3-pin vs 4-pin [Variable Speed] Fans

[wraith]

Hi All,

I've recently successfully completed my first FreeNAS build. During the build I noted comments in the X10SL7-F manual on the preference for 4-pin fans to allow the board to control the speed. I've looked around the forums and haven't found much specifically on this topic [apologies if it is there and I missed it]. All fans in my build are currently 3-pin and am contemplating whether to replace with 4-pin.

I'm interested in feedback/opinions on whether replacing the 3-pin with 4-pin fans would make a significant difference to power usage, reliability, noise etc. Power usage would be my primary focus given the server will be running 24/7

Thanks in advance

 

[Ericloewe]

significant difference to power usage

Judging by what you likely have in terms of fans at the moment, no.

reliability

Maybe, but it's an academic discussion.

noise

Sure, it'll cut noise a bit. Might not be noticeable, depending on what you have now.

 

[m0nkey_]

So, the difference between 3-pin and 4-pin fan headers. Well, 3-pin headers are controlled using voltage. If your motherboard only had 3-pin headers, chances are they either control the fan speeds by adjusting the voltage, or not at all (not all 3-pin headers do this). The 3rd pin is for RPM monitoring. With 4-pin headers, you get into PWM (pulse width modulation) which pulses the fan to reach the desired speed. That doesn't mean your fan is going to pulse, it'll still run at a constant RPM.

If you have 4-pin fan headers, go with PWM. It allows for better control of the fan speeds. 3-pin headers usually aren't dynamic or run 100%.

 

[jgreco]

Also, don't mix 3- and 4-pin fans.

 

[Ericloewe]

Also, don't mix 3- and 4-pin fans.

Other than a general preference for 4-pin fans, I can't think of a reason why mixing them would be bad - obvious limitations aside.

 

[jgreco]

Because usually the BMC firmware isn't designed to mix and match.

Some of the more recent boards do not bother providing variable voltage (3-pin) support to fans anymore. Supermicro has been in the process of not supporting variable voltage fans for awhile now and I don't know if any current products even do; I took a look around and it seems like X10 and X9 do not, while a manual for X7 implies it does:

http://www.supermicro.com/manuals/motherboard/3210/MNL-0970.pdf

See page 2-19, "2. Please use all 3-pin fans or all 4-pin fans on the motherboard. Please do not use 3-pin fans and 4-pin fans on the same board."

The general gist of it seems to be that the design of a combination 3-pin and 4-pin system is that there's a single voltage controller for most of the fans (Fan1-4), and during BIOS initialization, it attempts to determine if any of the fans are 4-pin, by setting the voltage to full, pulsing the fans at various PWM, and seeing how the tachs read. If a fan appears to be 4-pin, then the system will lock the voltage at full, and manage fans by PWM. If the fans do not appear to be 4-pin, then PWM is ignored and voltage is varied appropriately. The situation of mixed 3- and 4-pin fans is undefined. Will the system decide to run all the fans at 12V? Nonoptimal. Will the system try to voltage control the fans? Also nonoptimal plus potentially undefined behaviour from the PWM fans, which might not even run if they're not seeing PWM. That could be real bad actually. ;-)

 

[Ericloewe]

X9 and newer, at least, don't even do voltage control. It's PWM only, so all fans get +12V directly from the PSU rail.

 

[jgreco]

X9 and newer, at least, don't even do voltage control. It's PWM only, so all fans get +12V directly from the PSU rail.

Actually I misspoke, it appears that some of the X9 do.

http://www.supermicro.com/manuals/motherboard/C202_C204/MNL-1311.pdf

page 2-29 and one other example that I had already dismissed. I actually spent a little while dredging around because I knew that when I built our Xeon hypervisors in 2011, there were some issues related to that, and we had gone to some effort to acquire all 4-pin gear and to build some special 4-pin wiring harnesses. Your eyes go cross-eyed after you read enough manuals.

Anyways, the end results are:

1) Do not mix 3-wire and 4-wire fans unless you've got good reason to know it is safe to,

and

2) Try to acquire 4-wire fans everywhere because 4-wire rocks.

 

[Ericloewe]

Actually I misspoke, it appears that some of the X9 do.

http://www.supermicro.com/manuals/motherboard/C202_C204/MNL-1311.pdf

page 2-29 and one other example that I had already dismissed. I actually spent a little while dredging around because I knew that when I built our Xeon hypervisors in 2011, there were some issues related to that, and we had gone to some effort to acquire all 4-pin gear and to build some special 4-pin wiring harnesses. Your eyes go cross-eyed after you read enough manuals.

Anyways, the end results are:

1) Do not mix 3-wire and 4-wire fans unless you've got good reason to know it is safe to,

and

2) Try to acquire 4-wire fans everywhere because 4-wire rocks.

Hmm, that's odd, since the product page says "Support 3-pin fans (w/o speed control)".
http://www.supermicro.com/products/motherboard/Xeon/C202_C204/X9SCM-iiF.cfm

Maybe the Nuvoton BMC craps out if it some fans don't respond to PWM?

 

[jgreco]

I don't know. I recall it was a bit of a mess. I also seem to recall that the 1U SKU (can't think of it) seemed to be different, and my guess at the time was that Supermicro had done different things because the boards were essentially intended to go into chassis solutions that they already had available. It's one of those edge cases where the knowledge isn't worth the effort needed to ascertain what was actually going on; rather than spending an hour or two of tech time in the shop, it's cheaper just to source the four wire fans and be Done With It.

 

[wraith]

Hey guys,

Appreciate your input; I've learnt something new already [not mixing 3-pin and 4-pin amongst others].

I took a punt, yesterday, and purchased two 4-pin fans [Noctua NF-A14] and installed them as exhausts at the top of the case. The remaining two [supplied with the case] are 3-pin and performing as intake. My local only had two fans with more coming in the next week. I'll replace the supplied fans with 4-pin fans next weekend.

With the four fans running, it is making more noise than what I'd like. I wouldn't say it was excessive though; This being my first build, I haven't any comparison, though. Maybe, as recommended by jgreco, replacing the two remaining 3-pin with 4-pin will eliminate any confusion and allow all fans to drop in speed.

I'll report back :)

Would it be prudent to track temperatures for a week [obtain a base line], turn a fan off and monitor for awhile to determine whether four fans are necessary?

 

[jgreco]

More fans than less are always prudent; your fans are protecting a lot of $$$$ in hard drives, so you want your system not to turn into an EZ-Bake Oven if a fan dies on that weekend where you've taken a mini-vacation, turned down the air conditioning, and don't discover the problem for several days.

It would certainly be reasonable to unplug the two 3-pin fans and carefully monitor temps. This would give you some idea on how close you are on cooling. It would also let you find out whether your system is more likely to be quiet with 4-pin fans.

 

[wraith]

I was able to purchase and install two more Noctua NF-A14 PWM fans today [replacing the fans that were shipped with the case]. The fans are certainly quieter and I was able to further reduce noise by installing the Low Noise Adaptors that came with the fans. Future plans installing an additional fan in the front once I load up the bays with nine more HDD's.

My current setup for air cooling is attached; would welcome any feedback or suggestions. One potential improvement I can already see is to move the front fan to the top location.

The server with the new Noctua fans has been operating for about twenty minutes now. The fans seem to be regularly speeding up and slowing down. At their slower speed, they are inaudible [awesome!]; at their higher speed they do make some noise. Will this eventually settle down [hopefully into the quieter mode unless required]?

Are there any best practices on location of server within a home? I'm contemplating placing it in the hallway cupboard near the existing QNAP TS-410.

 

[Ericloewe]

The server with the new Noctua fans has been operating for about twenty minutes now. The fans seem to be regularly speeding up and slowing down. At their slower speed, they are inaudible [awesome!]; at their higher speed they do make some noise. Will this eventually settle down [hopefully into the quieter mode unless required]?

You have to change the IPMI sensor thresholds. Check the X10/X11 FAQ (link in my sig) for details.

 

[wraith]

@Ericloewe : Thanks for that. The fans appear to have settled into a steady state, unfortunately at the higher speed [some noise]. None-the-less, I follwed your FAQ and have modified the FAN1, FAN3 and FAN4 lower sensor thresholds to LNR=40, LCR=140 & LNC=240. I'll see how this affects the fans' performance over time.

 

[Robert Riedel]

I had 3-pin fans hooked up to the supermicro x10sl7-f and they where louder then I wanted. Then I swapped them for Noctua PWM fans and adjusted them in ipmi and they are silent most of the time unless the machine needs more air. My fractual design fans where pretty quiet on a 3-pin connectors on a different board but really loud on this one.

 

[wraith]

@Robert Riedel : Thanks for the tip. I went into the motherboard IPMI and changed the fan setting to a lower speed setting. Once the fans had slowed down, I turned optimal back on. The fans have not, yet, increased their speed. Fingers crossed the correct interpretation of this is that the logic went a bit screwey initially and set the fans on a higher speed when it didn't need to.

I, currently, have a nice, quiet server :)

Temperatures have increased slightly [no more than 5%] but nothing I'm to concerned about at the moment :)

 

[Ericloewe]

I'll add "properly document IPMI fan control behavior" to my todo list for February.