Nvidia GeForce RTX 5090 FE review: Next-Level Gaming

Today ends the information embargo on tests of Nvidia’s top-tier GeForce RTX 5090 Founders Edition graphics card. It delivers extreme performance but at the cost of unprecedented power consumption and the highest price yet. Even so, it is expected to attract immense interest from gamers and professionals for its 32 GB of memory and exceptional performance.

A closing summary

The increase in raw gaming performance understandably doesn’t match the marketing graphs from launch, which show a mix of results from different rendering modes with interpolated frames. It can be said that the cards are capable of rendering higher frame rates, but this is not due to higher rasterization performance, but due to technologies like DLSS. Even so, the increase in raw performance is not insignificant.

This will vary mainly depending on how demanding of a rendering the card is running. If it’s a classic rasterization without ray tracing, the performance gains will be lower than for demanding scenes with ray tracing. There are several reasons for this. The main one is that developers use ray tracing for more demanding effects, and you won’t find many new games that are performance intensive without it anymore. To that account, Nvidia and, with the new generation of Radeons, AMD have been beefing up performance especially in the area of ray tracing. On high performance cards, games that don’t use ray tracing often run into limitations on the CPU side and are not fully utilized. When you then compare cards that are not fully loaded, the performance gap and scaling will be smaller. The smaller performance increase applies to lower resolutions as well, but very few people will probably buy such cards to play in full HD or 1440p.

DLSS 4: Two extra frames and similar latency

Intuitively, it might seem that if the card only renders every fourth frame in 4× multi frame generation mode, the response time will be longer than when generating one frame. This is not the case, I want to emphasize that the responsiveness does not significantly degrade with three or four frames generated versus what we know with frame generation from the last generation.

So the same applies as for frame generation with RTX 40 – it’s not suitable for situations where you’re starting from 25 frames per second and interpolating to 80. In such a case, while the image is smooth, it has a noticeable lag behind mouse movement, similar to when you’re playing at 25fps, because when rendering at 25fps, the long response time is natural, and adding more frames between renders won’t improve it.

Therefore, it is a good idea to set the details to achieve higher base frame rates and only interpolate from there to higher frame rates for fast monitors. If the graphics card doesn’t have enough horsepower to play games with that setting, it means either lowering the detail or using classic DLSS in one of the Quality, Balanced or Performance modes along with the frame generation. Once you get to some 40–50fps, it still won’t be for pro gaming or frenetic shooters, but everything else can be played comfortably in terms of responsiveness.

Record-breaking performance, record-breaking power consumption

The power consumption could also be the subject of a separate article. Significant improvements usually come with a new generation of the manufacturing process. That hasn’t happened yet, and so the performance increase largely replicates the increase in card power consumption. In absolute numbers, the wattage is extreme for what we’re used to, but the card is far from always running at the maximum 575 W. Rather, even on the powerful Ryzen 9 9800X3D, you have to make a bit of an effort to get the card to the maximum values and keep it there. It behaves differently when rendering demanding scenes with ray tracing, differently without it, slightly differently with frame generation or upscaling enabled, and differently with adaptive vertical sync active. And a lot depends on how much the processor can assist the card and how many frames the monitor can display in case you don’t want to render “blanks”. In short, having a near 600W card doesn’t mean it will consistently consume around 600W when gaming, rather it will usually be lower.

But even the full 575 W does not mean a dramatic deterioration of the cooling of the entire build or problems with overheating of the card or other components. I’ve only tried it for a short time so far, but the card otherwise didn’t significantly overheat or lose performance even with only the four supplied system fans in the case or even with the fan speeds reduced to 600 rpm.

The case vented the heat away into the room without any problems. Of course, this is then noticeable in the room temperature, in the winter it is not that much of a problem, but in the summer it could get less comfortable. We will hopefully get to the issue of how 600W cards behave with different system fan performance in the articles dedicated to system cooling.

You have to expect higher power consumption than lower-end cards even under no load, but this depends largely on chip size, memory and the number and resolution of the connected monitors. In this respect, the RTX 5090 behaves similarly to high-end cards from previous generations.

The cooler continues to impress

In the context of the card’s power consumption, the performance of the Founders Edition cooler is particularly fascinating. Not to say that it is somehow extremely quiet. But in terms of noise, even at its size, it’s still slightly better than average cards and still significantly better than most previously released reference cards or the cheapest RTX 4080 or RTX 4090 models. But if you get a premium card that’s five centimetres longer and one and a half slots taller, it’ll definitely be quieter. But at the same time, you have to get the heat emitted by the card out of the case somehow, and that too can’t be completely inaudible at such power consumption.

If you don’t have a 360mm liquid cooler on your CPU, you’ll notice more of the CPU cooler fans, which have a fast and intense onset when they’re spinning due to high temperatures of one of the cores or another hotspot. Typically, when running games, the CPU cooler, when using auto-regulation, will be heard first, only after the game and saved positions are loaded does it calm down and the card comes on with slightly less noise. This is especially true for powerful multi-core processors, which are difficult to keep below thermal throttling under heavy load with classic air cooling. However, even the lower-power Ryzen 9 9800X3D gets to high peak temperatures just below the thermal limits, even when the extreme Noctua NH-D15 G2 sits on top of it.

As a result, the cooler is not much noisier than the rest of the build. In short, the RTX 5090 Founders Edition isn’t quiet, but even using your ears you won’t be able to tell that it’s a 575W card, it’s not significantly noisier than what you’ll encounter on regular and much slower cards. Slightly more pervasive was the sound of the power delivery coils compared to the RTX 4090 Founders Edition. However, they’re not significantly more audible than the cooler itself, nor are they any noisier than those found in many RTX 4090 models. It remains to be seen how other manufacturers will deal with the higher power consumption of the RTX 5090.

What I fear most is the prices in stores. It looks like demand is going to exceed supply for quite a long time to come. Even at 2000 USD, a new GeForce with 32GB of memory is an extremely attractive product, especially for professional use. Especially nowadays when everyone is playing with AI models and flirting with how to make money on AI or have it work for them. It’s also attractive to people revolving around the audiovisual industry, which is also a growing market. The market is extremely hungry right now, RTX 4090s have been sold out in our neck of the woods for a few weeks now. You can hardly expect retailers to stay grounded with the price under these circumstances. But the laws of the market are relentless in this regard, it will sell for what people are willing to buy it for. There was no reason to discount the last generation, and it was even the case that towards the end of sales, the RTX 4090s were getting more expensive rather than cheaper. Unfortunately.

In terms of performance, the RTX 5090 has no competition, and in terms of features, it can only be compared to much more expensive professional models. It also leads the pack in terms of offered technologies. Power consumption is high, but proportional to the increased performance. The only real criticism is its extreme price, which is simply a result of supply and demand. Few would say they wouldn’t want a card like this at home, and you can hardly get anything better—making the award well-deserved.

Pros and Cons

English translation and edit by Jozef Dudáš