A little while back we were looking at this tiny little console size computer, and crammed a single slot, half height graphics card into a very tight space. Even with special measures, that card remained quite toasty so we couldn’t get the full performance from it. I’m not completely happy with that as a resolution, so let’s revisit that today.
Welcome to Esk Computers, I’m Scott and today on the desk of Esk, let’s unleash the true power of this tiny graphics card by exceedingly impractical means.
I’d recently swapped out this 240mm all in one Thermaltake cooler from my 9800X3D system, and, well, I’m sure you can see where this is going.
Removing the stock cooling solution we see the mighty beast that we must tame, a GP107 die also found in the 1050 and 1050Ti, albeit with a trimmed shader count of 512 vs the full fat 768 units, and it ought to be running at 1354 MHz, with boost clocks of 1392 MHz, and the 4GB of GDDR5 memory is running at 1502 MHz.
As the fan which presumably gives at least some airflow over the circuit board as well as the heatsink is part of the assembly we’ll be removing, we’ll see what we can do about getting some airflow over the power delivery components later.
For some shortsighted reason Thermaltake doesn’t ship this with mounting hardware for a near decade old small form factor GPU that’s significantly smaller than the radiator of the cooler alone, so we’ll need to improvise. In scenarios like this, improvising always means cable ties.
There we go, eat your heart out Zotac 5090 ArcticStorm, you surely will have nothing on the untrammelled power of this fully unleashed P1000. It’s a shame there’s no particularly easy way of overclocking this card, or we could have gone for some extreme results, but I don’t feel like flashing the VBIOS with randomly sourced internet files today.
Not only is this cooler bigger than the GPU it’s cooling, it’s also a good percentage of the case’s size. So, we’ll pull this out and set it up testbed style. Oh, and I wanted to see if this stock AMD cooler works as well as the Metalfish unit I had on it, and the answer’s yes, at least on this low powered 2400G, but I couldn’t think of a way to milk a video out of that. It’s running with 32GB of relatively slow 2666MM DDR4 RAM, and a 512GB CUSU NVMe SSD.
Unsurprisingly the first run at this wasn’t entirely successful, not making great contact with the GPU die. Or any contact at all probably, given how quickly it thermal throttled.
Still, nothing a massive clamp can’t fix, just to maintain this very sensible project’s heady mix of practicality and aesthetic appeal.
Our video got us to a point where it still didn’t quite pass a 3Dmark stress test, and seemed to level out at 82°C, so that’s the initial goal I suppose, and we are indeed getting a passing grade of 98.4%, so that’s something.
We seem to be getting constant clocks of 1544 MHz, at least that’s what 3Dmark reports as GPU clock, but I think it’s getting memory clock and GPU clock confused. What it’s showing as the constant 1252 MHz fits better with the datasheet. Albeit 100mhz less than it ought to, if that datasheet was to be believed.
Testing the cooling capacity with Furmark, I’ve let it render this furry doughnut for 10 minutes and it’s topping out at 71 degrees C, with hotspots at 78 so about ten degrees cooler than our previous best, with completely stable clock speeds.
Let’s batter though the same set of benchmarks as last time as see if there’s any difference.
In 3DMark’s Fire Strike we are now getting 5120 points, up a massive 0.27% from the last result of 5106.
The hits keep coming in Time Spy, where the 1752 points we get this time eclipses the last result of 1746 by 0.34%.
Rounding of the synthetics with Superposition, it’s another thumping victory for this S-tier cooling solution, a colossal 0.93% increase from 2898 to 2925.
Okay, I’ll lay off the hyperbole for a bit. Shadow of the Tomb Raider’s benchmark at 1080p High settings gives exactly the same 28 average fps in both tests.
We do actually see a statistically significant improvement in Counter-Strike 2, where at 1080p Very High settings there’s an increase of 8.33%, going from 48 to 52 average fps, and similarly at Competitive, Low settings with Dynamic shadows there’s an 8% increase from 125 to 135 average fps. I’m not ruling out that being due to an optimisation to the engine rather than being driven by temperature.
Inexplicably, disaster strikes in the Black Myth Wukong benchmark tool, where at 1080p High settings without framegen or upscaling, we drop to 7 average fps, down a colossal 30% from the blistering 10 fps we saw last time.
It’s redeemed somewhat by the run at 720p Low settings, with frame generation and 66% FSP upscaling, where the 53 fps is at least consistent with the last run.
There’s a similar story in the Marvel Rivals benchmark, where at 1080p Medium settings, without framegen or upscaling we get 13 fps, technically down 7.14% on the last time where we got 14 fps. Splitting hairs really, but again, dropping the settings to Low and adding ultra performance FSR showed the same 62fps we got last time.
Cyberpunk 2077 at least returns to sanity, albeit hardly massive gains, where the benchmark run at 1080p Ultra improves 2.43% from 15.2 fps to 15.57 average fps. No, you’re clutching at straws.
The same benchmark at 720p Low settings gives us a 1.5% improvement from 40.7 to 41.3 average fps although there’s an interesting result in some actual gameplay where the usual run around Kubiki and cop bothering goes from 31.6 to 38.2, a near 21% improvement.
However, we’re soon back to the familiar pattern of little or no improvement. Forza Horizon 5 at 1080 Medium settings holds steady at 39 average fps, and it’s the same in Horizon Zero Dawn, where at 1080p Favour Performance settings we get 37 average fps in both scenarios.
There’s another technical improvement in Total War: Warhammer III, where the Mirrors of Madness benchmark at 1080p Ultra goes up 3.19%, but when that means going from 9.4 to 9.7 average fps it’s hard to get excited about it.
At 720p Low settings if anything it’s even less exciting, going up 1.1% to 55 from 54.4 average fps.
Lastly, we are going to Baldur’s Gate 3, where there’s technically a regression of 1.2% at 720p Low settings, dropping from 40.2 to 39.7 average fps, but like most of the results today, this is margin of error, run-to-run variance sort of stuff.
There is, however, a substantial jump at 1080p Ultra settings, going from 12.6 to 17.1 average fps, a pretty impressive 35% improvement, although still not really bringing it to playability even by my forgiving standards.
So, there we go, an overall average improvement of a humongous 2.34% and all it takes was a few simple, practical, and sensible steps of ripping the computer apart and assembling this hideous, monstrous fire hazard of a setup that no sane person would have disgrace their desk. It’s almost as if whoever designed the stock cooler thought about what the practical use cases would be and took that into account in the design.
Some would say this has been a pointless video, an impractical exercise that only shows the obvious point that cooler chips can run better, and there’s no point testing cards that only exist to be compact with bulky, overkill cooling solutions. And they’d be right.