GPU Thermal Tests in an Inverted Case

[evanevanevan]

I posted a picture of my build on a Discord server, and got multiple responses from people who were incredulous at the idea of top-mounted intake fans. Here's the picture in question:
View attachment 20240126_190209_1.webp
Didn't seem all that offensive to me, honestly.

The case is a be quiet! Silent Base 802, changed into its "inverted" configuration, hence the upside down motherboard. My theory here was that top intake is mostly unrestricted, and having the GPU pointing fans-up will lead to better GPU temperatures. So, let's test that!

The case comes with two top panels: a mesh one, and a closed-off one. I'll do some tests with both arrangements:

I'm running Linux, and I don't have any real games installed on this system right now, so options for graphics benchmarks are a bit limited. I decided to go with Unigine Superposition as my benchmark... except Superposition kept freezing as soon as it opened. Okay, I guess I'm using Unigine Valley instead. Valley isn't a very heavy load for an RTX-series GPU, but it's enough of a thermal load for the purposes of this test.

Here are some graphs showing how different top panel/fan configurations performed. I ran four different test configurations:

• Blue: mesh top panel, fans set to the "1" setting on the fan controller. They're inaudible at this speed, barely spinning.
• Green: mesh top panel, fans set to the "3" setting on the fan controller. This seems to be 100% fan speed.
• Red: mesh top panel, fans turned off (unplugged them from the fan controller)
• Orange: solid top panel, fans turned off

This first graph is the GPU temperature reported by nvidia-smi during the test. You can very clearly see where Unigine Valley starts and ends. All four tests are pretty close temperature-wise, but the configurations with mesh panel + fans are running are a couple degrees cooler. Also notice how the temperature trails off at the end of the test: the temperature declines much more quickly when the fans are feeding air directly into the GPU. The "mesh, no fans" and "sealed, no fans" configurations see very similar temperature dropoffs at the end of the test, indicating that the chimney effect isn't doing much to help the "mesh, no fans" configuration:

This graph uses a thermistor that I shoved under the GPU heatsink, trying to get it as close as possible to the VRAM modules. This isn't a true "VRAM temperature" measurement - it's more like a "board surface temperature near the VRAM" measurement. The "mesh top / fans on" configs are once again ~3-5 degrees cooler than their fanless counterparts. I should have let the tests run longer to fully capture the temperature decline at the end, but you can still clearly see that the intake fans are doing their job.

My takeaways:

• Top intake with inverted cases works! Force-feeding a GPU with cold air makes it run cooler, go figure.
• There's virtually no difference between "mesh top / no fans" and "sealed top / no fans." In other words, the chimney effect isn't doing anything meaningful here.
• All of these configurations would be fine with an RTX 3060 Ti - it isn't getting anywhere close to its thermal limits in any of the tests.

Hopefully this will be helpful to somebody considering an inverted case layout. I'm coming from a case with a huge side intake fan, so it's reassuring that I can still blast cool air directly into my GPU.


Side note: something really remarkable with this setup is the "thermal barrier" the GPU makes within the case. I can reach my hand into the case, and feel a major temperature difference between the air above the GPU, and below it. The CPU is water-cooled, so I'm not really worried about the relatively warm area under the GPU.

 

[evanevanevan]

In the last week, I've made a good number of tweaks to this PC, and I wanted to run a similar battery of tests again. First, let's go over what I've changed since last week:

• The GPU has been flashed with a VBIOS that allowed me to bump up the power limit, which allows it to get closer to its thermal limits.
• Enabled PBO on the CPU (oops)
• Adjusted the CPU fan curve - it was waaay to aggressive before.
• Updated Nvidia drivers
• Got Unigine Superposition working.

This time around, I decided to do a true "intake vs. exhaust" head-to-head comparison. The results were... um... decisive, to say the least.

Key:

• Green line: mesh top panel, top fans set to "3" using the fan controller.
• Blue line: no top panel, top fans set to "3" using the fan controller.
• Red line: solid top panel, top fans set to "1" using the fan controller.
• Yellow line: mesh top panel, top fans set to "3," but set up as exhaust instead of intake.

First, let's look at the the GPU temperature reported by nvidia-smi:

The "exhaust with mesh top panel" and "exhaust with no top panel" tests show pretty similar results - just a degree or two's difference under load. That's good! It means that the case's dust filter isn't causing unnecessary thermal troubles. Predictably, the sealed top panel test runs significantly hotter than the two tests with max-power intake fans.

The really surprising thing is that yellow line: the setup with two max-power exhaust fans is almost identical to having no ventilation at all! I figured exhaust would perform worse than intake in this setup, but I didn't expect the difference to be that significant. Also, notice that temperatures plateau at ~83C for both the exhaust and sealed tests. This is a symptom of some form of throttling, be it thermal or power. Meanwhile, the top-intake test runs (blue and green lines) are still slowly ramping near the end of the test. I didn't actually save the Superposition benchmark results, but I should have, because there may be some measurable loss in performance when running top exhaust here.

The thermistor chart is even more damning for top exhaust:

This thermistor is crammed under the GPU's heatsink, and is pressed up close to a VRAM module. It's not truly a "VRAM temperature measurement," but it's as close as I can get with the tools available to me. These results are even more surprising: For exhaust, the termistor reads a few degrees hotter than a sealed-off top panel!

So, why is this happening? Obviously, having exhaust fans fighting with your GPU is probably bad, but I didn't expect it to be as bad as having almost no ventilation near the GPU at all. Here's my guess: top exhaust pulls air upwards. This may draw heated air from the front-mounted AIO up towards the GPU, when it otherwise would have been pulled downward and out by the PSU or rear exhaust fan. In any case, the results are pretty decisive: if you have an inverted case with an air-cooled GPU, you should probably run top intake.

 

[Vya Domus]

and got multiple responses from people who were incredulous at the idea of top-mounted intake fans.

For a good reason, the internals are inverted in your PC, in 99.9% cases they aren't going to be and it makes no sense to invert the top fans. Your test isn't really that insightful, this is going to be the same optimal configuration in any PC.

 

[evanevanevan]

For a good reason, the internals are inverted in your PC, in 99.9% cases they aren't going to be and it makes no sense to invert the top fans.

Well yes, that's why the title of the post says "inverted case," the photo originally shared with the folks in question shows an inverted build, and I mention the inverted-ness of the layout no less than 3 times in the original post.

It's intuitive that it should work this way, but the goal is to see "how well does it work." There's a lot of pseudoscience with case layouts. Positive pressure, negative pressure, intake and exhaust placements. Everyone has a hot take (and some even have a hot case), but there isn't much actual data to make layout choices with.

 

[Veseleil]

Well yes, that's why the title of the post says "inverted case," the photo originally shared with the folks in question shows an inverted build, and I mention the inverted-ness of the layout no less than 3 times in the original post.

It's intuitive that it should work this way, but the goal is to see "how well does it work." There's a lot of pseudoscience with case layouts. Positive pressure, negative pressure, intake and exhaust placements. Everyone has a hot take (and some even have a hot case), but there isn't much actual data to make layout choices with.

I've tested full negative pressure build and it was awesome as I thought it'll be, so decided to make custom cases with that layout in mind for all my PCs. If you have time, check that thread, and if no one else, I encourage you to continue with your testing. Cheers.