0.78

          1. it does that by convection, a molecule of air bumps into the radiator and bounces off with a little bit of the energy.

            emission is when photons are emitted directly from the surface to carry away energy, it is also largely ineffectual below a couple hundred kelvin.

          2. PS: Emissivity and radiation are different things.

            Heat is simply electro-magnetic waves in the infra red region. Because a material can emit these waves, that same material will re-absorb a proportion of the emitted waves. Emissivity is the total difference between emission and re-absorption.

          3. to clarify:
            yes there is some heat loss from radiation when looking at the whole system, but the process of convection does not involve emissivity, look up the equations and it is there plain as day.

          4. “The heat has to radiate into the air before the fan can move the heated air away.”

            Air is transparent to IR at these length scales and does not pick up the heat. Heatsinks inside a closed box are cooled by convection (forced or free air movements), not radiation.

            Surface roughness is actually worse for this kind of heat transfer because it increases the boundary layer thickness and prevents the incoming air from actually touching the heatsink.

          5. you do realize that the article you linked states that

            “As expected the percentage of heat dissipated via radiation increases as the temperature of the heat sink increases. At higher temperatures the heat dissipated via radiation is over 5% of the total. In certain critical situations this may mean the difference between meeting or not meeting the temperature rating of the component being cooled.”

            which is 100% consistent with the statements i made earlier that emission wasnt the primary cause of cooling,
            nor is it a part of the process of convection, emissivity is not present in any of the equations i can find for convective cooling.
            no one claimed that emission cooling didn’t happen, only that the emissivity of a material is not related to its convective cooling ability.

          6. What’s interesting is how seemingly the makers of coolers and fans also have the hardest time figuring it all out, they keep releasing new models and often enough they aren’t better than the previous model even when using more expensive materials. Sometimes you even see the release of a liquid cooler that is no better than a normal air cooler, it’s weird.

            And that’s across the entire industry it seems from my observations.

          7. @[oodain]

            There you go again!!!!

            Your quote states “radiation”. Your next sentence states “emission” and then you state “emissivity”.

            NONE of these terms are interchangeable and until you can understand that we are not even talking about the same things. You are simply arguing with your self.

          8. they weren’t used interchangeably, at all.

            emission is the process in which radiation (including the infrared you talked about) is released from a material in the form of photons, emissivity is a figure that describes how much radiation, compared to a perfect black body, a given material releases at a given temperature.

            you are right that they aren’t interchangeable but they are highly related, to see them mentioned in the same sentence shouldn’t be a surprise.

          9. ROB : “Heat is simply electro-magnetic waves in the infra red region. ”

            You’ve complained that you think others are conflating emissivity with radiation, and meanwhile you conflate heat (thermal energy) with infra-red. You’re quite wrong, I’m afraid. Heat can be transferred via EM, it can occur at all frequencies (depending on temperature of course), but this is not the ONLY form of heat transfer, nor the dominant one in play for most heatsinks.

            If heat were only transferred by radiation, then cooling of electronics in space would be very simple – the surroundings are ~4K, right? But it’s actually a hard problem to move the heat away because there’s no air to provide convection.

            Your own example of changing the surface finish of a polished copper heatsink would have been a good example to check how much cooling is performed by radiation. If you had measurements from the polished state to compare to the high-emissivity state, I am quite confident there would be only a small change. Now try turning the fan off…

        1. >”You have this completely confused as you your are attempting a static analysis. The reality is a CPU heat sink is dynamic. ”

          The system will eventually reach thermal equilibrium, where the static case applies.

          The specific heat capacity of the heasink material doesn’t matter. Its emissivity doesn’t matter – what matters is the contact area with incoming air.

          1. > what matters is the contact area with incoming air.
            Which brings us back to boundary layers.
            Some turbulence is beneficial so the flowing air gets:
            – more mixing with the heated boundary layer air,
            – more contact with the heatsink surface,
            instead of relying upon conduction between the headsink surface > boundary layer > flowing air.
            Often a pin heatsink will cool more than a plate.

            Also, the greater the turbulence, the greater the noise.

            Minor tip for noise: if you can do without the screens on the case intake/exhaust, lose them.
            A curved intake/exit edge, like the curved edge surrounding some round speakers, can smooth the flow and noise of intake (smoother, hence quieter, flow into the case to the fan blade) and exhaust injection into the surrounding air is smoother/quieter.
            Those are minor compared to the fan blade noise and that of the air flowing past the braces holding the fan motor, but it all contributes and adds up.

            I wonder why we’re not seeing heat-pipes taking the heat of the CPU or GPU to outside of the case (with insulated lines), then heatsinks and/or fans on their radiators.

  3. “The good thing about using a server-grade machine as your desktop is having raw computing power at your fingertips. The downside is living next to a machine that sounds like a fleet of quadcopters taking off. ”

    I find that more so true of the rack mounted equipment, than the regular case servers.

      1. There are plenty of reuses for server fans, I mounted one of these blower fans under my barbecue for blowing fresh air from the bottom. It works like a charm.
        You can also race styrofoam boats by putting server fans and old batteries, You gotta carefully select the power/weight ratio (more batteries do not always make it faster due to the added weight and drag), but it’s plenty of fun.

  4. The whole idea behind server grade hardware is that is is designed to be on 24/7/365. The fans are built to much better specs than your typical desktop fan. They are generally ball bearing fans and not sleeve bearing fans. The last servers I used all had thermostatically controlled fans, so if the fan did not need to be on full speed, it would not be on full speed. You have just un engineered a rather elegant solution.

    1. that might have been the case in the 90’ies or earlier, i agree on the duty cycle but i have never seen a server fan that used anything i haven’t seen in normal desktops, i have taken apart a fair amount of servers, all of the ones i have seen used standard PWM fans, they might be insanely powerful and in a weird form factor but in essence they arent different.

      that is unless they are on/off fans for gross cooling, those usually have an independent temp sensor, i have also never seen them used as primary cooling and i have never seen them used in the server itself, always on a rack or a room and usually they are set to only kick in at emergency temperatures, think hot summer day when everything is running at full and the AC is out.

      ball bearings are also common in even fairly cheap consumer fans, the bang for your buck category today includes ceramic bearings.

    2. Modern desktop fans – the quiet ones – use fluid film bearings and/or magnetic bearings which can easily exceed 100,000 hours MTBF.

      Ball bearings don’t have a long life, because they can’t have very good tolerances.

  5. Another option is to relocate the workstation to somewhere where the noise it makes isn’t a problem. I got the idea from reading about how recording studios deal with noisy workstations.

    Mine sits in my loft. The trickiest connection to make long distance is the display. A combination of extra long displayport cables and repeaters lets me have a 2560×1600 display about 20m away in my lounge. Since I put the setup together it looks like things have progressed a bit and 4k display ought to be an option now. USB3 using active cables with repeaters is also possible over that distance, and a USB combination optical drive and card reader provides everything that I’d actually want to be close to the machine for. Sound is via the displayport. USB webcam with an alright mic does the other side of the sound setup.

    The loudest sound I’m aware of from the whole setup is a slightly whiny power supply in the Dell monitor. Meanwhile the machine with six fans and eight spinning disks makes my loft sound like a server room. Gets to about 40C up there in summer, but everything’s carried on working for the last couple of years.

      1. I hardly need the button – wake on lan from my phone and software suspend button.

        But I do confess to having Intel AMT enabled (so it’s probably not really my computer etc.), which allows power cycling via its web interface.

  6. You could get real smart and add a temperature sensor in to the mic and adjust the fan speed based on the temperature, then the unit is not under load and not needing as much air flow the fans would throttle back. add to that a sense to check that the fan was running and flash the LED if there was an issue.

    Now for a but about thermal dynamics, Copper sinks heat faster than most other materials, gold also sinks heat prety quickly but is somewhat cost prohibitive :) Using fins on of copper will dissipate heat quickly however thin copper fins are not exactly mechanically stable or not as stable as the same thickness aluminium so mating a copper slug with a set of Ali fins gives you the best of moth worlds.

    surface area is a key factor in getting rid of heat as well, If you sand blast the surface of an aluminium sheet with a course grit you can dissipate more heat than a polished surface. I suspect this is why high quality heat sinks are often anodized as the anodizing process creates somewhat of a rough surface as well as makes the surface a little more stable.

    Years ago I did some tests on some solar hot water elements and what kind of metal was best to use as the reflector, sand blasted ALI foil outperformed all other materials tested. Initially I thought that polished ALU would be better but it was no where near as responsive as the sand blasted version. Even black ALU did not perform as well.

    1. > sand blasted ALI foil outperformed all other materials tested. Initially I thought that polished ALU would be better but it was no where near as responsive as the sand blasted version. Even black ALU did not perform as well.

      Some years back, I had communication with a NASA scientist who said the best they had at the time for emitting heat was:
      – sand/media blasted Al (for increased surface area), with
      – a thin WHITE heat-conducting paint (so thin it doesn’t fill the textured surface) so the textured surface reflects away heat emitted by its surrounding textured surfaces.
      The net between emitted and re-absorbed was said to be much better with a white than a black surface.
      So I could see that working well for conductive flow applications too.

      If the material is thick enough, machine it to have valleys for increased surface area, then media-blast the resulting surfaces?

      DIY, I use coarse sandpaper with white ceramic brake caliper paint sprayed into a mist above the surface, descending like a super thin overspray. Any formed larger overspray droplets may fill some of the sanded texture, but at least introduce their own texture/surface-area too.

      Come to think of it, that descended mist may be a solution to partially texturizing a polished surface you can’t abrade.

  7. Make a dinglebopper that emits 2 or 3 pulses for every one received; most (all?) fan controllers don’t care very much about the regularity of the pulses, only the total pulses per sampling interval.

  8. Didn’t see it written or maybe it’s just industry standard, but will the server just refuse to turn on if it’s not receiving a fan signal? Is it a bios-level fail-safe kind of thing that can’t be disabled?

      1. This happened to me when I disconnected fan for cooling hard disks in a Lenovo workstation. HDD was hardly used so cooling it wasn’t necessary, and I wanted to reduce noise, but BIOS refused to boot.

  9. Then there’s the old “undervolt” trick…

    Instead of putting the dan between 12vdc and ground for 12 volts, put it between the 12 and 5 lines to run it at 7VDC.
    It makes a world of difference

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s