Some Powder Toy CFD

[BlivetWidget]

First off, let’s establish that Powder Toy (https://powdertoy.co.uk/) is not a calibrated computational fluid dynamics engine. I have professional experience with the industry standard CTH and LS-DYNA but that’s a bit beyond the scope of this post. But, Powder Toy is way less clunky, it’s free, and anyone can grab it and do some testing of their own with very little practice (I just discovered it for this project, myself). And the basic principles of fluid flow are there. So long as we don’t try to derive specific velocities/temperatures, etc. it should provide a good tool for comparison.

Anyway, I was thinking about computer cases and came across some Gamers Nexus testing where the Fractal Torrent (Torrent — Fractal Design) and Phanteks Eclipse P500A (the newer version of that case, nearly the same: Phanteks Innovative Computer Hardware Design) were reviewed. They were both high airflow cases with excellent temperatures, but for some reason, the Fractal Torrent lagged behind on GPU cooling (still good though). It was a bit of a head scratcher, as most of us would probably assume that those 3 fans on the bottom would give it the best GPU cooling since they blow right into the GPU!

So I decided to model these cases in Powder Toy and see what was going on. 1 small grid space = 1 cm, if anyone is interested in doing some playing around of their own with other cases.

Torrent on the left, Eclipse on the right. Stock fan configuration (this means not all fans shown are active). Red is high pressure, blue is low pressure, white is air velocity. No point worrying about temperatures; remember this isn’t a calibrated simulator and the temperatures will be controlled by airflow anyway. I added a CPU cooler of the type that’s needed for modern high-end CPUs, and a modern, beefy GPU based on the 4000-series RTX cards (with some top ventilation because this is a 2D simulation and there’s no way for air to get around the card otherwise).

Not much objective we can say from this so far except that there’s a lot more pressure in the Torrent (no surprise there, all fans point in). The Eclipse seems to be shooting more air out, though, and the air under the GPU appears to have a lower pressure. So I added streamers to the simulation that will give us a better idea of air movement directions.

From what I can tell, it looks like the Torrent is pushing so much air to the back with those two huge 180 mm fans that it’s bypassing the GPU and just exiting the case. On the Eclipse, we can see more of the streamers getting sucked up through the GPU. Now, I will caveat this by saying that varying the fan speed can have some interesting effects on the simulation, and there’s really no good way to calibrate it to reality.

That said, I think this provides a good theoretical explanation for the better GPU cooling performance of the Eclipse.

Next I plan to have a look at those top fans on the Eclipse. Conventional wisdom is they need to be blowing out (there are plenty of reddit posts proclaiming anyone who thinks otherwise to be an idiot), but I feel like that’s going to just suck out the air before it even gets to the CPU.

[GDBringer]

I have always been curious about optimizing my own desktop case. But I never had the anemometer to measure these kinds of tests. Interesting stuff.

The only recent video I saw on the subject was of a guy 3D printing a funnel to force air through the case.

I fixed PC cooling.

Though my unproven anecdote from living in a hot place was to first open 2 windows at opposite ends of wherever I want air to pass through. Then run a box fan a foot away from one window facing outside. So instead of air being pushed indoors, the air is pulled out the window. I don’t recall where the idea came from. But my thermostat tends to show me it is cooler otherwise.

@BlivetWidget

Now, I will caveat this by saying that varying the fan speed can have some interesting effects on the simulation, and there’s really no good way to calibrate it to reality.

Oh, but there is! It is my favorite program!

GitHub - Rem0o/FanControl.Releases: This is the release repository for Fan...

You can manually control the RPM of each fan in your case AND it can be programmed to automatically increase or decrease their RPM in relation to the CPU and GPU temperatures. So if you have an ideal airflow, for low temperatures or high temperatures, it should be possible with this.

[BlivetWidget]

Thanks, yes, it’s interesting toy =)

That was a really cool video. The ducting sends the air exactly where it needs to go so it can’t bypass the critical systems. Would be interesting to see if the mainboard and other systems are getting hot though, since they see less airflow that way. They produce a lot less heat so maybe it’s not an issue. As an aside, that angled fan on the bottom of the H5 is really neat. Kind of a shame the H7 dropped it.

@GDBringer

Oh, but there is! It is my favorite program! You can manually control the RPM of each fan in your case AND it can be programmed to automatically increase or decrease their RPM in relation to the CPU and GPU temperatures. So if you have an ideal airflow, for low temperatures or high temperatures, it should be possible with this.

Ah, I think perhaps I was unclear. When I say the fan speed cannot be calibrated to reality, I mean that Powder Toy does not have a means of setting RPM or unitless “CFM”. The simulation’s fan speed is entirely guesswork, and not numerical. It’s not obvious to me how to translate the click and drag speed control into the unitless space of the simulation in order to make a real world equivalency.

[markwr]

@BlivetWidget

lot more pressure in the Torrent (no surprise there, all fans point in). The Eclipse seems to be shooting more air out, though, and the air under the GPU appears to have a lower pressure

My guess upfront would have been that whichever case aimed more at “pulled” air out (likely also meaning lower pressure) would (in general) be the case more likely to be the best at cooling the interior (and in general, the components). That is of course assuming that the room temperature outside the case is cooler the the average temperature inside the case (otherwise you’d be in trouble, heh).

When I say “in general”, I mean “all other things being equal”. A factor that could disrupt my assumption is the volume of air flowing against the surfaces of the components you aim to cool.

My intuition is based on a few factors:

1. The more internal air being displaced by external air (assuming external air is cooler), the lower the average temperature of the internal internal air should become.

2. As that internal air flows against the surfaces of the targeted components, the cooler that air is, the more heat the components release.

3. I assume (I did not look this up) that the lower the air pressure surrounding an object (in this case the target components), the higher the rate of heat released from that object. Of course, in this instance, the difference in air pressure is so small as to make very little difference.

I predict using fans aimed more at pulling air out of the case, as opposed to pushing air in, would be the more effective design (most likely to replace the highest volume on internal air with external air). The second key decision would be to add internal surfaces that would direct the air movement inside the case in a way to efficiently move it over the surfaces of the targeted components.

[BlivetWidget]

Regarding your intuitions:

@markwr

The more internal air being displaced by external air (assuming external air is cooler), the lower the average temperature of the internal internal air should become.

Yes. I wouldn’t think of it in terms of averages, though there’s a way to make that work with more steps. The external air needs to be cooler than the heat-producing elements inside the enclosure. If you want to think of it in terms of averages, the heat-producing elements raise the average internal temperature given time, but we don’t need to wait for that to happen.

@markwr

As that internal air flows against the surfaces of the targeted components, the cooler that air is, the more heat the components release.

Yes. Energy transfer is relative, so a greater difference will mean more energy transfer. But, for a PC we generally have no control over the temperature difference, so it’s not really our focus. What we can control is quantity of air.

@markwr

I assume (I did not look this up) that the lower the air pressure surrounding an object (in this case the target components), the higher the rate of heat released from that object. Of course, in this instance, the difference in air pressure is so small as to make very little difference.

This one is backwards. A denser fluid takes heat away faster. This is why you can get hypothermia from enough time in 70 F (21 C) water but the same temperature air is just about perfect indefinitely. It is also why a vacuum flask holds temperature better than a mug. However, the pressure buildup in a positive or negative pressure PC case is really so small I think it has no measurable effect on density. Cases are simply too unsealed, and the fans are not turbine pumps able to fight any significant pressure differential. Most discussions on positive vs negative pressure have to do with the path the air will take to enter/leave the case, and are usually focused on dust.

[BlivetWidget]

Okay, some top fan testing on the Phanteks Eclipse G500a. All fans are 140 mm. Default fan configuration for the “performance” model is 3 in the front, and 1 out the back.

Adding 3 top fans pointing out:

3 top fans pointing in:

First 2 top fans in, rearmost out:

Setting the top fans to outwards of course places us in a negative-pressure scenario. Internet hates that but this case doesn’t have proper dust filters anyway, so I doubt the air coming in through the gaps is much dirtier than the air coming in through the fans. I’m hoping they make a G600A based on the P600s that actually has some dust filters, though.

Interestingly, not that much difference between the other test cases. Almost as if the stock configuration is well-designed… I think my only complaint is the open top in that case, though I understand why they put it there (to allow for liquid cooling radiators). Maybe there should be a solid panel that can be swapped with the mesh for an air setup, because it doesn’t really seem (from this limited simulation) to have much airflow benefit.

The thing that is difficult to control, and why there are endless internet arguments over fan configurations, is exactly where the air is going. This is highlighted in the video linked by @GDBringer - getting air into the case isn’t the same as directing it over the components of interest. Case fans really aren’t that powerful so the air they push in doesn’t stay a nice tube. You pull it into the case, or push it out, but as the above shows, it’s hard to make a difference at the actual components.