Cooling Noise and Power

For my last few tests, rather than focusing on in game performance, I like to check out other aspects of video card performance. These are also the most important ways to differentiate the performance between cards that have the same GPU. To start things off I took a look at power usage.

For this, our test setup utilizes the Nvidia-designed PCat v2 along with cables to handle both traditional 6 or 8-pin connections as well as 12VHPWR. The PCat also utilizes a PCIe adapter to measure any power going to the card through the PCIe slot so we can measure the video card wattage exclusively, not the entire system as we have done in the past. I test with a mix of applications to get both in game, synthetic benchmarks, and other workloads like GeekbenchAI and AIDA64. Then everything is averaged together for our result. I also have the individual results for this specific card and I document the peak wattage result. I do have to note that during my testing of the RTX 5090, I ran into issues with a few tests, especially with OpenCL which is what AIDA64 uses to do their stress tests. I ended up having to replicate our test results using OCCT. To get as close as possible I used the 3D Adaptive stress test in OCCT and using the 4090 I stress tested it and played around with settings to get to a setting where OCCT replicated the same power usage and card temperatures. In the future, once that issue is fixed I will be redoing our 5090 tests with AIDA64, but this issue was confirmed to be reproducible by Nvidia the day before launch and I didn’t have time to retest the whole list of cards tested. That said the OCCT result wasn’t the peak power usage here, so any difference between the tests will only change the average result slightly (but that same testing affects our temperature tests and power efficiency graphs. The RTX 5090 Founders Edition peaked at 648 watts and averaged 613, this is without a doubt the highest power draw I’ve seen to date and is 136 watts more than the RTX 4090. At one point in my testing, I had issues with 3Dmark Fire Strike Ultra crashing and as it turned out, it was because our motherboard needed the supplemental power to the PCIe lanes plugged in. So going in, you are going to need more power to run a 5090 than cards in the past. The average across all of the tests was 613 watts and the results ranged from 550 watts in FarCry 6 up to 648 in Time Spy Extreme.

graph34

graph35

With having exact peak wattage numbers when running Time Spy Extreme I was also able to put together a graph showing the total score for each watt that a card draws which gives us an interesting look at overall power efficiency in the popular and demanding benchmark. Even with the 648 watts pulled, the overall performance of the RTX 5090 still put it right at the top of our efficiency chart and with a good margin between it and the RTX 4090 which was below it. 

graph36

My next round of tests were looking at noise levels. These are especially important to me because I can’t stand to listen to my PC whirling. Especially when I’m not in game and other applications are using the GPU. For my testing, though I first tested with the fan cranked up to 100% to get an idea of how loud it can get, then again at 50% to get an idea of its range. The RTX 5090 Founders Edition with its new cooler design was a little noisier than the other Founders Edition cards at 50% fan speed but interestingly at 10)% fan speed it dropped back down into the middle of the chart. The reason for it being lower in the 100% chart can be seen in the fan RPM chart, it is down in the bottom ¼, they have the fans running slow but not slower than the previous generation 4090 and 3090 Ti. To be fair though, this is a 2 slot card and is using a lot more power I’m surprised they weren’t running the fans even higher.

graph37

graph38

graph39

I also take a look at noise performance while under load. For that when running AIDA64’s stress test I wait until the temperature of the card has leveled off and then measure how loud things are when the card is at its worst-case scenario with the stock fan profile. Here the RTX 5090 Founders Edition was once again up near the top like it was in the 50% fan speed chart. That makes sense though given that when under load the fans were running at 46%. That tells us even before getting into the temperatures that the card isn’t working too hard to keep things cool. You can see for example the last generation of AMD reference cards that were also two-slot were pushing their fans a lot harder, the 5090 is tied with the RTX 4090 here.

graph40

graph41

To finish up my testing I of course had to check out the cooling performance. To do this I ran two different tests. I used AIDA64’s Stress Test run for a half-hour each to warm things up (on everything except the 5090 which was tested on a similarly matched OCCT workload). Then I documented what temperature the GPU leveled out at with the stock fan profile and then again with the fans cranked up to 100%. With the stock profile, the RTX 5090 Founders Edition came in at 72c. For comparison, the RTX 4090 FE was at 77c in the same test. For GPU memory it was running at 83c which is warm for sure. Then with the fans cranked up, the RTX 5090 Founders Edition ran at 58c, again noticeably cooler than the 4090 in the same test and below the 3090 Ti and 3080 as well. The memory cooled down to 68c. The delta between the two results was 14c which is on the high side showing that if you want to play with the fan profile and don’t mind a little noise there is a lot of cooling headroom to be had especially if you want to overclock.

graph42

graph43

graph44

graph45

While running the stock fan profile testing I also took the time to get a few thermal images so we could see what is going on. The new thinner cooler design is able to pack in large fans by keeping the large length and height of past flagship Founders Edition cards and you can see just how much cooler things are running on the front and back side of the card where the fans are running. In the center with the covered sections, things are warmer with the power connection and then the rear triangle on the back side of the card being the two hottest spots. Even then on that back side picture, the thermal camera picks up the hotspot as being just below the card where there is less airflow between it and the M.2 heatsink. I’m sure larger aftermarket designs are going to improve on this, but I am impressed that Nvidia was able to design a card that cools this well while being a dual-slot card. I still wish it was a normal card height, but it's good to know they are paying attention to the SFF guys as well.

thermals 1

thermals 2

thermals 3

 

Log in to comment

We have 1260 guests and no members online

supportus