(Re)Engineering active cooling solutions for servers - A fan mod guide with a case study of the Brocade ICX7250-24P

verysneaky · Nov 19, 2025

1. Introduction

I've recently been through the process of trying to modify the fan configurations of a number of my servers from the stock OEM configuration to something else to reduce the amount of noise they produce. These servers were often designed without noise output as a meaningful consideration, and introducing that as a constraint has some trade-offs. My initial approach was to buy as many of the largest, highest CFM fans that I could, work out a way to mount them, and see if system temperatures stabilised with these new fans installed.

This approach left a lot of questions for me - what were the trade-offs that I'd made in this bargain? Was I still able to use these servers under full load without damaging the hardware? Had I selected the right hardware? Did I get enough fans to cover the thermal load? Were they placed geometrically in the right positions? Would these fans be satisfactory when I increase the number of hard drives in the server? In some cases, these fans weren't cheap either, and so exploring multiple solutions was becoming prohibitive.

So I decided to take a more engineering-based approach to better understand the design and solution space, so that I could make active decisions around some of these trade-offs. To be specific, the trade-offs that exist in this space when selecting a new fan are:

Temperature load on the system (choosing a lower noise fan will place more heat load on the system)
Fan noise (higher noise fans typically relieve heat load more effectively, but may not pass the wife test)
Ability to load the system to its full performance (in the case of a switch, choosing to keep the heat load small and fan noise low may mean you can't use some/any ports for PoE, depending on where your chosen fan sits in the solution space)

This resource will present the approach I took to doing this for the Brocade ICX7250-24P, a 1U network switch. In this particular case, a number of attempts to modify the fans in these switches have been made by others on this forum, with mixed results. A number of people replaced the two stock fans in this switch with lower airflow, lower noise fans, and ended up realising the thermal throttling problems. I wrote a quick summary of some of the fans that I'd researched and provided an initial assessment of performance difference and noise difference, however knowing that one configuration has x% of the performance didn't really give me much confidence that making one decision over another would actually solve the noise problem whilst providing effective cooling.

This certainly isn't the only way to tackle this problem, but I thought others might get value from the process that I went through. If you enjoy over-engineering your homelab; this guide is for you!

Disclaimer: I have a background in physics and engineering but I'm not a HVAC engineer - I learnt a lot of this in the process of my exploration of the problem. There's every possibility I'm misinterpreting parts of the theory and how it should be used appropriately.

2. Concepts of air cooling performance analysis

There are two elements to air cooling performance as it relates to fan selection:

The properties of the system, specifically how densely populated it is and the resulting system airflow impedance value (a pressure value, usually measured in Pa, in H2O or mm H2O)
The properties of the fan, specifically the pressure-airflow (P-Q) curve - how much airflow (usually measured in CFM or m^3/min) can the fan induce at different static pressure levels

The system airflow impedance (often referred to as just "system impedance", or "static pressure") is the amount of resistance the system exerts on the flow of a fluid (in this case air). For all fans, which are trying to move air, they will be more or less successful at inducing an airflow depending on how effectively they deal with this resistance.

In a free stream of air, there is no system impedance. This is where the Max Airflow measurements are taken, and is often what's reported in performance summaries. That means that the maximum airflow metric may be a relevant to you if you're intending your cooling solution to operate in a largely sparse/unpopulated environment (think single CPU 4U/5U or ATX form factor systems, though it always depends on the specific system). If the system is densely populated on the other hand (think 1U and 2U servers, particularly if they have a lot of hot-swap HDD bays), this metric is likely less valuable. What you're interested in is the airflow (CFM) that can be achieved at the system impedance, and so inspection of a P-Q chart is imperative.

System impedance is made up of two elements:

The static pressure of the system
The dynamic pressure resulting from the flow of air

This means that as air flows through a cavity, the system impedance increases proportionally to the flow rate of the air. This also means that as more objects impact the flow of the air (like in the case of a tightly packed 1U server), the system impedance increases.

On a fan specification summary data sheet, the "maximum static pressure" is also often cited, which is the system impedance (static pressure) at which the fan is unable to induce a flow of air (CFM=0). It is not possible for a fan to achieve its maximum airflow (CFM), and it's maximum static pressure at the same time. The airflow that a fan is capable of producing is determined by the static pressure of the system, and so performance of a fan is definitionally dependent on the system it is being used in and the role it is playing within that system (is it a case fan, a CPU fan attached to a heat-sink, a radiator fan?).

This is made a bit clearer by looking at the difference between placing two fans in parallel or in series (discussed later).

Note: I'm pretty inconsistent with units in this write-up. I have a strong preference for using SI wherever I can, but the resources and data around this are a bit of a mess. I've used SI wherever possible, except in the representation of performance where I've stuck to CFM and in H2O. Note that for the calculations to work you can't mix and match unit systems. For reliable results make sure you're only ever using SI when calculating quantities.

2.1 A note on determining system impedance

Having some kind of estimation of system impedance is critical to performing analysis this way. Without it, it's impossible to know where on the P-Q curve the system is and you have no basis for decision making. In order to get an estimate, the gold standard here would be to do this experimentally and take measurements of the system using a pitot tube and a manometer to properly characterise it. More generally though (in the absence of an ability to fully derive this experimentally), the system impedance follows the "square law", characterised by this equation:

P = KρQⁿ

where:

P : Static pressure (in H2O)
K : The load factor
ρ : The density of the fluid (lb/ft^3)
Q : The flow rate (CFM)
n : A constant, where n=2 approximates a turbulent system. n=1 characterises laminar flow, which is not the case here.

Given a system characteristic coordinate (P, Q) (from someone else's effort to characterise a similar system experimentally), we can solve this equation for K and then build up a system impedance curve. To illustrate:

K = P / ρQⁿ

Once we have K, we can substitute it back into the original equation for a sweep of Q values and plot the resulting P values. The utility of this relies on obtaining data from elsewhere, but this will be shown on graphs later using the limited information I was able to find.

2.2 How does placing fans in series or parallel affect cooling performance?

The general principle is discussed in the following sections.

2.2.1 Two fans in parallel

When two fans are placed in parallel, this primarily improves the airflow of the system. However, when compared to a single fan, it also sees a small improvement to static pressure performance, as demonstrated below:

As you can see, the system impedance curve intersects the parallel fan performance curve at both a higher static pressure, and it also produces a higher airflow. However, the intent of this configuration is to increase airflow in low static pressure systems, and the effect of this in the example above would have been more effective if the system impedance curve was lower.

2.2.2 Two fans in series

In contrast, when two fans are placed in series, this improves the ability for a fan system to operate with high system impedance. Like in parallel, we see an increase to both static pressure performance and airflow performance, however biased towards static pressure:

As can be seen, the system impedance curve intersects the fan system in series at both a higher static pressure and a higher airflow.

Both the concepts of parallel and series arrangements will be explored in the solution space later on to support selection.

3. Characterising the system

Before making changes to the system, it's good to understand how it currently performs, at least as the basis for evaluating alternatives, but also to better understand the designed-for performance regime.

The Brocade ICX7250-24P contains two Foxconn PIA040H12P 40x40x28mm exhaust fans arranged in parallel, each of which has a maximum air flow rating of 26.33 CFM and a maximum static pressure rating of 2.04 in H₂O.

In terms of the system impedance, I was unable to find any information specific to the Brocade ICX7250-24P, or any other Brocade product for that matter. This is considered a major limitation of this analysis. In fact, it was shockingly difficult to find any values for 1U servers more generally with any basis in reality. However, I did manage to find some for 1U and 2U servers that were produced by IBM and published in "Analysis of Airflow Distribution Across a Front-to-Rear Server Rack" (Radmehr et al., 2007):

Server	Airflow (CFM)	Static Pressure (in H₂O)
1U	43	0.15
2U	155	0.54

There are a few limitations associated with using these values. A non-exhaustive list includes:

They apply to high end IBM servers, not Brocade switches
They apply to servers, and not switches. Switches often lack the depth of servers, and have quite a different distribution of internal componentry (i.e., switches don't contain hard drives or large heat sinks).

As such we should take this curve with a giant chunk of salt. Nevertheless, this will be useful to give us some indication of the design and solution spaces. My intuition is that given that these values apply to servers, they would be a conservative estimate for a switch, but I really don't know. To provide additional context, I've made up a factor of safety curve for sanity checking any insights that come out of this analysis. If anybody has additional information they can provide here to support this kind of analysis it would be appreciated. For completeness, the factor of safety curve uses (Q=20 CFM, P=0.15 in H₂O) as the design point.

This current system performance is summarised in the following P-Q chart (fan characteristics digitised from data sheets):

Note that the single fan performance is only shown here for informational purposes. The total system fan performance is represented by the orange line. As can be seen, at the system fan curve intersects with the system impedance curve at (Q=49.3 CFM, P=0.2 in H₂O). Since these fan performance curves indicate airflow potential when operated at 12V (100% duty cycle), this means that the system designers anticipated that 49.3 CFM would be sufficient to keep the system within acceptable thermal thresholds when operating it at maximum load.

At idle, in Speed 1 (not the full speed shown above delivering 49.3 CFM), the system has the following measurements:

Code:

SSH@icx7250-24p-router>show chassis                                                                                                               
The stack unit 1 chassis info:                                                                                                                  

Power supply 1 (AC - PoE) present, status ok
Power supply 2 not present

Fan 1 ok, speed (auto): [[1]]<->2
Fan 2 ok, speed (auto): [[1]]<->2

Fan controlled temperature:
Rule 1/2 (MGMT THERMAL PLANE): 60.0 deg-C
Rule 2/2 (AIR OUTLET NEAR PSU): 33.5 deg-C

Fan speed switching temperature thresholds:
Rule 1/2 (MGMT THERMAL PLANE):
Speed 1: NM<-----> 93 deg-C
Speed 2: 82<----->105 deg-C (shutdown)
Rule 2/2 (AIR OUTLET NEAR PSU):
Speed 1: NM<-----> 58 deg-C
Speed 2: 49<----->105 deg-C (shutdown)

Fan 1 Air Flow Direction: Front to Back
Fan 2 Air Flow Direction: Front to Back
Slot 1 Current Temperature: 60.7 deg-C (Sensor 1), 33.0 deg-C (Sensor 2)
Slot 2 Current Temperature: NA
Warning level.......: 100.0 deg-C
Shutdown level......: 105.0 deg-C

Demonstrating that the idle temperature is at approx 60°C. Observing the fan duty cycle in this configuration:

Code:

SSH@icx7250-24p-router#dm fan
Fan 1 Speed at 6279 RPM.
Fan 2 Speed at 6398 RPM.

shows that these are spinning at ~6300 RPM. We know from the spec sheet that 100% duty cycle for these fans is 18000 RPM, so Speed 1 corresponds to ~35% duty cycle.

4. Understanding the thermal requirements

Before we get stuck into analysing different fans, it's useful to have a meaningful lens through which to view that performance. Namely, how much airflow do I actually need to cool my system? Armed with that information, we'll be able to identify the alternatives that are "good enough". Fortunately, the amount of heat produced by electronic components matches closely with the power consumed by the system, so we can derive the airflow requirements using the system power consumption.

4.1 Power Load

First, let's understand the requirement. We can look at the power consumption figures provided in the technical specifications for the Brocade ICX7250 series of switches, which provide the following consumption values for a few performance regimes (I assume 200VAC input here):

Configuration	ICX7250-24P (W)	ICX7250-48P (W)
Idle	48	70.8
Typical	428	862
Maximum	439	880

Note that the "Typical" configuration assumes that all ports have a PoE load and 10% traffic. The difference between this and the "Maximum" configuration is that the Maximum configuration has 100% traffic. In this case, PoE is obviously the biggest consumer of power.

Note: I've only included the PoE variants of the switches here because they are the most interesting. If you have a non-PoE variant, you should be able to follow the same process but your requirements will be far less constrained, so your performance will be impacted less by your fan selection.

4.2 Required airflow calculation methodology

The sensible heat equation gives us a model for understanding the trade-offs between power consumption, designed temperature rise and required airflow:

hₛ = cₚρQΔT

where:

hₛ : sensible heat (W). This is often also represented as BTU/hr in imperial.
cₚ : specific heat of air (assumed constant standard atmosphere 1006 J/kgC - this assumption could probably be improved as cₚ varies with temperature)
ρ : density of the air (1.202 kg/m^3 standard atmosphere)
Q : airflow (m^3/s - we'll use this to find the representation in CFM later)
ΔT : temperature rise from the inlet (°C)

This can be used a couple of ways:

1. to find out what airflow requirements we have to meet a temperature goal:

Q = hₛ / cₚρΔT

2. to understand what temperature impact using specific fans with specific airflow performance have

ΔT = hₛ / cₚρQ

4.3 Results

I've performed a sweep of parameters Q, hₛ and ΔT in the above equation to make the relationships a bit easier to visualise. The figure below shows how the airflow requirements vary for different heat loads, with different design temperature rises. I've also shown where the power load for the ICX7250-24P and ICX7250-48P lie on this chart. To read this, look at the CFM associated with the intersection point of these power loads and the desired temperature differential.

There are a few interesting insights to pull out of this:

As per above, the design airflow for the ICX7250-24P is 49.3 CFM. Looking at the intersection of the ICX7250-24P "Maximum" line with ~50 CFM, we can see that this corresponds to a design temperature rise of ~15°C. However, since the ICX7250-24P shares a lot of design commonality with the ICX7250-48P, it's possible that the design goals of these switches are not independent. For validation, I also generated the system characteristics for the IXC7250-48P (which includes a third Foxconn fan in parallel). This showed that the ICX7250-48P cooling system generates ~68 CFM, which intersects with the "ICX7250-48P Maximum" line between 20°C and 25°C - lets assume that this is approx 22.5°C. I suspect that this is actually the true thermal design range, and they just removed a fan in the ICX7250-24P since 2 fans still exceeded the designed-for thermal performance, but 1 fan would not have.
As little as 3.3 CFM can support the thermal load of the ICX7250-24P at idle for a ΔT=25°C target (note that this assumes no traffic)
As little as 5 CFM can support the thermal load of the ICX7250-48P at idle for a ΔT=25°C target (note that this assumes no traffic)
Interestingly, for the ICX7250-24P, small gains in CFM at the Maximum threshold seem to have a large impact on the cooling performance of the system. The span between ΔT=50°C (15 CFM) and ΔT=35°C (22 CFM) is only 7 CFM. Likewise, the span between ΔT=20°C (~38 CFM) and ΔT=35°C (~22 CFM) is only ~16 CFM. This indicates that at the lower end, small changes are likely to make a large difference. This also complicates things for us from a selection perspective, because our uncertainty in system performance will reduce our confidence that we can accurately identify whether the temperature differential is ΔT=35°C or ΔT=50°C.
Unfortunately, I'm missing information about the relationship between fan RPM and CFM for most of the fans under assessment, including the OEM Foxconn fan, so it's not perfectly clear what conditions led to my initial measurement of 60 degrees at idle. For the purposes of this, I'll assume:
1. Speed 1 corresponds to a 35% duty cycle
2. 35% duty cycle generates 35% the total airflow = ~18 CFM, which means 60°C is approx a 5°C temperature differential.
3. Therefore, a 35 degree rise corresponds to approx 55°C+ 35°C= 90°C
  1. Of course, this will depend on ambient temperatures in your locations. The 55°C here represents ambient temperature + heat transfer from adjacent server elements in the rack
Given observations (4) and (5), it seems to me that the desirable temperature range to opt for is between 20°C and 25°C, which appears to align with the Brocade design engineers perspective too. You could probably extend this to 30°C if you're running short on options, however this is pushing very close to the critical threshold of 93°C at which point the ICX7250-24P ramps up into Speed 2 (100%), which will not serve us well to reduce noise. There's a concern here as well in that if Speed 2 is not able to reduce the temperature back down sufficiently low, the switch will never revert back to Speed 1, so it is in your interest from a sound perspective to choose a fan with enough performance headroom to do this. There is also more uncertainty about performance in this selection region.

Therefore, for the ICX7250-24P, our desired minimum CFM to satisfy these design objectives is where ΔT ≤ 25°C intersects with the "ICX7250-24P Maximum" threshold. This appears to be at approx Q ≥ 30 CFM. Unfortunately for the ICX7250-48P users (not me!), maintaining this same temperature rise of ΔT= 25°C requires a doubling of the airflow to 60 CFM. Relaxing this to ΔT=30°C still requires 50 CFM.

To summarise:

Design ΔT (°C)	ICX7250-24P Airflow Requirement (CFM)	ICX7250-48P Airflow Requirement (CFM)
20	38	76
25	30	60
30	25	50

To state the obvious, the other dimension that hasn't been discussed here is operating the switch in a manner that doesn't take it to its design load. Managing the amount of energy consumed is another way to bring down the airflow requirements without sacrificing the design temperature. To plan your cooling solution around some fractional capacity, simply take the intended power consumption and identify the airflow requirements at the ΔT of your choice. In this case, an easy win for users of the ICX7250-48P would be to adopt a policy of using some fraction of the total PoE ports. To illustrate, if you chose to only use 24 of your 48 ports with PoE, your requirements would relax to something approximating the ICX7250-24P.

5. Fan performance
Below are a selection of fans for comparison. These fans are a mix of:

Fans that have been recommended as good candidates for substitution in Brocade ICX Series (cheap & powerful 10gbE/40gbE switching)
Fans that have been recommended by YouTubers and eBay vendors
Fans identified as within the desired performance regime (noise, airflow) from a search on Digikey

I don't provide any guarantees about these fans - I have not tested most of them. Independently of their performance, they may not be suitable for other reasons that I'm not aware of (tach features, low fan speed, etc).

Ref	Model	Availability	Max Airflow (CFM)	Max Pressure (inH2O)	Max Sound Level (dBA)	Estimated noise (2 fans, dBA)	Estimated noise (3 fans, dBA)	Relative noise intensity (3 fans) wrt. 2x Foxconn fans	Comment	Status
1	Foxconn PIA040H12P-F02-AB	Aliexpress	26.33	2.04	53.8	56.8	59	1.7x louder	Stock OEM selection	Tested OK
2	Sunon MF40201VX-1000U-G99	Digikey	10.8	0.29	27.5	30.5	32.3	282x quieter		Tested OK
3	Same Sky CFM-4020V-180-275-20	Digikey	10.8	0.29	27.5	30.5	32.3	282x quieter	Identical P-Q curve to Sunon MF40201VX-1000U-G99	Tested OK
4	Mechatronics MR4020M12B1-RSR	Digikey (bulk only) Bravo Electro	11.5	0.24	29.8	32.8	34.6	166x quieter		Untested
5	Comair Rotron Gryphon GDA4028-12BB	??	11.26	0.34	31.44	34.45	36.2	115x quieter	"Unobtainium", but here for comparison	Reported OK
6	Mechatronics G4020E12B1-RSR	Digikey (bulk only) Bravo Electro	11.7	0.23	32	35	36.8	100x quieter		Untested
7	Delta AFB0412HHB-F00	Digikey (bulk only)	11.5	0.359	32	35	36.8	100x quieter		Untested
8	Delta FFB0412HN-F00	Digikey	10.6	0.248	32.1	35.1	36.9	98x quieter		Reported OK
9	Delta EFB0412VHD-F00	Digikey	10.1	0.416	32.5	35.5	37.2	91x quieter		Reported OK
10	Same Sky CFM-4020CF-195-342-20	Digikey Mouser	11.9	0.41	34.2	37.2	39	60x quieter		Untested
11	Mechatronics MR4020H12B1-RSR	Digikey (bulk only) Bravo Electro	13.6	0.34	35.1	38.1	39.9	49x quieter		Untested
12	Delta AFFB0412VHB-F00	Digikey (bulk only)	13	0.459	36.5	39.5	41.3	35x quieter		Untested
13	Same Sky CFM-4015B-1105-365-20	Digikey	12.3	0.56	36.5	39.5	41.3	35x quieter		Untested
14	Sanyo Denki 109P0412H3013	Digikey	11.3	0.414	37	40	41.7	32x quieter		Untested
15	Mechatronics G4028M12B1-RSR	Digikey (bulk only)	13	0.37	37	40	41.8	32x quieter		Untested
16	Wakefield DC0402812J2B-3T0	Digikey	13.8	0.46	38	41	42.7	26x quieter		Reported fan failure in Speed 1
17	Mechatronics MH4028L12B1-RSR	??	12.8	0.33	39	42.01	43.8	20x quieter		Untested
18	Mechatronics MR4020E12B1-RSR	Digikey (bulk only)	15.8	0.45	39.1	42.1	43.9	19x quieter		Untested
19	Sanyo Denki 109P0412B3013	Digikey	13.4	0.574	40	43	44.7	16x quieter		Reported OK
20	Sanyo Denki 9L0412H301	Digikey	13.1	0.426	40	43	44.7	16x quieter		Untested
21	Mechatronics MR4020X12B1-RSR	Digikey	17.3	0.54	41	44	45.7	13x quieter		Reported OK
22	Mechatronics G4028H12B1-RSR	Digikey (bulk only)	16	0.57	41	44	45.7	13x quieter		Untested
23	Sanyo Denki 109P0412G3013	Digikey	14.8	0.719	42	45	46.7	10x quieter		Reported OK

In order to get a better view of how each of the candidate fans performed relative to each other, I digitised all of the performance charts that the vendors provide (shout out to Engauge Digitizer), and overlayed this with the system impedance curves described earlier to help guide our assessment of the performance regime of these fans that we should care about.

Note: Like-for-like performance should be compared at the points at which each fan intersects the "Typical 1U system impedance" curve, which follows a square law. You should not just compare values at the same static pressure; these are not comparable due to the dynamic pressure component of system impedance produced by the airflow itself.

We shouldn't read too much into this figure; these represent the single fan performance, and what we really care about is the total system performance. Having said that, it's exceedingly obvious just how much better the OEM Foxconn fans are than any of the other alternatives being evaluated, both from a pressure and airflow perspective. It's also surprising how the performance regime demonstrated by the "typical 1U system impedance" curve is so close to the maximal airflow performance of the fans. If this assessment of the system impedance is realistic, then the first order analysis described in the introduction is probably sufficient; comparing the max CFM values (and ignoring static pressure as a metric) is probably "close enough". Additionally, looking at the chart, the distance between the OEM fan and the next most performant stays roughly equidistant for the entirety of the profile.

As mentioned earlier, the default configuration for the ICX7250-24P is two exhaust fans arranged in parallel at the rear of the chassis. This is the baseline from which performance will be compared. In the case of the ICX7250-24P, there is an unused mounting point for a third fan, and usage of this should form part of the solution space. The following configurations will be assessed in the following section:

Removing one of the existing OEM Foxconn fans and operating the switch with a single fan to reduce noise
Applying a low noise adapter to reduce the duty cycle of the existing OEM Foxconn fans to reduce noise
Replacing the two existing OEM Foxconn fans with an alternate and keeping them in a parallel configuration
Replacing the existing two and adding a third fan in parallel
Replacing the existing two and adding an additional two fans in series (4 fans total)
Replacing the existing two, adding a third, and adding an additional three fans in series (6 fans total)

6. Configuration performance
6.1 Single Foxconn fan
As can be seen in the single fan performance chart above, the usage of a single Foxconn fan would generate ≈26 CFM at the system impedance. This is a viable alternative, generating a ΔT ≈ 30°C at the maximum load of the ICX7250-24P, if a little hotter than I would prefer. This is also effective at reducing the sound output by (2 times quieter) at maximum load. However, another consideration is the noise when not loaded up to the maximum load. At Speed 1, these fans operate at 35% duty cycle. This generates 9 CFM from one fan, which is enough to support 128W of heat at ΔT = 25°C or 156W at ΔT = 30°C. I don't have any data to quantify the noise level at this duty cycle (I haven't been able to find noise/RPM curves for any of the fans assessed), but in my personal anecdotal experience it is too loud, and offers limited improvement over just keeping the stock configuration.

The following are the measurements I took after letting the switch reach a steady state with only a single fan:

Code:

SSH@icx7250-24p-router>show chassis
The stack unit 1 chassis info:

Power supply 1 (AC - PoE) present, status ok
Power supply 2 not present

Fan 1 ok, speed (auto): [[1]]<->2
Fan 2 failed

Fan controlled temperature:
Rule 1/2 (MGMT THERMAL PLANE): 74.5 deg-C
Rule 2/2 (AIR OUTLET NEAR PSU): 39.5 deg-C

Fan speed switching temperature thresholds:
Rule 1/2 (MGMT THERMAL PLANE):
Speed 1: NM<-----> 93 deg-C
Speed 2: 82<----->105 deg-C (shutdown)
Rule 2/2 (AIR OUTLET NEAR PSU):
Speed 1: NM<-----> 58 deg-C
Speed 2: 49<----->105 deg-C (shutdown)

Fan 1 Air Flow Direction: Front to Back
Fan 2 Air Flow Direction: Front to Back
Slot 1 Current Temperature: 74.5 deg-C (Sensor 1), 39.5 deg-C (Sensor 2)
Slot 2 Current Temperature: NA
Warning level.......: 100.0 deg-C
Shutdown level......: 105.0 deg-C

Code:

SSH@icx7250-24p-router#dm fan
Fan 1 Speed at 6282 RPM.
Fan 2 Speed at 182 RPM.

So, ≈75°C at ~6300 RPM.

6.2 Two fans in parallel

As a reminder, the way to read the performance on the following chart is to identify the airflow generated at the intersection point of each fans curve and the relevant system impedance curve. I have also marked where the predicted temperature rise thresholds exist for the full load configuration of the ICX7250-24P. These thresholds will be different for the ICX7250-48P. In order to identify which fans meet the ΔT=25°C threshold, look at the fan performance curves that intersect the system impedance curve to the right of this threshold.

The legend in these charts shows each fan model in a unique colour. I appreciate these charts are quite busy and can be difficult to read, so I've used distinctipy to make these colours as distinguishable from each other as possible (if not pretty, make it functional). I've also shown the predicted noise for the entire solution in the legend entry (i.e., noise for 2 fans in this case, more in the subsequent configurations).

There are two configurations to talk about in this section:

6.2.1 Using a low noise adapter to reduce the noise of the stock fans

As you can see in the above chart, two duty cycles were evaluated for the Foxconn fans: 35% and 50%. In terms of performance, it looks as though operating 2 Foxconn fans at 50% duty cycle would be sufficient to ensure ΔT < 25°C when the switch is under full load. However, this would be louder than the current Speed 1 configuration (35%), and so doesn't represent the kind of sound gains that I want to see.

6.2.2 Replacing the stock fans with another alternative

As it turns out, for just a direct replacement of the existing fans there is only one alternate fan that meets our requirement of ΔT < 25°C and that's the Mechatronics MR4020X12B1-RSR. At a total noise for 2 fans of 44 dBA, this represents a noise intensity 19x quieter than the stock fans. This solution is not without risk however, given the uncertainty associated with the system impedance curve. Looking at where it intersects the safety factor system impedance curve shows a ΔT ≈ 33°C, which is a bit warmer than I would like. If you're not feeling particularly rigid about the ΔT=25°C requirement and are willing to relax it slightly, the Mechatronics G4028H12B1-RSR has a slightly more forgiving P-Q curve. This fan has the same noise performance (44.0 dBA), but you can see that it intersects the safety factor system impedance curve at 24 CFM, which is much closer to the ΔT=30°C threshold.

Still, this safety factor performance for both fans isn't a show stopper - that's going to be a similar story for most of the fans in this assessment. I think either of these are a good solutions, but leaves some room for further noise optimisation as 44 dBA still isn't particularly quiet. Of the two, in this configuration, I would probably go for the Mechatronics G4028H12B1-RSR and sacrifice slight performance on the "typical" system impedance curve under the assumption that reality is slightly worse than this.

On reflection, a number of fan mod solutions proposed in these forums for the ICX7250-24P are represented in this chart, and it makes sense that people saw temperature issues. The Mechatronics 4020E12B1-RSR, Sunon MF40201VX-1000U-G99, and Delta EFB0412VHD-F00 generate between 13.7 and 15 CFM and are unable to meet the ΔT=50°C threshold. The Mechatronics MR4020M12B1-RSR, Comair Grphon GDA4028-12BB, Delta AFB0412HHB-F00 and the Delta FFB0412HN-F00 generate between 15.3 and 15.8 CFM in this configuration, and place them in the ΔT=45°C and ΔT=50°C temperature rise thresholds at best. These are amongst the quietest solutions of the alternatives, but none of these are acceptable solutions in this configuration without making other accommodations such as reducing the power load.

6.3 Three fans in parallel

With the addition of a third fan in parallel, a number of additional options are opened up to us as the airflow generation capability of the system is improved, as demonstrated below.

The following fans now meet our temperature threshold requirement of ΔT=25°C. These include the Same Sky CFM-4015B-1105-365-20 (41.3 dBA), Same Sky CFM-4020CF-195-342-20 (39.0 dBA), Delta AFFB0412VHB-F00 (41.3 dBA), Mechatronics MR4020H12B1-RSR (39.9 dBA), Mechatronics MH4028L12B1-RSR (43.8 dBA), Sanyo 109P0412B3013 (44.8 dBA), Mechatronics G4028M12B1-RSR (41.8 dBA), Sanyo 9L0412H301 (44.8 dBA), Mechatronics MR4020E12B1-RSR (43.9 dBA), Sanyo 109P0412G3013 (46.8 dBA), Wakefield DC0402812J2B-3T0 (42.8 dBA), Mechatronics MR4020X12B1-RSR (45.8 dBA), and the Mechantroics G4028H12B1-RSR (45.8 dBA).

The quietest of these are the Same Sky CFM-4020CF-195-342-20 (39.0 dBA) and the Mechatronics MR4020H12B1-RSR (39.9 dBA). Of the two, the Same Sky CFM-4020CF-195-342-20 intersects the safety factor impedance curve at 20 CFM, which corresponds to ΔT ≈ 37.5°C, whilst the Mechatronics MR4020H12B1-RSR intersects the safety factor system impedance curve at 21.2 CFM which corresponds to ΔT ≈ 36°C. I think that this temperature difference is probably marginal, and so in this case my preference would be the additional 0.9 dBA and the Same Sky fan. Having said that, if you are a bit less risk tolerant than I, then there a few more options further up the curve. For example, the Wakefield DC0402812J2B-3T0 (42.8 dBA) provides an impressive 38 CFM in the 3 fan configuration, which edges it across the ΔT=20°C threshold. Against the safety factor system impedance curve, it intersects at 26.9 CFM, corresponding to ΔT ≈ 28°C. This fan probably represents the least risky balance between performance and noise, however it is 3.8 dBA louder than the Same Sky alternative. There are of course quieter alternatives between these two, however the performance difference in the case of the safety factor system impedance curve is marginal; the Wakefield DC0402812J2B-3T0 is really the first that offers a substantial enough difference.

In comparison to the original Foxconn fans, three Same Sky CFM-4020CF-195-342-20 are 60x quieter, three Mechatronics MR4020H12B1-RSR are 49x quieter, and three Wakefield DC0402812J2B-3T0 are 26x quieter.

For those willing to run things a little hotter, or reduce the load on the switch for still-quieter alternatives, the Mechatronics MR4020M12B1-RSR (34.6 dBA) will run at 27 CFM for ΔT ≈ 27.5°C on the typical curve, and 18 CFM for ΔT ≈ 42.5°C on the safety factor curve. Quieter still is the Sunon MF40201VX-1000U-G99 (32.3 dBA) which will provide 26 CFM for ΔT ≈ 28°C on the typical curve, and 16.7 CFM for ΔT ≈ 45°C on the safety factor curve. You would really only want to consider these in a three fan configuration if you live in a location with a low ambient temperature, other active cooling solutions, or you intend to operate the switch at partial capacity.

6.4 Two fans in parallel, two fans in series

By adding two additional fans (four fans in total) in series with the original two, we see an improvement to static pressure, without an increase in maximum airflow. This doesn't do a great deal in the way of expanding the list of available options to us that meet the ΔT=25°C threshold, however what it does do is make all of the options more resilient; they have an increased ability to handle static pressure and are thus now able to maintain airflow to a much more similar capacity in systems that have more impedance.

In comparison to the 2 fans in parallel configuration, we now have both the Mechatronics MR4020X12B1-RSR (47.0 dBA) and the new addition of the Mechatronics G4028H12B1-RSR (47.0 dBA) that satisfy the ΔT=25°C threshold. The new addition is due to the increase in the slope of the curve due to the new static pressure resiliency of the fan system, which has increased the effective airflow the system is able to achieve under the system impedance. Additionally, due to the increased slope of the curve, the general trend here is that whilst there is a slight improvement in the airflow capability at the predicted typical 1U system impedance, the performance at the safety factor system impedance curve is substantially improved. This means that some fans that had good performance at the typical 1U system impedance curve but had high ΔT values at the safety factor impedance curve, now have much more reasonable performance at the safety factor impedance curve. To illustrate, the Mechatronics G4028H12B1-RSR produced an airflow of ≈24 CFM at the safety factor system impedance curve in the two fan configuration, corresponding to ΔT ≈ 31°C. In the 2+2 configuration, this same fan now generates ≈28 CFM, corresponding to ΔT ≈ 27.5°C, which is a shift across thresholds. Likewise, the Mechatronics MR4020H12B1-RSR (41.1 dBA) has transitioned from ΔT ≈ 42.5°C to ΔT ≈ 35°C. This improved resilience comes at the cost of additional noise, however.

Given that this configuration doesn't really change the calculus for fans at the typical 1U system impedance curve, my conclusion from this data is that going straight to this configuration would be cost inefficient. My suggestion would be to go for a parallel configuration first, and then if the realised system impedance deviates substantially from the modelling here, add additional fans first in parallel, and then in series. This will ensure that both cost and noise are kept to a minimum, and avoids the addition of unnecessary fans. The key takeaway from this though, is that in systems with high airflow impedance it is more valuable to add fans in series rather than in parallel (though both are valuable).

6.5 Three fans in parallel, three fans in series

Now we look at a 3+3 configuration (6 fans total). This broadly just exaggerates the effect we saw in the 2+2 configuration, however we can now see that we are much, much closer to the stock performance of 2x Foxconn fans.

This configuration provides a substantial boost to both airflow and static pressure performance. In this configuration, the Sunon MF40201VX-1000U-G99 (35.3 dBA) provides the quietest solution, representing a total solution 141x quieter than the original. This is followed by the Mechatronics MR4020M12B1-RSR (37.6 dBA), which represents a total solution 83x quieter than the original. In this configuration, the Sunon has slightly improved performance over the 3 parallel configuration, achieving a ΔT ≈ 25°C against the typical 1U system impedance curve. It performs much more reliably against the safety factor system impedance curve as well, at ΔT ≈ 32°C (improvement from ΔT ≈ 45°C). Note that this appears to be an identical fan to the Same Sky CFM-4020V-180-275-20 (35.3 dBA), which has an identical P-Q curve. It's worth noting also that there's still some cause for concern with performance against the safety factor system impedance. ΔT ≈ 32°C is quite close to the Speed 2 threshold, indicating that this cooling solution may not be sufficient to cool the solution sufficiently to return it to Speed 1. The Mechatronics MR4020M12B1-RSR performs slightly better than the Sunon/Same Sky, however it also adds 2.3 dBA to the noise; almost double the noise, and only has marginally better performance in the case where the safety factor system impedance is relevant.

Notable mentions here include the Com Air Gryphon GDA4028-12BB and the Delta FFB0412HN-F00; these improve the cooling performance tangibly for a typical 1U system, however fail to provide substantial additional tolerance to a system with higher static pressure as demonstrated by the safety factor system impedance curve, operating at ΔT ≈ 30°C. This may not be an issue, but selection of these depends on your risk tolerance and how accurately you think this analysis characterises the system impedance.

In order to develop confidence in the face of the uncertainty around system impedance, it seems as though the Same Sky CFM-4020CF-195-342-20 (42.0 dBA) offers the best solution, achieving ΔT ≈ 22.5°C in the typical 1U system impedance case, and ΔT ≈ 26°C in the safety factor impedance case. I wouldn't expect this to be silent at 42.0 dBA, however it does offer a substantial noise improvement over the Foxconn configuration (30x quieter).

7. Discussion and conclusion

To conclude, my broad thoughts of each of the possible solutions is:

Removing one of the existing OEM Foxconn fans and operating the switch with a single fan to reduce noise
- Whilst seemingly effective, performance wise, it doesn't appreciably reduce the noise, so not a viable solution
Applying a low noise adaptor to reduce the duty cycle of the existing OEM Foxconn fans to reduce noise
- As with (1), the fans are too noisy even at low RPMs for this to be considered an effective treatment
Replacing the two existing OEM Foxconn fans with an alternate and keeping them in a parallel configuration
- One solution stood out in this space; the Mechatronics MR4020X12B1-RSR which satisfied the design temperature threshold, however did so at 44 dB. There are quieter solutions in the remaining options, so this shouldn't be considered.
Replacing the existing two and adding a third fan in parallel
- A reasonably large number of fans stood out in this space. The Same Sky CFM-4020CF-195-342-20 (39 dBA) and the Mechatronics MR4020H12B1-RSR (39.9 dBA) performed similarly and yielded the quietest results, however suffer from poor performance in environments with higher static pressure, leading to a risk of overheating if the "Typical 1U system impedance curve" has been too poorly estimated.
Replacing the existing two and adding an additional two fans in series (4 fans total)
- Only two fans met our design requirements in this configuration; the Mechatronics MR4020X12B1-RSR (47.0 dBA) and the Mechatronics G4028H12B1-RSR (47.0 dBA), however I don't consider this an actionable solution given how non-competitive 47 dBA is in comparison to other solutions.
Replacing the existing two, adding a third, and adding an additional three fans in series (6 fans total)
- Surprisingly, this solution turned out to be the one that yielded the most resilient and quietest solution. The candidate solutions resulting from this are:
  - 6x Sunon MF40201VX-1000U-G99 (total 35.3 dBA) or 6x Same Sky CFM-4020V-180-275-20 (total 35.3 dBA) for the quietest solution
  - 6x Mechatronics MR4020M12B1-RSR (37.6 dBA)
  - 6x Same Sky CFM-4020CF-195-342-20 (42.0 dBA), which offers the best performance guarantees with the cost of poorer noise performance

Now, this brings us to what this means in terms of fan selection. For the purposes of this I'm assuming a desire to be able to run the switch with a full PoE load. For the ICX7250-24P, we're aiming for >30 CFM total to stay within the ΔT=25°C to ΔT=30°C range. This seems to be reasonably achievable by adding a third fan in parallel. Of course, this makes some large assumptions about the system airflow impedance, and so whether this is 3 fans in parallel, or 3 fans in parallel and 3 in series (6 total fans) to make sure the airflow requirements are satiable is another question. If anybody has any more realistic information about system airflow impedance (static pressure), that would be greatly appreciated. Nevertheless, there's also a risk profile question. Given the uncertainty with the system impedance, are you more comfortable buying noisier (but more performant) fans to mitigate the risk that the cooling performance would be insufficient in a quieter solution? This is everyone's trade-off to make.

For me, I need my lab to pass the wife test as it sits in my home office next to mine and my wife's desk. With that in mind, I'm open to a riskier solution where the fans may not meet the temperature targets outlined above for a fully loaded system. My mitigation to this risk is that I'm unlikely to load the switch with full PoE any time soon, and so I won't realise the error of my ways for quite a while. Therefore, my selection will be 3x Sunon MF40201VX-1000U-G99 initially. This will provide approx 26 CFM of airflow for a design temperature of ΔT ≈ 28°C. If the system impedance characterisation is too optimistic, I'm anticipating that the realised temperature rise could be as much as ΔT ≈ 45°C, which would be too hot to operate the switch at full capacity. In the case of this eventuality, I will either add more fans in serial to improve this performance (to an estimated ΔT ≈ 32°C at 24 CFM), replace the Sunon's with the Wakefield DC0402812J2B-3T0 (for an estimated ΔT ≈ 27°C), or add an additional fan to the ASIC as has been described in other successful fan mods.

For the ICX7250-48P, we're aiming for >60 CFM to stay within the ΔT=20°C - 25°C range. None of the fans I evaluated fall within this range. The closest was the Mechatronics G4028H12B2-RSR which with 3 fans in parallel generates 41.5 CFM at the typical 1U system impedance design point. If you were to use these fans, you would expect to see a temperature rise in the order of ΔT=35°C, which would likely push the switch into critical territory. Placing another 3 of these fans in series increases the total airflow to ~46 CFM, which improves things, but not enough to change things drastically. This still has a ΔT ≈ 32.5°C rise. Three of the Foxconn OEM fans in parallel only generate 68 CFM at the 1U typical impedance, so I'm hard pressed to imagine a quieter solution is available that satisfies the performance threshold, without adding more fans, larger fans (like the 120mm mods I've seen), or fans in different locations (like on top of the ASIC).

Having said that, as you can see there is a solution space to play around with if you don't need the maximum thermal capacity of the switch. If you're only using half the PoE load, then I suspect the comments about the ICX7250-24P will mostly apply.

As a final note, I hope that this has also made it obvious where there might be utility in going for very low noise, low CFM fans like the Noctua NF-A4x20. This type of fan has very low noise (14.9 dBA) and produces a maximum airflow of approx 5.5 CFM. As per the temperature thresholds chart above, ~5 CFM can cool up to 100W of heat generation with a temperature rise of ΔT=35°C. I would consider this the extreme outer range of performance, and would not trust this to an individual fan. In the interest of maintaining a temperature rise of ΔT=25°C, I would limit my usage of a single NF-A4x20 fan to applications less than 60W. This likely excludes many PoE switches. These fans also do not have a large static pressure rating, and Noctua haven't published any P-Q charts for them so it's difficult to know what the relationship looks like, but I would assume practically you will get less than 5 CFM of performance out of them in most applications. It's also worth noting that there are other reasons not to select these fans for the ICX7250-24P - there are minimum RPM thresholds that need to be satisfied so that the switch doesn't think that the fan has died; I've seen some reports that this has occurred with these fans.

Additionally, it's worth pointing out that another major deficiency in this analysis is the effective noise. All of this analysis has made assumptions based on the maximum sound pressure. Practically though, these fans are likely to be operating at some fractional speed, ideally at their most efficient around ~50%. Each of them will have unique noise profiles that may be better or worse, regardless of the maximum sound pressure value. They may have harmonics or whines that make them less tolerable, even if they are quieter. Unfortunately, the vendors didn't seem to publish much on this to draw insight from so in the absence of any personal experience this hasn't been treated as a decision point.

8. Selection and Recommendations

NOTE: Not all of the fans in these results have been tested. Use these at your own risk.
NOTE: You should come back to these recommendations after reading the Initial Results post, which highlights some issues with arranging fans in series.

For the extremely noise conscious and with a mild to moderate PoE load
- 3x Sunon MF40201VX-1000U-G99 (32.3 dBA, estimated ΔT ≈ 28°C)
- Add additional 3 fans in serial if cooling performance insufficient for your needs (35.3 dBA, estimated ΔT ≈ 32°C)
- Replace fans with an alternative or add additional fan to ASIC if still insufficient
For the moderately noise conscious, but likely to have a reasonable (but not full) PoE load
- Solution 1 (preferred)
  - 3x Same Sky CFM-4020CF-195-342-20 (39.0 dBA)
  - Add additional 3 fans if cooling performance insufficient for your needs (42.0 dBA)
- Solution 2 (if 1 not available)
  - 3x Mechatronics MR4020H12B1-RSR (39.9 dBA)
  - Add additional 3 fans if cooling performance insufficient for your needs (42.0 dBA)
For the moderately noise conscious, but likely to have a full PoE load with a lot of devices
- 3x Wakefield DC0402812J2B-3T0 (42.8 dBA)
- You shouldn't need to add any more fans to this; this should really be enough in most cases. However, if you add another 3 in series, you should be at ~90% of stock fan performance (45.8 dBA)
- Note that there have been reports that the Wakefield falls below the minimum RPM threshold in Speed 1
For those who are not noise conscious at all and only want performance
- Add 3 more of the Foxconn fans in series

9. Future Work
~~I haven't implemented this design decision yet for the ICX7250-24P - I will edit this post with the results when they're available.~~
Initial results showing the outcomes of my purchases are available in the Initial Results post

Resources

Edit 15/05/2026: Update fan table to include confirmed and unconfirmed functionality in the switch
Edit 26/05/2026: Update selection and recommendations to (a) include a disclaimer about the recommendations being theoretical and unvalidated, and (b) resolve typo erroneously recommending the Sunon MF40101VX-1000U-G99 (10mm depth and not assessed) instead of the Sunon MF40201VX-1000U-G99 (20mm depth, assessed)
Edit 08/06/2026: Add links to initial results post

fohdeesha · Nov 28, 2025

holy shit, and I thought *my* mom took too much tylenol during her pregnancy. nice work

Mithril · Nov 28, 2025

Holy crap what a post! I'm going to have to re-read this when I have more caffeine, and IQ

Mithril · Dec 1, 2025

So, looking over this a bit more I'm not sure if the math on the fans when in series accounts for the many issues with putting fans in series (and why so many series configurations have the fans rotating counter to eachother, among other airflow design elements). At minimum you'd need significant space between series fans (external or internal).

IMHO without moving to 3d printed (or otherwise fabricated)options to use larger 60mm, 80mm, or 120mm fans the "go from 2 to 3" seems like the most realistic option.

If anyone has fancy fluid flow modeling software I wonder if we could mitigate some of the performance loss with airflow guides inside of the switch.

One thing I will note from my experience with many 120mm fans is that two fans can both be "40dBA" even as measured (lots of fan measurements by oems... bend the truth) but one will be more noticeable due to the actual sound profile, and this will change based on how restrictive the airflow path is. I have put 120mm fans on a water cooling radiator that sounded fine as case fans and then they sound terrible.

verysneaky · Dec 1, 2025

Yeah that's a good point to call out @Mithril. The air is generally turbulent coming out of the first fan in a series, which if stacked with no separation from the second fan will lead to more noise, and probably a loss in efficiency in the subsequent pressure drop. The two treatments to this are (a) as you say, have the second fan in series counter rotating which helps to minimise the turbulent air, or (b) separate the fans with a duct. I've done a little research into this and it seems like the size of the duct doesn't need to be too large - (5mm - 10mm?) to see a lot of the negative effects minimised/mitigated. From a sound perspective, I thought that optimum's video on stealing Noctua's fan mod was interesting, as it was essentially adding an exhaust duct and some space between the trailing fan edge and the exhaust grill.

Nevertheless, on the ICX7250-24P, adding space between the fans in series is only possible if you're willing to take destructive steps, like removing the fan clips, otherwise there's only 2-3mm of additional flexibility in the spacing (which still might be sufficient to have a tangible impact). I'm still waiting on appropriate screws to mount my new fans, so I haven't quantified any of this yet. Still, I think the modelling of the effect of placing fans in series described above is probably instructive enough from a "rough order of magnitude" perspective. My initial insight from my experimentation is that it does have a meaningful impact on cooling performance without any additional treatment (ducting, counter-rotating fans) for the turbulent air, at least for the ICX7250-24P. I'll leave the detail supporting that statement to when I've collected a few more data points though.

Mithril · Dec 5, 2025

Another thought, are there any blower/radial/cage fans with the right height we could adapt. Many radial fans are fantastic at static pressure compared to an axial fan of similar noise level. However, most of the ones I see have an intake on the wide/flat side and output to the narrow side; which would mean mounting internally and running into fitment issues or seeing if reverse airflow works better.

One of those times I wish i was good with 3D printing and modeling, I suspect adapting some level of "anti vortex" fins in a 3D printed spacer might be a huge benefit with stacked fans.

I also wonder if we are willing to mod the case to add a 3rd rear fan when it normally has 2 (the 24 port version), is there a good place to add an intake fan or two. I assume you'd need to keep it in-balanced enough to still get airflow over the SFP+ ports and POE board at the front (or very careful fan placement).

Unspec · May 15, 2026

This entire "study" entirely fails to account for the most important thing: actually working in the real world. I purchased 3 Wakefield DC0402812J2B-3T0 off the recommendation here and the switch will fail to boot, as the fan does not reach the minimum rpm and so the switch reboots after seeing no "working" fans. So i would take the fan recommendations with a massive grain of salt until and unless someone proves that the switch boots with them in the real world

40 bucks down the drain for nothing lol

verysneaky · May 15, 2026

Unspec said:
This entire "study" entirely fails to account for the most important thing: actually working in the real world. I purchased 3 Wakefield DC0402812J2B-3T0 off the recommendation here and the switch will fail to boot, as the fan does not reach the minimum rpm and so the switch reboots after seeing no "working" fans. So i would take the fan recommendations with a massive grain of salt until and unless someone proves that the switch boots with them in the real world

40 bucks down the drain for nothing lol

I think I caveated the analysis above pretty heavily - this was all done pre-purchase for me. As I stated:

I haven't implemented this design decision yet for the ICX7250-24P - I will edit this post with the results when they're available.

And also, with respect to the fan thresholds:

It's also worth noting that there are other reasons not to select these fans for the ICX7250-24P - there are minimum RPM thresholds that need to be satisfied so that the switch doesn't think that the fan has died; I've seen some reports that this has occurred with these fans.

A few questions for you:
1. Are you doing this mod on the ICX7250-24P?
2. If yes to (1), did you switch the fan pins to meet the motherboard requirements of the switch? (it expects ground and +12V to be swapped)
3. If no to (2), refer to the following videos which describe what you need to do: [1][2]

I've not tested the Wakefield fans, or most fans in the list above, but I am planning on providing a real world validation follow up to what I've presented. It's taken some time to instrument my set up to get the data I wanted, and it's far from perfect, but I'm hoping to write something up soon. Some additional comments:

1. I ended up buying the Sunon MF40201VX-1000U-G99 and the Same Sky CFM-4020V-180-275-20, both of which are less performant than the Wakefield's you bought.
2. The datasheet for the Wakefield DC0402812J2B-3T0 indicates it runs at 8000 RPM at 100% power. Speed 2 is 35%, so I would expect this to run at ~2800 RPM. This is about what my Sunon and Same Sky fans run at without issues, so I would be surprised if the RPM is too low for the switch

Unspec · May 15, 2026

verysneaky said:
I think I caveated the analysis above pretty heavily - this was all done pre-purchase for me. As I stated:

And also, with respect to the fan thresholds:

A few questions for you:
1. Are you doing this mod on the ICX7250-24P?
2. If yes to (1), did you switch the fan pins to meet the motherboard requirements of the switch? (it expects ground and +12V to be swapped)
3. If no to (2), refer to the following videos which describe what you need to do: [1][2]

I've not tested the Wakefield fans, or most fans in the list above, but I am planning on providing a real world validation follow up to what I've presented. It's taken some time to instrument my set up to get the data I wanted, and it's far from perfect, but I'm hoping to write something up soon. Some additional comments:

1. I ended up buying the Sunon MF40201VX-1000U-G99 and the Same Sky CFM-4020V-180-275-20, both of which are less performant than the Wakefield's you bought.
2. The datasheet for the Wakefield DC0402812J2B-3T0 indicates it runs at 8000 RPM at 100% PWM. Speed 2 is 35%, so I would expect this to run at ~2800 RPM. This is about what my Sunon and Same Sky fans run at without issues, so I would be surprised if the RPM is too low for the switch

You caveated the Noctua's, not other server grade fans.

Yep, 7250-24P. Also, for what it's worth, I do not have Foxconn fans, and instead have Nidec W40S12BGB5-52 T02.

Yep. You mean it expects ground and tach to be swapped compared to normal computer fans.

Who knows why it doesn't work. It spins up fine on boot, but once it's fully booted, the fans are completely still (tested with 1 original fan plugged in as well - still completely still). Probably some 0db feature on the wakefields. Regardless, they simply do not work and should be removed from the guide.

verysneaky · May 15, 2026

Unspec said:
Yep. You mean it expects ground and tach to be swapped compared to normal computer fans.

Who knows why it doesn't work. It spins up fine on boot, but once it's fully booted, the fans are completely still (tested with 1 original fan plugged in as well - still completely still). Probably some 0db feature on the wakefields. Regardless, they simply do not work and should be removed from the guide.

Yes, sorry - ground and tach. You're right. For what it's worth though, what you're describing is the behaviour I saw when I had ground and tach wired the wrong way. This is a photo of how I've got mine wired for reference:

For future readers, that's tach/power/ground into the board.
Edit: Note, I've updated the fan table to show whether fans have been tested or reported to either work or not - hopefully that adds some clarity for future readers.

Unspec · May 15, 2026

verysneaky said:
Yes, sorry - ground and tach. You're right. For what it's worth though, what you're describing is the behaviour I saw when I had ground and tach wired the wrong way. This is a photo of how I've got mine wired for reference:
View attachment 48784

For future readers, that's tach/power/ground into the board.
Edit: Note, I've updated the fan table to show whether fans have been tested or reported to either work or not - hopefully that adds some clarity for future readers.

I have it wired the exact same way. Appreciate the table.

I know this wasn't a tested fan, but the Delta FFB0412SHN-BF00 has been reported as working.

Nidec W40S12BGB5-52 T02 should also be added at the top of the table, as it is one of the other OEM fan types these seem to ship with. That said, specs are hard to find for them.

TRS-80 · May 25, 2026

I have been reading STH for a while now. I finally decided to register an account today so I could share my results in this awesome thread.

I really wanted to thank the OP (@verysneaky) for their work here, as well as @fohdeesha (and others) for all their original research and sharing of information on these awesome switches.

Unfortunately, my results were not as good... The fans were indeed dead silent (after initial boot, once they go down from 100% to 50% or whatever). But they did not exhaust enough heat, eventually creeping up to a temperature that caused the fans to kick into high speed (and then stay there, even after the temps dropped down below what I understood to be the lower threshold temperature(?!)).

So, now double checking part numbers, it seems I ordered the wrong ones? Although, in fairness it currently says:

verysneaky said:
8. Selection and Recommendations

For the extremely noise conscious and with a mild to moderate PoE load

3x Sunon MF40101VX-1000U-G99 (32.3 dBA, estimated ΔT ≈ 28°C)

Above is the exact part number I ordered (and installed):

MF40101VX-1000U-G99 (3x)

However, in the table (and linked to DigiKey), it says...

MF40201VX-1000U-G99 (3x)

Now it seems to me that the only difference is the thickness here (10mm for MF4010, and 20mm for MF4020). But that would explain the performance difference.

I could have sworn that I had ordered the right ones, I seem to recall double (maybe even triple?) checking. Maybe @verysneaky updated this in the meantime?

Well, I guess I will try and order the correct ones, and try again... Arrrgh!

Hopefully we can get this corrected, before someone else makes the same mistake?

I did also have to reverse engineer and re-connect the wires:

switch -> new fan
black ---> black (-)
yellow -> red (+)
green --> yellow (tach)

Finally, the OEM fans seem to have 4mm screws, but the replacements use 5mm. Although the latter seem to be the commonly available ones for PCs.

I tried but could not seem to upload any photos.

verysneaky · May 26, 2026

I could have sworn that I had ordered the right ones, I seem to recall double (maybe even triple?) checking. Maybe @verysneaky updated this in the meantime?

Well, I guess I will try and order the correct ones, and try again... Arrrgh!

Oh dear. You have my apologies! This was a typo as a result of manually typing the model number for the written conclusion. I didn't assess the 10mm fan, so it never formed part of the analysis (the charts all have the MF40201VX-1000U-G99 listed). I've rectified the typo now.

I would suggest that the original post shouldn't inform any purchases until I follow this post up with results; what's been presented so far is a purely theoretical analysis, and my own findings show some big gaps in what's in the original post. I'll actually update the recommendations with this warning.

TRS-80 · May 26, 2026

verysneaky said:
I would suggest that the original post shouldn't inform any purchases until I follow this post up with results; what's been presented so far is a purely theoretical analysis, and my own findings show some big gaps in what's in the original post.

I had ordered in between the time of your OP and when you published your first results. My intention was to help carry out some of the testing (and help you defray some of the cost of buying all these different fans!). I figured that would be the least I could do, after you put so much effort in to get this party started.

I'm not the slightest bit upset (unlike some people

), rather the contrary, I really appreciate your efforts thus far!

New 20mm fans should be here in about a week, and I hope to be able to confirm your results.

ksjh · May 28, 2026

Thank you @verysneaky and @fohdeesha for all the excellent information. I replaced all three fans in my ICX7250-48P with Sunon MF40201VX-1000U-G99 fans. I also added a 3 mm spacer similar to the ones from optimum's YouTube video I 3D printed using Sunlu Silk TPU. The switch is almost silent now, but I have not put significant load upon it so far. I do not plan to utilize a high PoE load anyway, I will plug in a few 10G LR SFP+ modules in near future.

Since the weather is quite warm here today, the switch just changed to speed setting two without any significant network or PoE load:

Code:

Fan 1 ok, speed (auto): 1<->[[2]]
Fan 2 ok, speed (auto): 1<->[[2]]
Fan 3 ok, speed (auto): 1<->[[2]]

Fan controlled temperature:
        Rule 1/2 (MGMT THERMAL PLANE): 75.0 deg-C
        Rule 2/2 (AIR OUTLET NEAR PSU): 41.5 deg-C

Fan speed switching temperature thresholds:
        Rule 1/2 (MGMT THERMAL PLANE):
                Speed 1: NM<-----> 95       deg-C
                Speed 2:        85<----->105 deg-C (shutdown)
        Rule 2/2 (AIR OUTLET NEAR PSU)***active***:
                Speed 1: NM<-----> 41       deg-C
                Speed 2:        34<----->105 deg-C (shutdown)

But the switch is still quite low noise with this fan speed.

verysneaky · May 28, 2026

ksjh said:
also added a 3 mm spacer similar to the ones from optimum's YouTube video I 3D printed using Sunlu Silk TPU.

Have you noticed a difference relative to no spacer? This is something I'm planning to experiment with as well

ksjh · May 28, 2026

Since both the fan speed and the noise are so low, I did not immediately notice any difference, but I will also experiment a bit more. I just ordered some of those Chicago book screws and those nuts (M3x20x4) to try two fans in series (and thin spacers in between them).

verysneaky · Jun 7, 2026

Initial Results

Based on my previous analysis, I decided to buy the following fans:

3x 40x40x20 Sunon MF40201VX-1000U-G99
3x 40x40x20 Same Sky CFM-4020V-180-275-20
1x 60x60x10 Sunon MF60101V1-1000U-G99

My thought process here was that:

The Sunon and Same Sky fans theoretically have identical performance curves
This would allow me to validate the differences between these two fans, given their suspiciously identical performance
This would allow me to test all fan configurations that I proposed above:
- Single
- Double
- 3 in parallel
- 2 in parallel, 2 in series
- 3 in parallel, 3 in series
In addition, it would allow me to assess the relative impact of attaching a fan directly to the ASIC, as recommended in other posts about fan mods. This one is a little more complicated to quantify analytically, so it was excluded from my original analysis.

The first thing that became apparent was that whilst the ICX 7250-24P has 3 slots for fans, the shroud blocks the exhaust of the third fan completely. This was a pretty large oversight on my behalf (penance for looking at a deconstructed switch for too long, I guess), and so the first observation is that the theoretical benefit of having three fans in parallel is entirely unrealisable without a dremel and the motivation to destructively modify the case. This is unlikely to be relevant to the ICX 7250-48P, which I believe does have an additional exhaust port in the case.

Nevertheless, I left this configuration in the test suite to see if it did impact performance at all, even without an exhaust port for the third fan.

Temperature

For each of my tested configurations, I recorded the following temperatures:

This was a difficult experiment to control for, as there were non-trivial fluctuations in the ambient temperature both during individual tests and also between different configurations. I recorded the ambient temperature at the beginning and end of each test though, and so I've attempted to correct for this by using a linear gradient across the test sample and subtracting it to yield the system temperature, i.e.

ΔTᵢ = Tₘ,ᵢ-Tₐ,ᵢ

As you can see, temperatures are underwhelming, and in general terms don't align particularly well with the original analysis. All configurations performed very similarly, except when an ASIC fan was added, which improved ASIC cooling performance substantially to approximate parity with the Foxconn fan. This indicates that my original analysis underestimated the system impedance curve. I'm not particularly surprised by this given the vacuum of information around characterising it, but it does change the outcomes materially.

To elaborate on this, these are the temperatures recorded at idle. There is no load either from a PoE or traffic perspective. Even so, we can see that a number of configurations are dangerously close to the Speed 2 threshold for the ASIC (93°C) in this case, which was not predicted for ~50W load. However, the difference between a ΔT=50°C temperature rise and a ΔT=20°C temperature rise at 50W is only ~2.5 CFM, so there is a lot of room for error in predicting this accurately. Even in this case, you can see that the only configuration that actually does ramp to Speed 2 is the single fan configuration.

My intention for this results post was originally to attempt to actually characterise the system impedance curve and provided an updated performance assessment based on my experimental measurements. This has not been as easy as I had hoped; as it turns out experimental design is hugely important to making sure that the airflow and pressure measurements are performed in a way that (a) doesn't introduce additional losses on the system, and (b) doesn't introduce an overestimation bias in the measurements. Essentially, it turns out that the act of measuring the system without influencing the system is a difficult one. I will follow this post up with some of my preliminary attempts at this and some of my findings on that, if for no reason other than hopefully it provides the next person interested in doing this sort of analysis with information about where the pitfalls are.

Noise

One of the big things that I failed to consider in the initial analysis was the difference between Speed 1 and Speed 2. Of course, if Speed 2 is quiet enough then I suppose it doesn't matter, however my observation is that even with the quietest fans in the list that I assessed, that's not the case. The case of making this office friendly then means that ideally these will be in Speed 1 for the majority of the time, where the initial analysis focused on the maximum noise impact in Speed 2 (100%).

I measured the noise of of my environment for each of the different fan configurations in my experimental design, which was within an operational rack. This is a reasonably silent rack; the switch was by far the dominant noise source in all of these experiments, however it is not a clean room test. The noise measurements for each of the configurations is shown below:

Similarly to temperature, it was difficult to maintain a controlled environment for this, so I'm not sure this provides a lot of meaningful information for comparison. The ambient noise varied between tests (and probably within tests), but I'm showing it here to illustrate that even though the fans I'm using are rated for a max sound level of 27 dBA, these spec-sheet numbers don't necessarily translate directly to your environment.

For instance, the estimated noise for 2x Sunon's was 30 dBA in Speed 2, but what was measured was ~55 dBA. Similarly, even though the OEM Foxconn fans are rated for 53.8 dBA each, the overall impact at Speed 2 was ~85 dBA. Neither of these is particularly desirable. However, because there was quite a large swing in the ambient noise for each of these experiments, I've attempted to adjust for that with the following formula:

Nₛ = Nₜ - Nₐ

This is performed in the logarithmic space, which does not properly characterise the effective power of the noise source required to achieve the measured sound pressure, but it does give us some indication of the difference in perception. The results are shown in the second subplot above. I think this has reasonably corrected for some of the ambient noise differences, which are clear in the first plot, and you can now see that most of the alternatives are in the same performance regime, which matches what I perceived.

However, I made these noise measurements using the "Sound Meter" Android app, averaging the samples across a 30 second period, and I think this chart reflects some of the limitations with this approach. The first is that the granularity of the measurement was 1 dBA, which is not particularly fine and makes comparison difficult. The second is that it seems like the accuracy of the measurements are within ≤3 dBA, which is probably not accurate enough to draw meaningful comparative conclusions from.

I have the following observations:

Pretty much all combinations with the exception of the Foxconn are only slightly perceptible above the ambient noise when in Speed 1 (qualitatively).
In Speed 1, whilst there may not be a substantial increase in the noise level, all combinations have unique acoustic properties that make them distinguishable from the noise floor (whine, hum, etc)
Some of the measurements for otherwise similar combinations turned out to be wildly different. I didn't perceive a large difference between these when I was performing the measurements, but to illustrate:
- 3x Sunon has ΔN = 3 dBA when compared to 3x Sunon + 1x Sunon 60mm in an exhaust configuration at Speed 2. My intuition was that the addition of another fan should make this configuration more noisy, not less
- 3x Sunon + 1x Sunon 60mm in an intake configuration has ΔN = 1dBA when compared to 3x Sunon in Speed 2. This makes sense, however we can see that this relationship is reversed at Speed 1, which is confusing for similar reasons to (1). Why would four fans be quieter than three?
- 3x Sunon + 1x Sunon 60mm in an intake configuration is uniformly louder than the Same Sky in the same configuration. This also doesn't make a lot of sense, and wasn't my perception at the time
- A single Same Sky fan is approximately 1dBA louder than 2x Same Sky fans in Speed 1, and 3dBA louder than 2x Same Sky fans in Speed 2. This also doesn't make a lot of sense.

All of that to say that I don't think that there's a lot that can be gleaned from this experiment except the broad observation that all of these configurations were substantially quieter than the Foxconn fans. I think my experimental procedure probably (a) wasn't consistent enough, and (b) I didn't use tooling of the appropriate granularity and accuracy to capture the differences quantitatively. I guess one of my anecdotal takeaways is that I didn't notice a huge difference between any of the alternate options (maybe because of the acoustic characteristics/whine which probably dominate over the noise).

As an aside, one interesting observation about the Foxconn fans is that whilst they have 28mm depth, the depth of the fan that they contain is 20mm - exactly the same as many other fans in this comparison:

The additional 8mm at the exhaust side is reserved for a static stator, the purpose of which is presumably reduction of the airflow turbulence and the associated noise. This effect is characterised in the following resources:

Which begs the question: Does adding a spacer or an axial stator between the outlet and the fan would improve airflow performance and reduce noise (or importantly, any vibration or turbulence "whine" effects as a result of interactions with the chassis)? This is a point for future exploration.

Fans in series

The results above show that there is very little benefit to putting fans in series. This is counter to the theory, or at least, it doesn't seem to be well publicised what conditions need to be satisfied to realise the theoretical relationship. My hypothesis (without digging into the theory a great deal) is that the turbulent air on the exhaust side of the fan is swirling substantially, and the ability of the following fan in series to effect an additional pressure drop is decreased by the turbulent and swirling nature of the flow (perhaps because the second fan in series is rotating in the same direction as the swirling air leaving the first). Some things that may help resolve this to some extent are:

Using a counter-rotating fan as the second in the series to oppose the flow state, though I suspect that this will have a negative impact on noise performance
Addition of a duct (and maybe a honeycomb insert) to separate the two fans in series, so that the flow has time/distance to stabilise back to something approaching laminar
Addition of an axial stator to reduce the rotational velocity in the airflow

Future work

As mentioned above, one of my longer term goals is to characterise the system impedance more effectively.
- Experimentally, using an anemometer and/or manometer
- Theoretically, using Computational Fluid Dynamics (CFD) - if there are any enterprising honours or masters students looking for an interesting project this one could be for you!
Explore the impact of adding a duct or axial fan stator between the fans and the exhaust wall
Explore the impact of adding a duct or axial fan stator between fans in series

Conclusion

None of the configurations that I proposed in my initial analysis really satisfy my requirements with these fans. This is largely as a result of erroneous assumptions around:

The system impedance curve for these switches - this seems to be much higher than initially estimated
The ability to use three fans in parallel on the ICX 7250-24P (so the maximum I can have in parallel is two)
The effect of placing fans in series - nil effect observed, but some additional work to see if ducts or stators make a difference

To mitigate these erroneous assumptions, my final configuration is:

2x Sunon MF40201VX-1000U-G99 as exhaust fans
1x Sunon MF60101V1-1000U-G99 mounted to the ASIC with zip ties as an intake fan

(though the image shows the Same Sky fans mounted - this is mostly to demonstrate how the ASIC fan is mounted)

This has the effect of cooling the temperature sensor directly, which provides a substantial impact on the cooling performance. There's still an open question in my mind about whether this resolves the problem entirely; I'm left to wonder whether the reduced airflow through the case is sufficient to cool other componentry that don't themselves have a temperature sensor. In particular, do the SFP+ ports receive the necessary cooling they need? What about the PoE distribution boards at the front? It's possible that the reduced airflow through the case generally manifests as a temperature rise in the PSU area (and maybe this is sufficient?), but this isn't something I've tested. I guess I'm just identifying that by compensating with additional cooling on the ASIC directly, this solves the ramping and reporting problem associated with the ASIC, but may still lead to additional heat damage to other components, especially under load. Additionally, this does nothing to avoid ramping in the PSU area, so if that's what's triggering speed ramping, then it's not clear to me that this will resolve that issue at all. It's worth noting that perhaps obviously, adding the ASIC fan reduces the temperature of the ASIC measurements substantially, but has very little impact on the PSU measurements; it's not solving part of the problem.

I certainly won't be running any 10GBaseT transceivers in the SFP+ ports.

TRS-80 · Jun 14, 2026

Hello again, everyone!

My results and analysis are not going to be nearly as thorough as the above, but I wanted to share them nonetheless.

I think that OP is trying to approach this in as proper/scientific of a way as possible, where I (and I suspect, many readers) are looking for more practical advice. Although the latter ultimately flows from the former, I suppose. Anyway...

verysneaky said:
The first thing that became apparent was that whilst the ICX 7250-24P has 3 slots for fans, the shroud blocks the exhaust of the third fan completely. This was a pretty large oversight on my behalf (penance for looking at a deconstructed switch for too long, I guess), and so the first observation is that the theoretical benefit of having three fans in parallel is entirely unrealisable without a dremel and the motivation to destructively modify the case. This is unlikely to be relevant to the ICX 7250-48P, which I believe does have an additional exhaust port in the case.

I can confirm that the -48P (which I have) has openings in the case for 3 fans.

The correct (20 mm) fans arrived, and I was able to get them installed yesterday, and did my first test last night.

My results were a bit of a mixed bag. Slightly disappointing (at first) but still better than the (noise levels of the) OEM jet engines at the end of the day. So... ¯\_(ツ)_/¯ I guess?

In my case, I have (3) SFP+ ports populated (with DACs), only a couple of the Gigabit Ethernet ports populated, none of which are under any sort of significant load, nor using any PoE (yet). So I consider this a pretty light load. And this is in a small bedroom office with a Midea U-Shaped (mini split in a window unit form factor) set at 70 F (actual ambient probably being some degrees above that, depending on where in the room).

As a reminder, I am testing (3) Sunon MF40201VX-1000U-G99 in an ICX 7250-48P.

So, given the above, the results I experienced...

Fans were essentially dead silent at low speed (or at least low enough I could not hear them at all over ambient noise in the room, especially the A/C).

Unfortunately, after about an hour[0], the temperatures in the case had crept up to the point where the fans kicked up into high speed. So I can only conclude that these (3) fans cannot keep up with the cooling demands, even at this low load.

In my case (both in this test, and my first one), I observed that the fan speeds never did go back down, either, even well after the temperatures had gone back down within the range where I would have expected this to have occurred. Not sure what's up with that, and I am curious if anyone else noticed the same?

So, as I said, bit of a mixed bag.

Therefore, going forward, the next things I am considering are to:

1. Maybe just lower the A/C settings a couple degrees, to try and keep the ambient temperature lower in the room overall. It seems I am hovering right around the temperature limits where the fans kick into high speed, so maybe this will be enough to keep them on low(?). This would be the easiest thing for me to do, and requiring the least time/effort, so I might try this first.

2.a. Add a large, slow (quiet) intake fan on top of the ASIC. In my case, I don't mind cutting a hole in the case, as I figure I am the terminal owner of this switch. And maybe more air (in general) helps keep other components cool, as well? As I also share OPs concerns about sufficient cooling of other areas of the board which are not measured.

2.b. Add some temperature probes around elsewhere in the case, due to concerns (mentioned above) that we are just moving the problem somewhere else. In my mind at least, I wouldn't really want to do (a) without doing (b).

2.c. This will require more time/effort, so might have to wait a while.

[0] That was the result of my first test, last night. I am now doing a second test. Even though it is afternoon here (and warm), I have been watching the internal temperatures and they are hovering just under the limits for kicking the fans into high speed. And the switch has been on for 1.5 hours now. So I think this is something which will require further experimentation and monitoring. BTW, this is with A/C set at 70 still.

TRS-80 · Jun 14, 2026

This is not super scientific, but if you feel around with your hand on top of the unit, after it has been on for a while, it's pretty clear where most of the heat is coming from (that ASIC or whatever it is with the big heat sink). Which seems to concur with the temperature measurements I am seeing, as well as conventional wisdom around these switches.

Maybe later I will get around to adding some more temperature sensors, but after thinking about it some more (and spending the rest of my Sunday reading all the posts from people who did this before), I went ahead last night and ordered a little 60x10mm fan to install on top of that ASIC.

(Re)Engineering active cooling solutions for servers - A fan mod guide with a case study of the Brocade ICX7250-24P

Member

Kaini Industries

Active Member

Active Member

Member

Active Member

New Member

Member

New Member

Member

New Member

New Member

Member

New Member

New Member

Member

New Member

Member

New Member

New Member