A few weeks back we tried to cram a single slot, half height GPU into a compact case with very little airflow, and got all surprised Pikachu face when it started overheating. Let’s see if we can’t improve on this situation.
Welcome to Esk Computers, I’m Scott and today on the desk of Esk, let’s harness the amazing power of wind to try and tame these wayward electronics.
We have the P1000 in here today, and to aid plans we’ll get to later I’ve taken off the backplane where the mini DisplayPorts live to encourage some airflow out of the case. I think we’ve established last week that the T1000 is held back too much by the Ryzen 5 2400G for it to make sense in here.
Just to set a baseline, with everything set at the defaults, I’ve set Furmark running for 5 minutes where it’s topping out at a roasty 97 degrees C according to its own measurements.
I’ve got Hardware Info running alongside the 3DMark Steel Nomad Light stress test. It’s saying it’s running at a more sensible 84-ish degrees C, although it’s doing that by only drawing 30-ish watts, not the 50W it should be, and as a result the effective clock speed is around 1270 MHz, so it’s losing around 100MHz of its supposed base clock speed.
Unsurprisingly, it gets a failing grade of 88.7% stability, which I think is the worst score of the loops divided by the best. I’ve also re-ran the Fire Strike and Time Spy benchmarks to see if we can improve on them later. It’s getting 5048 and 1672 points respectively.
Before taking the obvious step of slapping additional fans in here, let’s see if we can’t work with what we have. Currently the only fans in the system are this one sat on top of the CPU cooler, which is doing a solid job of keeping the relatively low powered Ryzen 5 2400G under control, and this tiny fan on the tiny heatsink of the P1000, which isn’t having quite the same success.
Who could blame it? The CPU’s putting about 65W of heat out at maximum load, and the GPU should be putting out nearly the same at 50W and is significantly smaller. Let’s find out if the CPU cooler can help share some of the load.
The fan control of the CPU cooler is, naturally enough, normally tied to the CPU’s temperature. The hotter it gets, the faster it spins, and the more it cools. However, using the excellent and free Fan Control utility, we can also tie its behaviour to the graphics card temperature, so if either start to heat up on load, the CPU cooler will ramp up.
The idea here is that as it’s a downdraft cooler, it will move more air about the case which should help the little GPU fan, which will already be running flat out but can also be tweaked in the Fan Control utility if you want. Now, I don’t expect this to work but seeing as it’s free why not try it, and run the same benchmarks.
My phone tells me that with both fans maxed out it’s hitting 60 dBA, and while that’s louder than I’d like it’s not disgusting. While Furmark is giving me pretty much exactly the same readings as before, surprisingly there’s movement in 3DMark.
Running the stress test, we see it’s now holding at 82 deg C, so that’s about 2 degrees cooler, and that’s allowing it to run at higher effective clocks of around 1430 MHz, which is higher than the standard clock speed. This means it’s actually holding a boost clock. This is stunning. I am stunned. The stress test result was 93.8%, still not a pass but a solid 5% improvement just from tweaking a few settings.
Running the benchmarks again, we’re now getting 1747 points in Time Spy, about a 4.5% improvement, and in Fire Strike we now get 5088, which is technically a percent better but that’s margin of error stuff.
Well, that’s actually much better than I was expecting, but maybe not game changing. Let’s go back to plan A, and add another fan in to the mix. The obvious problem is where to put it, as there’s no mounting points to add one, and precious little space.
I think there’s just enough space to squeeze this small 40mm Noctua fan in here at one end of the graphics card, and that can blow across the length of the card and exhaust out the back, hence the removal the backplane of the mini DisplayPort connections. I’ll mount it through the magic of blue tack, until we figure out if this is worth making more permanent.
We can use Fan Control to give us whatever control curve we want, here I’m going with a pretty straightforward straight line one that will run at the minimum fan speed when the graphics card is at 40°C or below, and proportionally increasing to full speed when it reaches 70°C and higher, although in practice this card pretty much maxes out its temperature under any load of more than a few seconds. Seeing as it’s seemingly making a marginal difference, let’s leave the CPU cooler linking scheme as well.
We’ll execute the same tests as before, and now we’re getting – drumroll please – basically the same 82 deg C. Boo. However, I think the effective clocks are higher, frequently getting over 1500 MHz, so, I suppose, yay?
It’s finishing with an improvement of nearly 3% for a 96.7% result, and given that a passing grade is 97%, I’m willing to give it 0.3% of leeway and say we’ve pretty much succeeded here.
Again re-running Fire Strike we get a fractionally better 5106 points, and Time Spy gives us 1746, one point lower but again, margin of error stuff here.
Let’s re-run the same test suite from the video a few weeks ago, and see where we have landed. Turns out that Fire Strike result was 5107, to all intents and purposes exactly the same as we get here, and Time Spy got 1732 points, so we’re now a whole 1% better. Which is weird, seeing as we’ve seen improvements from a lower baseline here. Well, let’s not get dispirited and see what’s happening in the rest of the results.
The Superposition benchmark at 1080p High settings last time around got 2701 points, now we’re at 2898, a 7% improvement.
Shadow of the Tomb Raider’s benchmark at 1080p High settings is now giving us 28 average fps, that’s 8% better than the previous 26 fps result.
Counter-Strike 2 next, with the FPS Heaven Benchmark at 1080p Very High settings sees us get 48 fps an improvement of 0% over the 48 fps last time. Marginally better news at Competitive settings, gaining 2 fps to 125 average fps. Technically 2% better, but that’s margin of error stuff.
The dark world of Unreal Engine 5 next, where Black Myth Wukong’s benchmark at 1080p High Settings, with no framegen or upscaling sees us jump 67%. Don’t get too excited, that just means going from 6 to 10 fps. Dropping to 720p Low, with framegen, and 66% FSR upscaling gets a more reasonable 13% boost from 47 to 53 fps.
The other stretch goal, the Marvel Rivals benchmark at 1080p Native resolution, Medium settings now gets 14 fps, from the last result of 8 fps, again, a not particularly useful but impressive sounding jump of 33%. Dropping to 720p Low settings, with Performance FSR and a smidgen of framegen sees us get 62 fps, up 15% from the last result of 54 fps.
Forza Horizon 5’s benchmark at 1080p Medium settings see us get 39 fps, up 5% from the last result of 37 fps.
Horizon Zero Dawn’s benchmark at 1080p Favour Performance settings gets us an extra 2 fps, up 6% to 37fps from 35 fps.
Given that it’s a CPU test, it’s to be expected that there’s no improvement at all in the Civilization VI Gathering Storm AI Benchmark at 1080p, both runs getting a late game turn time of 63 seconds.
Total War: Warhammer III’s Mirrors of Madness benchmark at 1080p Ultra settings remains a bit of a slideshow, in fact it’s giving a reduction from 12 fps to 9.4 fps, a 22% drop but at these low numbers I’d not place much stock in it. Normal service is resumed at 720p low settings, where it improves 5% from 52 fps to 54.4 fps.
I think our usual Baldur’s Gate 3 walkabout in Lower City need a faster processor as well as a faster GPU, certainly at 1080p Ultra settings where it’s now getting 12.6 fps, compared to last time’s 14.8 fps, an 15% drop. More realistically, dropping to 720p Low settings has us move from 32.7 fps to 40.2 fps, a difference of 23%
We’ll finish off with Cyberpunk 2077, where the in-build benchmark at 1080p Ultra now hits 15.2 fps, up 13% from the 13.5 fps of the last run. Dropping to 720p Low settings has us at 40.7 fps, that’s 4% up from last time’s 39 fps.
A bit of real world performance with the run around Kubuki, again at 720p Low settings but with high crowd density sees us get 31.6 fps, that’s functionally identical to the last run of 32.6 fps.
So, what’s all that mean? Well, in the results I think are actually meaningful there’s a roughly 5% improvement, which isn’t quite game changing, but it’s clearly better than not being improved, so it would be churlish to complain. I can’t remember my statistics courses well enough to work out if that’s statistically significant, but it feels like it is, and I always trust my gut. I have so much of it.
More important than the peak burst performance is the much better sustained clock speeds, as seen by the improvement in the stress test result. This means a more consistent and predictable experience, so I think it’s been worthwhile doing it. Now I need to work out a way to more permanently mount the fan. Our work is never done.
I think we’ve now pursued this idea as far as it can go – there are similarly small fans that move more air that are used in server environments, but at a noise level I do not want to countenance. Maybe an extract fan on the top of the case might help, but that would look a bit too industrial, as would drilling a bunch of air holes in the side of the case. Worth trying if this was the only case we had available, but not something I’m eager to try.
Still, if you have any questions or further suggestions please leave a comment down below, and if you enjoyed this videotronic missive then consider subscribing. Until next time, take care of yourself, and each other.