Page 1:Modifying a GeForce GTX 1050 Ti for Passive Cooling
Page 2:GeForce GTX 1050 or GTX 1050 Ti?
Page 3:Installing An Aftermarket Cooler
Page 4:The Right Case & Placement
Page 5:Clock Rates, Voltages & Temperatures
Page 6:Push it to the Limit!
Page 7:Infrared Temperature Measurement
Page 8:Final Thoughts
Modifying a GeForce GTX 1050 Ti for Passive Cooling
Three years ago, we published Passively Cooling Nvidia's GeForce GTX 750 Ti...With An AMD Sink. The result was, given the performance metrics of the time, a successful operation to turn first-gen Maxwell into a silent gamer by putting our trust in the company's ability to monitor temperature and scale frequency/voltage to obey a defined target.
But now GeForce GTX 1050 and 1050 Ti are available and we have a new challenge to face. After all, GTX 750 Ti was a 60W card, and the two 1050s are rated at 75W. We even went so far as to dig out our 2013 setup up with its modified cooler to compare the previous generation, plus GTX 650, in a passively-cooled shoot-out.
At a basic level, this is a battle of architectures: Kepler versus Maxwell versus Pascal. But it'll be interesting to see how each generation copes with the build's thermal limitations, and measure how far hardware improvements have come over the past three years.
MORE: Best Graphics Cards
MORE: All Graphics Content
GeForce GTX 1050 or GTX 1050 Ti?
Although Nvidia rates the GeForce GTX 1050 and 1050 Ti for 75W, you might imagine that the 1050 would be better-suited to passive cooling. However, the 1050 Ti's performance advantage, helped in part by twice as much on-board memory, compels us to pick the higher-end model. A maximum of 2GB just isn't enough these days.
MSI's fairly simple GeForce GTX 1050 Ti OC stands in as our test subject. It has no separate power connector and features some extra mounting holes that we'll discuss shortly. Using a power target of 80%, the hardware never exceeds an average of 52W, and it often draws significantly less. If you're truly worried about heat, you can dial the power target back even further to 60%. The performance loss imposed by a more conservative ceiling isn't as bad as the percentage suggests.
|Power Target||Metro Last Light, 1080p||FurMark Full-Screen|
To recap so far:
- The GTX 1050’s 2GB of on-board memory is not especially future-proof.
- The card's power draw can be tuned easily through the power target setting.
- The simplest 1050 Ti will suffice for this project, and it requires no overclock.
Installing An Aftermarket Cooler
Let's get back to the 1050 Ti’s previously-mentioned mounting holes. If you read our story about the GeForce GTX 750 Ti, you know that we had to modify the distance between screws by several millimeters in order to find a compatible cooler. Now we're in the same regrettable situation because the 1050 Ti's original cooler has a similar grid of drilled holes measuring 48mm. Finding a third-party heat sink with that measurement probably isn't going to happen.
But wait! The MSI card presents a unique opportunity in that it sports a grid of four, unused holes measuring 58mm. This layout proves extremely convenient, since practically every cooler conforms to it. As a result, new life is breathed into the Arctic Accelero III that we planted on a Radeon R9 290 three years ago. If you want to replicate our results using another cooler, just bear in mind that there should be at least 15mm of space between the bottom of the heat sink and the fins to prevent contact between components like coils and capacitors.
The image below is notable because half of the cooler body is hanging off MSI's card. This setup allows for optimal airflow through the process of convection. It's also one of the reasons why passively-cooled cards should not be installed in a vertical configuration; heated air rises, after all. A graphics card pointed up and down would cause that air to spend more time in contact with the fins, creating inefficiency.
The short MSI card and longer cooler paint a beautiful picture together. All that's left now is installation, after which we can put these two components to work.
- The 48mm distance between screw holes presents an issue when it comes to compatible aftermarket coolers.
- There are several cards with unused screw holes in a 58mm grid configuration, one of which is the MSI board we're testing today.
- The longer the cooler, the more it will overhang the card's PCB, increasing airflow and cooling efficiency.
The Right Case & Placement
There needs to be at least 34cm of horizontal space available in whichever case you choose. After test-fitting the card’s positioning during installation, it is clear that, due to the cooler's extreme length, any chassis with a motherboard mounted horizontally is going to be impractical. Just to be sure we're covering our bases, we'll measure how the modified card/cooler behaves with and without plenty of airflow inside the case.
In our first run-through, we installed the card in a workstation. Its case fans were all disabled, except for two 240mm radiator fans that spin at 300 RPM. In addition to a slight under-pressure condition caused by those fans pulling air upward, the card enjoys plenty of room inside the chassis, which is ideal for the passive sink's convection-based heat dissipation.
For the second test, we pulled out our build from 2013, which employs passive cooling exclusively, including the CPU. We also armed it with a temperature-controlled fan setup as backup. Our GeForce GTX 650 project needed the extra cooling on and off. Can the GTX 1050 do without it, and deliver better performance in the process? We're excited to quantify the improvement in efficiency over the past three years.
Compared to the GTX 650 and GTX 1050 Ti, our passively-cooled 750 Ti had to make do with an aftermarket Sapphire heat sink that we modified to fit since the card's screw holes wouldn't allow for anything larger.
Mounting the card/cooler combination horizontally allows for optimal cooling via convection. Vertical mounting would be less efficient.
A large amount of free space in the case is advantageous for any passive cooling experiment.
The card/cooler’s own convection process should not be impacted by other hardware in the case, if possible.
Clock Rates, Voltages & Temperatures
Measurements collected in a passively-cooled rig require significantly more test time than in actively-cooled systems. The passively-cooled system can take up to an hour for all of the components inside to reach their final operating temperature. But after 57 minutes, the GPU temperature we recorded drops from our forced limit of 83°C down to 77°C, and then continues to hum along right there. Even after two hours, our measurements show the card posting a constant 77°C, which is astounding.
The original card/cooler combination can function in passive systems with a convincing 63°C result, while the extreme-length cooler in the cavernous workstation with two slow-spinning radiator fans achieves a downright chilly 54°C.
If you're wondering why the passively-cooled, fanless system would suddenly drop in temperature after two hours of running, worry not. We performed more tests to get to the bottom of this.
Our first order of business was recording the GPU frequency, which revealed even more useful information. After running for a while at the factory temperature limit of 83°C, GPU Boost was no longer working to push clock rates higher. The processor leveled off at 342 MHz and stayed there.
The same symptoms are observable in the voltage, which switches to a constant 0.812V. This value, similar to the static clock rate, appears to be some type of fail-safe. Power consumption even drops under 40W. Does this emergency response affect gaming performance, though?
FurMark exposes similar results, though the various limits for power consumption and temperature are reached significantly earlier due to that utility's intensive load. Fortunately, the passively-cooled GPU runs smoothly at maximum load without ever approaching dangerous temperatures.
There's a downside, of course. Cooling a GPU passively and maintaining reasonable temperatures necessitates dialing back clock rates, and ultimately, performance. We certainly don't want to circumvent the safeguards that limit frequency, voltage, and power consumption. They're in place for a reason, and working around them is a recipe for damaged hardware.
Our modified cooling solution provides better results than the original, even with limited airflow.
A completely passive configuration is entirely possible, though it creates performance limitations.
- Clock rate and voltage are automatically lowered to protect the card.
Push it to the Limit!
All three games we tested (Metro: Last Light in 1440p, GTA V and The Witcher 3 in 1080p) take a 21% performance hit compared to MSI's stock configuration. At the same time, though, power consumption drops below 60% of the original measurements. Taking into account the 4% slow-down imposed by our 80% power target, the potential loss of performance due to passive cooling doesn’t seem as disheartening as before.
Let's have a look at an example where the GPU negatively affects our benchmark sequence's performance. These limitations are not outside the norm by any means, and they occur routinely in demanding games like the ones we're using to test. We see again that, as of the moment when the GPU’s emergency measures kick in, something else appears to be bottlenecking the system’s performance.
It cannot be our power limit, since there's constant intervention right up until minute 53, when this appears to evaporate.
In the end, this has to be caused by the card's consistently high temperature, and it persists even after the fail-safe brings temperatures down below our preset limit.
Performance degrades more slowly than power consumption.
Performance is reduced up to 21% as a result of preset limits.
- There is a performance loss of up to 4% in the benchmarks due to our 80% power target.
Infrared Temperature Measurement
In order to obtain a more detailed understanding of exactly how our modifications affect performance, we measured temperature changes through an infrared camera. Up until now, we were only measuring the GPU's thermal situation. But a graphics card contains many other components that are sensitive to overheating, and may critically influence performance.
At idle, low clock rates and voltages yield a nice, cool board. Temperatures across the PCB are virtually undetectable by the thermal camera. Only the VRMs register a bit of warmth.
If we put the card into a case with minimal airflow, the passive cooler does its best to keep Nvidia's GP107 cool. Even next to MSI's stock configuration, it fares well enough, too. The card glows orange in our infrared image, but these temperatures are typical of any graphics card cranking through a game like Metro: Last Light.
It's only when the card is installed into a completely passive PC and allowed to run for an hour that it saturates from edge to edge with heat. In spite of the thermal challenges this environment presents, the memory still runs cool enough to prevent any heat-related damage.
We have to conclude that this passive cooling mod is quite successful from a purely thermal standpoint, even if we had to battle a little bit with its placement.
Our small battery of tests shows again that completely passive cooling configurations in the 60 to 75W category are only possible after a handful of compromises. Choose one of two options: either install a set of low-RPM fans to create a little airflow inside your case, or allow the GPU to hit its temperature limit in a passively-cooled enclosure and live with the reduced performance. Leaning on a fail-safe isn't really ideal though, so we don't love that idea.
Realistically, the upper limit for completely fanless operation remains around 40W, where almost all of the card's power is dissipated as waste heat. In that situation, the best option is to manually set a power target of 50%. Otherwise you're going to see protection mechanisms kick in and force the power limit anyway.
Given the circumstances presented, we must recommend at least some air circulation in your case for a project like this one. With that in mind, when we think back to our original passively-cooled GeForce GTX 650, not much has changed in the past three years. After all, we aren't picking cards that use dramatically less power. What they do, however, is deliver significantly better frame rates, improving the performance to power ratio versus three years ago. The frustration of coming up short on a completely passive solution is dampened somewhat by the GeForce GTX 1050 Ti's excellent response to a humble, low-RPM fan.