devLap

How Close Is Too Close? Applying Fluid Dynamics Research Methods to PC Cooling

hackernews

In April 2025, LinusTechTips visited NASA's Langley Research Center in Hampton, Virginia at the invitation of the Flow Physics and Control Branch, where they were hosted by Dr. Louis Edelman. Louis served as a researcher at NASA Langley from 2018 to 2026 but has left NASA since we filmed the LTT video. He is now an assistant professor in the University of Tokyo Department of Aeronautics and Astronautics.

The intent of this article is to accompany the LinusTechTips video (and Floatplane exclusive videos) on the testing conducted. Not all of the content could be included and it would be a shame for it not to be shared. Louis generously offered to write a article paper delving into the test setups, methodologies, and conclusions. We hope you find it interesting and educational.

NASA Langley Research Center was founded in 1917 as the Langley Memorial Aeronautics Laboratory, predating NASA itself. It was the first government research laboratory for aeronautics in the United States and has been involved in nearly every advancement in the science of flight for the last 109 years. There are lifetimes worth of stories to tell about the rich history of Langley and the future its personnel are working to build today. For the LTT visit, the challenge before us was to distill some part of that immense legacy into a single video. We decided that the most impactful and honestly fun approach would be to perform an experiment that bridged the world of leading-edge flight research to the (hopefully leading-edge) PC sitting on your desk. An experiment that highlighted the tools, technology, and process behind the everyday in NASA’s “first A”, aeronautics. We also squeezed in some tours; many of those moments can be seen as Floatplane Extras.

The most obvious aerospace-adjacent component in a gaming PC is the humble case fan. So we set out to answer a deceptively simple question: does a restrictive front panel or placing your PC too close to the wall hurt the cooling potential of your fans? How close is too close? What if my intake fan is also my radiator cooling fan? In this article, we detail how we carried out the experiments in the LTT video, the history of the techniques and facilities we used, and some of the behind-the-scenes process.

Before we proceed any further, a disclaimer for the legally minded among you. The opinions expressed herein are those of the author and do not reflect the opinions of NASA or the Government of the United States of America. The detailed description and use of any hardware and software in this article is purely descriptive of their use and does not imply an endorsement of those products. With that out of the way, on to the science!

Experimental Methods

The core of the three experiments is a single Noctua NF-A12X25 120 mm diameter PC fan powered at 12V from a benchtop DC power supply and given a 100% PWM control signal through a Noctua NF-FC1 fan controller. A 3D printed bracket rigidly mounts the fan perpendicular to the table, lifting it up and away from any ground effects without disrupting the inlet or exhaust flow paths. A 3 mm thick acrylic sheet is laser cut to the outer dimension of the fan frame with a 105 mm square 4x6-32 clearance hole pattern for mounting. Four 75 mm long 6-32 screws are inserted into the fan frame and the acrylic plate is then slid onto them in front of, or upstream, of the fan face. This keeps the experimental space to a single parameter: the gap between the front plate and the fan face, as noted in Figure 1. 3D printed snap-on spacers are placed on the screws between the fan frame and the acrylic plate; these ensure the plate is nominally parallel to the fan face. The fan is set at 100% 12V PWM to reduce the complexity of the variable space. This is representative of typical PC case or PSU intake fan operating conditions during heavy load with the fan working at its maximum airflow and cooling potential.

1 of 2

Figure 1 Diagram of the mounted fan test model with adjustable acrylic front plate and a photo of it on the optical table.

With just one day of testing and several testing methodologies to demonstrate, keeping the parameter space one-dimensional was an absolute necessity. These considerations mirror the challenges of a production wind tunnel. There are always more questions to ask and parameters to test than you have tunnel time - or funding - to answer. Thus, you must minimize the number of variables while extracting the maximum amount of knowledge. This is called the design of experiments. It is its own field of study and one every lab environment must master. 

Adjusting the fan to plate gap, we applied three testing techniques in this experiment: aerodynamic tufting, particle image velocimetry (PIV), and aeroacoustic measurements. We will discuss in detail some of the history of each technique as we discuss the methodologies and thought process behind implementing them.

Component Summary:

Strings in the Wind: Aerodynamic Tufting

In fluid dynamic research, it is usually best to start simple. As with a large-scale wind tunnel or flight test, the first measurement technique we applied in our benchtop fan testing was “tufting”; the practice of affixing short strings on an aerodynamic surface to visualize flow separation. The first documented use of this technique was by Melvill Jones at the University of Cambridge in 1929 who glued “tufts” of wool to the wing of test aircraft to identify the onset of flow separation and stall. NASA evolved this technique to be non-intrusive and easier to image with fluorescent micro-tufting in the 1980’s. Because of its simplicity and detailed qualitative results, fluorescent micro-tufting remains a central methodology at NASA today, further enhanced by modern LED UV lamps and high-speed camera technology. 

We used a custom 3D printed tufting jig that allowed us to weave fluorescent, glow-in-the-dark thread into equal length tufts and apply thin Kapton tape. It is important to ensure that the tufts are long enough to be sensitive to the flow but not too long or too tightly packed that they collide and stick together. For this application, three 50 mm tufts were mounted along the motor mount arm on the exhaust side of the fan. The objective of this stage is to narrow down the range of front plate gaps that create a noticeable impact on the flow with a minimum of effort. 

We started with a 30 mm gap, where there was no observable impact on the tufts. We then halved the gap to 15 mm, where we observed some twitching on the innermost of the three tufts. Stepping in again at 5 mm, all three tufts began flopping around with the innermost tuft being pulled completely backwards into the fan indicating that the fan is sucking in air from the back. This initial exploratory mapping tells us that our area of interest is a fan face to front plate gap of approximately 5-15 mm. Before proceeding with more complex tests, we confirmed this exploration by stepping the gap from 5 to 15 mm in increments of 2.5 mm.

We have extracted a lot of good qualitative information about the parameter space from this quick Mk. 1 Eyeball based test. However, to see the behavior in more detail, we turn to high-speed fluorescent tufting. Leveraging the Chronos 4K12 and a large 400 nm UV-LED light source, we observed the behavior in detail at gaps of 5 mm and 15 mm, which you can see slowed from 1,400 FPS to 40 FPS in Figure 2. For this particular case, it tells us nothing more than what we could see by eye. However, in more complex flows, such as the wing-fuselage junction of an airliner, high-speed fluorescent tufting can identify the onset of detrimental phenomena such as flow separation. With just some very scientific string blowing in the wind.

Figure 2 Side-by-side of high-speed tufting.

Component Summary:

Smoke in the Wind: Particle Image Velocimetry (PIV)

Now, with a more focused parameter space, we can look at the flow behavior with more detailed and complex measurement techniques. The current gold standard for measuring the velocity field around an object in air is Particle Image Velocimetry (PIV). This technique was first developed by Ronald J. Adrian and Yao Chung-Sheng of the University of Illinois Urbana-Champaign in 1984. After completing this foundational work as part of his Ph.D. thesis work, Dr. Yao came to the Flow Physics and Control Branch at NASA’s Langley Research Center, the organization hosting LTT for this experiment. Dr. Yao retired in 2025 after 35 years of service, continuing to make significant advances to the field he pioneered as a graduate student and mentoring countless researchers in the field, including myself. Today, PIV is a widely commercialized technology at the heart of every advanced fluid dynamics program, from fundamental fluid dynamics studies in government and academic labs to the high-stakes production wind tunnels of every Formula One Team.

PIV works by filling the flow with small light scattering particles, or seeds, and then taking a pair of images with exposures defined by short, bright pulses of light, typically from a frequency doubled neodymium-doped yttrium aluminum garnet (Nd:YAG) laser. The core concept is that velocity equals distance divided by time. The distance these particles move in the image frame, converted from the frame to the real world through a camera calibration, divided by the time between the two light pulses, yields velocity. In the 1980s, researchers used a pair of film cameras to achieve the required speed and sensitivity for PIV. Then they scanned the developed prints with a 128x128 pixel camera for digital processing. Processing a single PIV frame pair could take days. For the LTT fan experiment, we used the DaVis FlowMaster PIV system with a high sensitivity, low noise 5.5 Megapixel sCMOS camera. This is one of numerous commercially available PIV systems that span a wide range of price and capability. Our use of it here was primarily for convenience, as it was the system already available in the lab. 

We deviated from the typical PIV implementation in one major way. Instead of using a laser to produce the double light pulses, we used a pulsed LED to avoid damaging any human eyes or camera sensors. For expediency, we borrowed the pulsed light source from a high-speed schlieren imaging experiment (explaining schlieren is a different article sorry). The custom-built unit couples a high power laser diode driver with a red/amber LED rated to 36 A peak current. Because we do not particularly care about diode longevity, we use the laser diode driver to shove over 240 Amps through them in bursts of 1-10 microseconds. With appropriate cooling, these LED’s handle overcurrent like a champ, producing high intensity bursts of light perfect for freezing the image frame.

This decision turned out to be less expedient than we planned. We first attempted 3D stereo PIV. This would have resolved the swirl and profile of the flow across the fan face. However, Mie scattering, the mechanism by which seed particles are illuminated, is less intense and more asymmetric at longer wavelengths. This made it difficult to obtain a similar signal-to-noise ratio between the two stereo cameras. Although we could have replaced the LED with a shorter wavelength model to produce a better PIV signal, time constraints pushed us to revert to ever reliable 2D planar PIV with a single camera. We position the light source downstream of the fan producing a light sheet that cuts a perpendicular slice of the flow that exits the fan, as shown in Figure 3. This configuration does not resolve the “out-of-plane” component. I.e., any flow to the left or right of the stream. As such, it does not resolve the swirling flow that a fan produces, just the bulk movement of the air stream. However, that is the bulk streamwise component that matters most for cooling. For seed particles, we positioned a handheld fog machine upstream of the fan to seed the flow with particles. The fogger thermally atomizes a glycerine and propylene glycol liquid to produce a cloud of particles that are drawn into the fan and illuminated by the light sheet.

Figure 3 Streamwise 2D PIV experimental setup.

The final piece of the PIV puzzle is the timing of the light source and camera exposure. PIV cameras use a special dual buffer for image data that allows for two frames to be taken in sequence with an adjustable interval. The light pulses are set to straddle these two exposures with the timing between the light pulses effectively controlling the time change used in computing the velocity from the PIV cross-correlation displacement field. Figure 4 below shows an example sequence of PIV image pairs with 4 microsecond light pulses separated by 800 microseconds.

Figure 4 PIV Image Pair Video.

Approximately 150 image pairs are recorded for each test condition. First, without a front plate, then with a gap of 22 mm, 15 mm, and 5 mm from the fan face. After these tests, we attached a 25 mm thick AIO radiator to the exhaust side of the fan with 15 mm and 5 mm front plate gaps. This represents the fan operating in a push radiator cooling configuration. The image pairs are processed through the Lavision DaVis software, although numerous free and open source alternatives exist to perform this analysis. Each image pair is intensity normalized, and then regions with no particles are automatically masked out. The key to PIV is to slice the image into a series of overlapping analysis “windows" rather than performing the analysis at once on the full image. Each, in this case, 32x32 pixel window is run through a cross-correlation algorithm that identifies the pixel shift of the particle pattern within that window and image pair. With a properly calibrated camera and a known time between the two correlated frames, each window yields a single velocity vector. This is a brilliantly parallelizable process that stress tests your CPU harder than Prime95.

The result is 150 snapshots with a velocity vector estimate at the center of each window. We then applied two layers of outlier detection before collapsing the data down to a single average velocity field. First, vectors with low correlation coefficients are removed from each individual velocity field. Then the Grubbs outlier test was applied across all 150 frames. Below is a representative flowfield from the unobstructed test case without a front plate or radiator installed. The field-of-view is focused on the upper half of the fan, with the center of the fan hub representing a radius of r = 0 mm and the top of the fan disk at r = 60 mm. The streamwise direction is from right to left along the centered slice of the light sheet. In a PC case reference frame, the outside of the case is to the left where the obstructing plate is, the fan is on the y-axis, and the plot shows the flow into the computer. The colormap shows the streamwise velocity component, u, with arrows overlaid to indicate the local direction of the flow. The streamwise velocity field for this and other cases will be explored further in the results section of this article.

Figure 5 Unobstructed example of PIV results showing a streamwise slice of the flow without a front plate on the fan.

PIV Experiment Equipment:

Waves in the Wind: Aeroacoustics

The tufting and PIV tests show the time-averaged structure of the flow behind the fan in each of the front plate gaps and the radiator test conditions. In addition to cooling capacity, the next most important factor for a PC cooling fan is acoustics. Does the presence of the front plate make the noise level of the fan better, worse, or just different? To answer this, we turned to the NASA Langley Small Hover Anechoic Chamber (SHAC), formerly known as the Small Anechoic Jet Facility (SAJF). The SHAC, shown in Figure 6, is an acoustically treated room of 3.87x2.56x3.26m for noise sources above 250 Hz.

Figure 6 Diagram of SHAC Facility with placement of linear array microphones and test model

As the name would suggest, this is an acoustic anechoic chamber originally designed to measure the noise output of small jets and later modified to test rotor blades for electric vertical take-off and landing (eVTOL) systems. These are both significantly louder noise sources than your average PC case fan, especially one as famously quiet as the Noctua NF-A12X25. Due to this, the A12X25 was close to the noise floor of the facility. To solve this, we swapped it for the Noctua NF-F12 industrial PPC-24V-3000 SP IP67 PWM. Note that in the images and video of this test, the normally black and brown Noctua PPC fan has pink blades. This is due to a thin layer of high-speed pressure sensitive paint (PSP) applied to the blades. The plan was to phase-lock the high-speed camera to the PWM tachometer signal to visualize the surface pressure distribution over the fan blade, much like we do on a helicopter or eVTOL rotorblade test. However, with only one day on hand and so much to do and see, this test did not make it. We drove the fan at 20 V and 100% PWM resulting in approximately 2600 RPM. The dominant tone for most fans is the blade passing frequency, the frequency at which a blade passes through any given point in the fan disk. Since this is a seven bladed fan, we expect a blade passing frequency of approximately 300 Hz to dominate the acoustic spectra.

We performed two types of acoustic measurements in SHAC. First, an acoustic directivity test with a linear array of eight free-field microphones along the top of the test cell, noted as M1-8 on Figure 6. The array measures the angle, θ, at which noise propagates outwards from a source, in this case the fan. Zero degrees is looking straight at the face of the fan, while 90° is looking at the right or left side of the fan, respectively. In the context of the computer on your desk, facing the front of your case dead-on vs. sitting to the side of it. Note that we did not run a calibrated noise source prior to these tests, so do not focus on the absolute decibel values that are shown. What is important is the relative difference between the two test cases: the unobstructed fan with no front plate vs. a front plate gap of 15 mm.

The second test was using a beamforming array, or acoustic camera, to identify the source of noise from the perspective of an observer. The particular array was a  

Signal Interface Group (SIG) ACAM 140 with a logarithmic spiral of 40 microphones on a 40x40 cm plate spatially calibrated to a digital camera at its center. The array was positioned 5° off-angle approximately 0.3 m away from the fan under test. The minimum resolvable frequency of the array is a complex function but generally depends on the size of the array, the distance from the object under test, and the constituent microphones. For this particular case, the floor to reliably localize a noise source is approximately 1 kHz. As such, when trying to identify the source and propagation of the 300 Hz blade passing frequency, it is necessary to use higher frequency harmonics for the tone. These are multiples of the base frequency that are akin to shadows of the original signal that become progressively weaker at each higher harmonic. For the purposes of this test we isolated the tonal blade passing frequency around the third harmonic at 1.2 kHz and then inspected non-tonal noise between harmonics at 2 kHz.

Aeroacoustics Experiment Equipment:

Results from Research Data to Actionable Insight

PIV Data Implications for Cooling Performance

Now that we have explained all of our methodology and best-laid plans, it is time to tackle the interesting bit: results and insights. We saw most of what there was to see from our design-space exploration with the tufts, so let’s get right into the PIV results. Studying the smoke, the math reveals the velocity field of a slice of air exiting the fan disk. The colors show the streamwise velocity (right to left) of the air, and the arrows show the directionality and relative strength of the flow. First, we look at the baseline case without a front panel. We see a smooth flow with a plume of fast moving air coming out the back of the with a small area of no flow immediately behind the fan hub. Any unevenness in the air as it is accelerated along different parts of the blade is quickly evened out to create a strong plume of approximately 2 m/s air. It is also interesting to note that Noctua is definitely doing some creative things with blade tips to throw more momentum inward. This improves the static pressure and acoustic profile of the fan instead of radiating outwards and buzzing the frame.

1 of 4

Figure 7 PIV Results for the fan with varying front plate gaps. Colors indicate streamwise velocity; warm for flow to the right, cool for flow to the left.

Now we add the front plate, first with a gap of 22 mm. Other than some minor differences expected from test-to-test variation, these are, for all intents and purposes, identical flowfields. This implies the 22 mm gap has no impact on fan performance at the 100% PWM test condition. Next, the gap is reduced to 15 mm, the condition in which we saw the innermost tuft start to flop around in separated flows. In the PIV velocity field, the stagnant, or not moving, area near the fan hub has increased in size. According to Bernoulli's principle, the faster the air moves, the lower the static pressure it has. Thus, the stagnant flow at the center of the fan pushes the lower pressure stream of air outward to create an upward curl to the flowfield. This combines with the momentum deficit behind the fan hub to create a downstream deadzone at the center of the fan that is not present in the 22 mm gap and no plate cases. This means less momentum in the air mass flowing over your components, convecting heat away from your 12VHPWR cable. In general, while the 15 mm plate gap case is not ideal, it is functional as an intake fan with mostly streamwise airflow at similar speeds to the unobstructed case.

Now let’s take things to the extreme with a front plate gap of 5 mm. A case representative of putting your PC against the wall or a PSU intake on the ground. Now, all the fan can do is barely suck some air in around the edges and sling it through at the tip. It is so starved for air coming in, it creates a reverse flow sucking the stagnant air near the hub backwards and outwards. While observable with the tufting previously, the PIV renders the flow full pattern allowing us to extract detailed physics insight into the cause of that reverse flow. We see a slow, swirling vortex with maximum flow speeds less than 0.5 m/s and virtually no streamwise flow to convect heat away from the PC components. It would just recirculate hot air within the PC case, adding to that heat by the waste energy of the uselessly spinning fan motor. Ultimately, just a Noctua brown work of art or RGB light, not a cooling fan.

The next question is what happens when you attach a 25 mm thick 120 mm radiator to the back of the fan in a push configuration. And let's face it, if you have a glass front panel to show off the brown majesty of your Noctua fans, you are using a push configuration. We tested our edge case with a 15 mm gap and the extreme case at 5 mm with the radiator attached, as shown in Figure 8. For the 15 mm gap case, the radiator acts as a flow straightener, so we don’t see that outward turning anymore. However, the maximum airspeed is halved to approximately 1 m/s with a significant increase in dead air towards the fan hub compared to the radiator free case at the same front plate gap.

Figure 8 PIV results with a 25 mm thick radiator and a 15 mm front plate gap. Colors indicate streamwise velocity; warm for flow to the right, cool for flow to the left.

You can think of blockage as a force pushing back on the momentum the fan can impart on the flow, as defined by conservation of momentum. In this case, there are two primary types of blockage that push against the fan. First, mechanical or physical blockage produced by the frontal area of the radiator fin array. The second and more important one is the frictional blockage. This is the momentum lost to skin friction through the boundary layer along the surfaces of the fins. This is the mechanism that causes thicker radiators to require higher static pressure fans. The surface area of the radiator with air flowing over it is reduced by about 40%, resulting in a marked reduction in the convective cooling capacity of the radiator. A further side effect if the radiator fans are your only intakes is lower airspeeds in your case. Hot air is likely to stagnate around and heat other components, like your precious RAM, if not extracted by exhaust fans.

Figure 9 PIV results with a 25 mm thick radiator and a 5 mm front plate gap. Colors indicate streamwise velocity; warm for flow to the right, cool for flow to the left.

Now, just for fun, let us see how bad it can get with your radiator fan being against the wall or floor with a 5 mm front plate gap in Figure 9. Perhaps we see a light breeze from the open door at the corner of the lab but from the fan: just a sea of purple with essentially zero flow going anywhere. Your radiators only hope is some natural convection to pull heat away here because that fan is now just a decorative RGB wall splash. Please don’t do this, think of the poor silicon.

From the PIV measurements, there are a few TLDRs. If you are just running the fan as a case intake, you can thermally get away with about a 15 mm gap between the fan face and a front panel, wall, or floor (see caveats in the methods section). However, if that same fan is on a radiator, you will take a significant cooling reduction, and a larger gap is recommended. Back your PC away from the wall or reconsider that glass front panel case. If you absolutely must show off the spinning RGB up front, put some pull fans on the back of the radiator and some exhaust fans to help your hardware perform at its best. In both intake and radiator cases, do not have your case slammed against the wall or use the front radiator mount on a glass case with a gap of less than 15 mm. Perhaps even give your PSU bottom intake fan a break and lift it a little bit above the floor, desk, or silicon gods forgive, the carpet.

Acoustic Implications of the Front Panel Gap

Intuitively, one might think that a solid front panel in your case would reduce the acoustic signature of the fan. You are adding a physical block between the spinning noise source and your ears after all. However, fluid dynamics is a harsh mistress and the reality can be quite the opposite. Figure 10 shows the measurements of the linear microphone array with each of the eight microphones at different angles relative to the fan face, as shown in Figure 6. The red curve with diamond markers is the unobstructed case with no front plate on the fan, and the blue curve with circular markers has the front plate installed with a 15 mm gap. The x-axis shows the angle relative to the fan of the microphone. The y-axis shows the overall sound pressure level (oSPL) in decibels at that microphone. Because we did not perform a complete microphone calibration on the linear array, we focused on the relative differences in oSPL between the two cases and not the absolute values in dB.

Figure 10 The acoustic directivity of the linear array and the acoustic spectra of the center M4 microphone. Red is an unobstructed fan without a front plate. Blue is with a front plate with a 15 mm gap.

The blue 15 mm gap covered case is consistently 4-8 dB louder than the red uncovered case across all angles. It is necessary to dig deeper into the data to find the source of this significant increase in fan noise. First, look at the lower plot in Figure 10 that shows the acoustic spectra of the M4 center microphone for the covered and uncovered faces in blue and red, respectively. The blue covered case curve has a broad spectrum bulge compared to the unobstructed spectrum in red. This is indicative of separated turbulent wake noise as opposed to the tonal noise of the blade passing frequency harmonics that create sharp peaks in the spectrum. Looking back at the PIV data for the two cases in Figure 7 confirms the source. In the time-averaged representation of the flow field, the region of stagnant air around the fan hub indicates that there is indeed a separated wake. One can think of a separated wake as a chaotic mix of swirling flows with vortices or eddies of different sizes that all interact together within the larger vortex of the separated region. This is a simplification but helps express the nature of the turbulent energy cascade as many different flow scales interact to produce a broad spectrum of turbulent pressure fluctuations or noise like that seen here.

Another notable feature of the acoustic directivity in Figure 10 is how much flatter the oSPL is with a range of 1 dB and 6 dB for the covered and uncovered cases, respectively. The prominence of several blade passing frequency harmonics in the front plate covered acoustic spectrum is reduced relative to the uncovered case. Both of these indicate that the presence of the front plate is changing the way the fan noise radiates outward. To better understand this behavior, we can turn to the acoustic beamforming images in Figure 11. These show us where noise in different frequency bands appears to be originating from the point of view of the array. First, on the left of Figure 11 we can see how the noise intensity around the 1.2 kHz blade passing frequency third harmonic intensifies. The plate appears to amplify this harmonic, creating a more intense and spatially larger noise source that extends beyond the fan disk itself. The overlap with the separated turbulent wake energy around the 1.2 kHz harmonic serves to dump more acoustic energy forward into the plate and outward to the listener. This effect is even more pronounced in the 2 kHz frequency band on the right of Figure 11. This band is away from the stronger blade passing harmonics, but still within the broad spectrum bulge we identified as the separated turbulent wake earlier. The pattern is consistent with the acoustic energy spilling out from around the edges of the plate and radiating from the unconstrained surfaces of the plate. The thicker lower lobe of the pattern also indicates some reflection upward from the mounting bracket.

Figure 11 Acoustic beamforming video of the uncovered (top) and 15 mm front plate gap (bottom) fans in SHAC. Apparent sources for the 1.2 kHz (left) and 2 kHz (right) frequency windows.

Without the front plate, there is relatively little energy in the 2 kHz band, with most of it originating from some blade-frame interactions in the direction of the listener. Most of the total acoustic energy is directed straight out of the fan and dominated by the blade passing frequency, resulting in an off-axis drop-off in oSPL. However, the front plate adds insult to injury. Not only does it disrupt the flow to create the separated wake, it blasts that energy out in all directions. You can no longer just point your computer away for a little acoustic relief. A more tailored case design with rubber in all the right places might mitigate this. It is a reminder that good case design requires serious engineering.

Concluding Thoughts: Always More Questions

Returning to our original question: how close is too close? Well, for a single fan Noctua NFA12X25 operating at 100% PWM, anything closer than 15 mm is probably too close. However, there are many confounding variables such as case design, fan-fan interaction, fan curves, radiator thickness, etc. that make the exact numbers we measured in this study less important than the general trends we observed.

First and foremost, we observed that the gap between a front panel or wall and the fan face does matter for both cooling and acoustic performance. In addition, you are more likely to induce acoustic performance problems at a wider intake gap than you are to cause cooling performance problems. We saw that radiators have a large impact on how tolerant a fan is to intake obstructions. So, where acoustics is no object, such as a server room, you can get away with closer spacing between intakes and obstructions. But for the tower that sits on your desk, give it some room to breathe. Pick it up off the floor, pull it out of the corner, and consider intake airflow when purchasing your next case. Adding a radiator complicates things with the detrimental cooling impact occurring at a much larger gap than when unobstructed. Keep in mind the blockage a radiator will cause when deciding where to mount it for the cleanest flow in your particular case.

It’s more knowledge than we started with and served as an excellent vehicle to highlight how we work at NASA Langley to understand the fundamental fluid dynamics that will unlock the faster, quieter, and cleaner aircraft of tomorrow. I personally came out of this day of testing with more additional questions than answers. We had one day to get as much done as possible and imposed on the generosity and time of many of my colleagues across Langley. As such, we did not have the time to really sit down and dig through the results of one test before rushing on to the next. For example,  It would have been great to take some of the insights we learned in the acoustic chamber back to the bench to run some more detailed PIV around the intake and fan face. Or to have some acoustic chamber time after I finished processing the radiator PIV data in the following weeks. I still wonder if the flow straightening effects of the radiator would prevent the formation of the separated wake and mitigate some of the acoustic problems caused by the front plate. This and a thousand more questions that we wish we had time to answer. What if we stick a piece of shag carpet in front of the intake instead of an acrylic plate? How does this translate to packing PCIe cards tightly together with your GPU cooler?

This mirrors many of the day-to-day realities of doing research. Research is always incomplete; every test leaves questions unanswered. Every answer sparks new questions we couldn’t have conceived before. Not every answer is a step forward and not every avenue of research ends in a world changing innovation. It is threads of knowledge on how to do something or, often more critically, how not to do something that weave together into technologies that will define the future. It intertwines with the needs and ideals of society and how society grapples with the implications of that research. It is nurturing a sense of discovery in ourselves and the next generation of thinkers through outreach like this collaboration with LTT.

Behind this collaboration and behind every paper, presentation, or patent at NASA is an immense workforce with numerous constraints. Key facilities like our wind tunnels are simultaneously irreplaceable and in high demand, while becoming increasingly expensive to operate. A shrinking cadre of highly trained technical staff keep these often decades old facilities operating and producing world-class data. Yet more individuals move the financial and logistical levers to sustain this enterprise in the near- and long term through constrained budgets and shifting winds. There are communications professionals like Brittny McGraw who are critical to enabling collaborations like this one and communicating the importance of NASA to the public. All are working together to build the future of flight and inspire the next generations of innovators and researchers.

If something in the video or article sparked a desire to learn more about NASA, its work, and its history, I encourage you to dig deeper. There are a ton of Floatplane extras from this collaboration, so head over there for a behind-the-scenes look at NASA Langley. Check out the NASA Technical Reports Server (NTRS) to access almost everything NASA and its predecessor the N.A.C.A have published for over 100 years. Explore the wonderful NASA History Series of ebooks that document the highs and lows. Some of my personal favorites are listed below. See where your curiosity takes you and what questions you will find. Thanks for reading and remember that NASA is with you when you fly.

Read Full Article open_in_new
arrow_back Back to News