Alongside our previous tests of premium 92 mm Noctua fans, we’re adding something for value-focused users. For the more affordable Prime-series card, these are a better match. We’ll measure how much quieter the card gets after swapping the stock fans for budget 92 mm Arctic P9 PWM PST models with sleeve bearings, and compare the noise to the Arctic P9 PWM PST CO variant with dual ball bearings for continuous operation.
Before we focus on the results with the Noctua fans, let’s look at the behavior of the card with the original fans. In the first graph, you can see how the card performed in terms of noise with the supplied fans. The card’s control allows you to set their output in the range of 30–100%. This corresponds to an operating range of 700–3200 rpm. Because one of the fans was clicking, the noise level of the cooler was 32.1 dBA even at the low 700 rpm. This increased the noise of the card across the entire speed range.
The next graph shows noise levels with the trio of Noctua NF-A9 fans. When connected to the control panel, they start at 10% PWM and run at 98–100% PWM at maximum.
At lower speeds, the noise curve is flat. The reason is that the fan noise is below the measuring range of the sound level meter (30 dBA). At low speeds, the fans are almost inaudible and the intrinsic noise of the sound level meter is higher than the noise of the fans
How the operating characteristics of the card change will be measured across the entire range of fan performance. For time reasons, I will use short loops of the Cyberpunk benchmark with the RT Medium profile and a resolution of 2560 × 1440 pixels with DLAA.
Before the whole batch, one warm-up loop runs with three passes and fan output set to 40%. This is not visible in the graphs below; they start with the first measurement with fans set to 100% output. Fan performance then gradually decreases in 5% steps from 100 to the lower limit of the measured range. With the card using factory fans, this ends at 30%, which corresponds to the minimum adjustable value. With cards with replaced fans, it ends at 20% output. Testing at 15 and 10% usually does not make sense, because at low fan speeds and GPU temperatures above 85 °C, regulation already starts to significantly throttle GPU power and with it clocks and performance, and noise can no longer be measured with the sound level meter used.
With each setting, three test loops are run. Between individual test runs, delays are only a few seconds, so the system remains warmed up during testing. Measured values will be read from the third loop, exceptionally from another one if there is some problem with the last measurement.
Asus Prime GeForce RTX 5070 (stock, 3× Axial-tech fan)
In this chapter, we will go through how the card behaved across the entire fan performance range. First, these will be simple graphs showing the entire test run, where everything is visible together. Specific measured values for each setting can be found in the interactive graphs in the following chapter.
At the lower limit of the tested range, that is at 30% fan output, the maximum GPU temperature was 85.2 °C and the average was 84.0 °C. At maximum speeds, the average temperature dropped to 59.2 °C.
The next graph shows fan speeds in the light blue data series. The fans were connected directly to the card, but their speeds were set manually using the Fan Control application.
At the lower limit, at 30% output, they start at around 700 rpm; at 100% output they run at 3200 rpm.
The purple curve in the graph shows the set fan output. The light blue values correspond to the current fan speed on the Fan 1 connector.
At minimum speeds, when the GPU temperature goes over 85 °C, power throttling of the chip occurs, clocks are reduced, and with that performance drops.
The next graph shows temperatures (in red) and total CPU utilization (in blue). As fan output on the card decreases, airflow in the case slows down, and CPU temperature rises slightly.
Performance differences do not change dramatically. With a cooler chip, clocks rise slightly and with them performance, but the difference increases only by tenths of frames per second.
This is more visible on GPU clocks than on performance. Toward the end, at low fan speeds, the card began to overheat; here you can see more significant drops in clock speeds. Between 30% and 100% fan output, the difference is roughly 100 MHz.
For reference, we also have graphs of power consumption for the entire system and for the card itself, according to what monitoring reports. In the first run, when chip temperatures were already too high, it is visible that GPU Boost also throttled the card’s performance. In subsequent runs, the card is clearly operating at the limits of its power budget.
⠀
I’m very grateful for this test. I always lacked motivation to check how much better my temperatures actually are after a deshroud and if using 92 mm fans might be a better option. I just always went straight for 120 mm fans, even though they were sticking out, because of the “bigger = better” assumption and both temperatures and noise were always better than with stock fans. Recently, however, I finally managed to gather the determination to roughly compare a deshroud with 2× Phanteks T30 versus 3× Arctic F9, and the results align with what was tested here. Is this the end of this saga/series, or are there still some other fan models lined up for testing?
note: the test with 2 x 140 mm fan was performed on a different graphics card. 😉
May I be curious and ask how the comparison went for you: 3 x 92 P9 vs 2 x 120 T30?
At similar room temperatures, 2x T30 (mounted in a printed shroud where fans was fully enclosed inside) reached a maximum of 85.2*C and an average of 83.3 °C with a peak power draw of 306.3 W, while 3x F9 reached 84.2*C max and 81.3*C average with a peak power draw of 312.7 W.
Test was ran by loading a CP2077 save at maximum graphic settings with path tracing enabled, without moving the character or camera, and letting it run like that for nearly an hour. The results were collected via HWiNFO64, which was reset immediately before pressing the key to load the game. Fans speeds were fixed at comparable noise levels, measured using a smartphone app. Iirc ~1250 rpm for T30 and ~1500 rpm for F9.
However, the result is not precise because I did not control temperature in the room. I only measured it with a simple rectangular meter with a humidity readout, which accuracy is questionable. The same applies to the consistency and accuracy of in-room sound level measurements taken with a smartphone app. That is why I am particularly interested in whether these results were just coincidence caused by external factors, or whether they reflect a real difference.
nice work, thank you very much for your input 👍 😉
I’d like to add my thanks as well—very nice documentation. Cyberpunk 2077 will (probably) be a serious stress test that really pushes the graphics card almost to its very limits. 🙂
Exactly. CP2077 was a good non synthetic stress for card as it never came close to 60 FPS and iirc never crossed 50 FPS mark and always was 99% loaded. Thanks for kind words and you’re welcome 🙂
Lower temperatures don’t necessarily mean more efficient cooling. Control over GPU clock speeds is also important—we address this in CPU cooler testing as well. Power draw in watts is, of course, important too, but it doesn’t allow for a clear-cut conclusion that “more is better.” Higher power consumption can often be driven by greater thermal losses precisely because cooling efficiency is lower. That said, I’m referring here to experience with processors running on motherboards. 🙂
I definitely agree! When I saw how long my comment become, I deliberately limited it to reporting only the power draw, assuming it would give a sufficient picture under assumption that higher power draw translates into higher clocks. As you rightly pointed however higher temperatures also mean lower power efficiency, which in turn leads to higher power consumption. This is very easly observable with CPUs. Der8auer once did a test on this and visualized it very well.
Thats said, to complete the picture, here are the recorded clock results for those configurations:
2x T30: core 3135 MHz max, 2833.5 MHz average; Memory temps: 76*C max, 74,1*C average.
3x F9: core 3202 MHz max, 3015.2 MHz average; Memory temps: 76*C max, 73,8*C average.
In both setups memory clocks were stable at 2833.5 MHz through whole time. I should also add that the GPU is a 5070 Ti with OC set to +400 on core and +3000 on memory in MSI Afterburner.
Now… the winner…in your case is incontrovertibly clear 😉
The fact that higher power consumption does not automatically mean higher CPU clock speeds can also be seen in our cooler tests. Beyond a certain point, as temperatures increase, power consumption may continue to rise while CPU core clock speeds sometimes start to drop. 🙂
As far as 92 mm fans go, that’s probably all for now—but tests with Arctic P12 Pro fans will likely appear soon. 🙂
…and maybe, if I’ve motivated Adam enough 🫣😀😉,
…a test with 2 140mm fans will appear later
Coolermaster has a custom ASUS built series of 5080 cards with a shroud specifically designed to enable swapping out the fans with standard 120mms with accommodations for standard thickness and other marques of fan. I wish other companies, like say, Sapphire would jump on this train.
I’ve seen that solution from CM and I really regret that it’s only a prototype/visual concept with no plans to bring it to market but that would be absolutely fantastic! Personally, I could even live with the same pricing and no stock fans included, as long as I could install my own, let’s say, 10-32 mm thickness range if only that became a standard. Ideally, though, the best solution would be something like with AIOs: you have the option to mount your own fans, but there are also “off-the-shelf” fans included out of the box.
Cooler Master likely has more motivation than Sapphire to produce graphics cards like this. They have their own fans that can be used, whereas Sapphire doesn’t really focus on offering standalone fans in its portfolio. 🙂
End user replacable fans *are* a feature that’s proliferating on modern radeons IIRC, so it may be less of a leap – especially as Sapphire has marketed CPU coolers outside of the global north before. Getting into the chassis fan market isn’t that improbable for them I’d think, if they want to diversify. IIRC, I’ve heard good things said about their cooler fan designs too, so they may be able to field some credible offerings too, with that experience.
Perhaps the possibility of entering the global CPU cooler market is closer than it seems
https://www.sapphiretech.com/en/consumer/nitro-s360a-aio-cpu-cooler
This is not new one, but it is on the global EN web
Review (TPU) from 2022
https://www.techpowerup.com/review/sapphire-nitro-s240-a-aio-liquid-cpu-cooler/
Hopefully, Sapphire AIO coolers will eventually make their way to our market as well, similar to their motherboards. 🙂
Yes, Sapphire has been involved in cooling solutions—at least in the past—when it comes to CPU coolers. However, the Vapor-X model apparently didn’t gain much traction on the market, and that may be one of the reasons why their motivation to continue in this segment declined. 🙂
The removable fans on Nitro+ graphics card coolers are excellent, and personally, I’m really rooting for Sapphire to push things to the next level. 🙂
Hi hwcooling,
What a gooood job that you made here !
I think that if you want to have a better cooling, one of the good update is to add extra rear fan (without the backplate) to help the right fan to extract hot air !
I think that if you try that, the cooling will be better !
(Sorry for my baaaddd english 😬)
well, existing examples do not support this assumption… 😉
https://www.hwcooling.net/en/aorus-geforce-rtx-5080-master-whats-the-fourth-fans-benefit/4/
the thickness of the heatsink and its density do not require a pull-push setup…
moreover, the proximity of two fans that would spin at the same speed can lead to so-called beat frequency (When two waves (like sound waves) with close frequencies combine, they constructively and destructively interfere.)
We addressed the benefits of a fourth fan on the RTX 5080 Master graphics card cooler here, as Bufo also points out. 🙂
Any chance of you releasing the stl files? Or, preferably step files so we can make changes. There are a few Asus prime shrouds out there but I really think the way you separated the legs as separate parts is clever. And looks good too.
Thx.
I plan to make it available for download here and on Printables in two weeks or so—would that be okay? I need to make a few adjustments: some screw posts on the legs don’t line up perfectly with the PCB holes—there’s a 0.5–1 mm offset. I’ll realign the screw posts to match the PCB holes, tweak the corner pockets for the silicone pads to improve printability, and add a version for two 120 mm fans.
And yes, I can provide STEP files. The model comes from Rhino 3D and will be a closed boundary representation (BREP)—a joined polysurface. It should import as a solid, but it may not behave the same as STEP files from typical parametric CAD tools.
Thanks for responding. I’d take whatever you have without you having to put more work into it.
Yeah, the shroud design looks good and it will probably work well too. There are no gaps between the fan and the shroud structure where static pressure could escape. That means airflow through the radiator should be high—though of course it still depends on which fan model is used and at what speed. 🙂
The Thermalright TL-P9 work pretty quiet on my 5070Ti Prime. They have 2200 rpm as maximum rpm and usally stay below 2000 under load through the MODDIY Adapter, which is still tolerable when gaming with desktop speakers.
I’m looking for a solution in pull configuration though, as i have the GPU in upside down orientation in an inverted sff case (McPrue Apollo SE).
Any tips on what to use there? Are Phanteks T30 worth a try?
If the 120 mm T30 fans fit, then using them as a pair would definitely be a good choice. That also applies to Noctua NF-A12x25 (G2)—or really any fans that rank at the top of our measurements on radiators. I wouldn’t be afraid of Arctic P12 Pro either, even though there are frequent user reports saying they get noisier in pull configuration. That may be true on case grilles, but it doesn’t necessarily apply to radiators. It might—or it might not. Either way, it would definitely be worth trying. Aerodynamically, the P12 Pro fans are very well designed, and given their low price point… 🙂
Based on your experience, what fan size would you recommend for the Asus TUF 5070? It has the same dimensions as Asus TUF 5070 Ti (330x140mm), which you deshrouded already. I’m leaning towards two 120mm fans (perhaps the low-speed version of the Noctua NF-A12x25 G2), but I worry that I will be far off from covering the length.
Based on what I’ve tested: if I were going with two fans, I’d probably choose the NF-A14x25 G2; if I wanted to save some money, I’d pick the Arctic P14 Pro PST instead.
https://www.hwcooling.net/en/tuf-rtx-5070-ti-deshroud-2x-140-mm-noctua-nf-a14x25-g2/7/
For a 120 mm setup, I’d install three NF-A12x25 fans or three Arctic P12 Pro fans.
In fact, it would be also interesting to compare a deshroud with two and with three 120mm fans, so it’s clear how much the third fan helps.
Thanks for the comment—I don’t think I could have replied to Taisho any better myself. 🙂
Yes, adding a third 120 mm fan could improve cooling efficiency, but compatibility then becomes an issue. The graphics card would become very long and start conflicting with many cases. With two fans, compatibility is already above average and much easier to maintain. 🙂