GeForce RTX 5090 reviews are out – finally high FPS in 4K, but you’ll have to fork out more

The new flagship graphics card GeForce RTX 5090 will be available on January 30. Nvidia has allowed specialized media and bloggers to publish today their reviews of the new product in the Founders Edition reference version, the recommended price of which is $2000.

Image source: TechPowerUp

The GeForce RTX 5090 is based on the GB202 graphics processor, built on the Blackwell architecture. The GPU contains 92.2 billion transistors. This is more than the AD102, which is the basis of the GeForce RTX 4090, with 76.3 billion transistors. However, the quantitative increase of the latter in the GB202 is not explained by the transition to a new technological process – for the production of GB202, like other processors of gaming video cards in the Blackwell line, the same 5-nm TSMC process is used as for the production of GeForce RTX 40 series video cards.

Image source: TechPowerUp

The increase in the number of transistors can be explained rather by a simple increase in the area of ​​the GPU itself. While the AD102 (Ada Lovelace) measures 609 mm², the GB202 measures 744 mm², making it the second-largest consumer GPU after the Turing-generation TU102 at 754 mm². At the same time, the transistor density of the new GB202 GPU is even lower than its predecessor – 123.9 million/mm² versus 125.3 million/mm² for the AD102.

Image source: TechPowerUp

The GB202 GPU features 21,760 CUDA Cores (33% more than the RTX 4090), 680 Texture Units, 192 Rasterization Units, 170 SM Units, 680 Gen 5 Tensor Cores, and 170 Gen 4 RT Cores. The base GPU frequency is 2017 MHz, the maximum is stated at 2407 MHz. But, as practice shows, the RTX 4090 in games, the new product will be able to provide a higher frequency. The same RTX 4090 produces 2750 MHz, despite Nvidia’s declared 2520 MHz. The flagship of the Blackwell series received 32 GB of new GDDR7 memory with a speed of 28 Gbit/s per contact and support for a 512-bit interface. Thanks to this, the memory bandwidth of the RTX 5090 is 78% higher than its predecessor, amounting to 1.79 TB/s.

The card’s stated power consumption is 575 W, which is 27% (125 W) higher than that of the RTX 4090. Nvidia recommends using a power supply of at least 1000 W with the new product. The GeForce RTX 5090 is equipped with one 16-pin additional power connector, a PCI Express 5.0 x16 interface, as well as external DisplayPort 2.1b and HDMI 2.1a connectors.

One of the key features of GeForce RTX 50 series video cards is exclusive support for Multi Frame Generation (MFG). Unlike the first generation frame generator, which appeared with the release of GeForce RTX 40 series video cards, the new generator creates not one, but three additional frames for every traditionally rendered frame to increase FPS. But more on that below.

Relative performance of GeForce RTX 5090 at 1080p, 1440p and 4K

The GeForce RTX 5090 is definitely a card designed for 4K gaming at maximum quality and ray tracing settings. Therefore, most reviews of this video card are more focused on this resolution. Thus, gaming tests from the TechPowerUp portal show that in terms of pure raster performance (without RT and DLSS), the GeForce RTX 5090 provides up to 35% more performance compared to the GeForce RTX 4090. This is certainly impressive, since traditionally the performance increase between Nvidia flagships of different generations was about 25%. And yet, this increase is less than what was seen when moving from the GeForce RTX 3090 Ti to the GeForce RTX 4090. According to TechPowerUp tests, the difference in performance between generations was then about 51%, and according to other reviewers – even more. The new GeForce RTX 5090 is twice as fast as the RTX 3090 Ti. The video card has no competitors and is not expected. The closest one, the Radeon RX 7900 XTX, lags behind it by up to 75%. Just think about it.

Since AMD’s new Radeon RX 9070 series cards based on the RDNA 4 architecture are expected to perform somewhere between the Radeon RX 7900 XT and RX 7900 GRE, the new RTX 5090 will be at least twice as fast as AMD’s highest performing next-gen card. Compared to Nvidia’s second-fastest Ada Lovelace-generation graphics card, the GeForce RTX 4080, the RTX 5090 has a 72% advantage. And all this is raster performance.

Below we will present the results of gaming tests in 1440p and 1080p resolutions, in which the new card is also the undisputed leader. But it’s worth repeating, the GeForce RTX 5090 is not needed for those who use monitors with resolutions lower than 4K. In this case, the best solution would be to first upgrade the monitor, or even better, the entire PC, and only then look closely at the new Nvidia flagship.

Nvidia has done a lot of work on Blackwell’s ray tracing. The level of performance loss when enabling this technology is lower than that of Ada Lovelace video cards.

The company has also added several new optimization methods that can be used by game developers. The most interesting of these is neural rendering, implemented through the Microsoft DirectX (Cooperative Vectors) API. This ensures that the feature will be available to all video card manufacturers and its use will depend only on the motivation of game developers. The RTX 5090’s 4K ray tracing performance is 32% faster than the RTX 4090. AMD’s fastest graphics card, the Radeon RX 7900 XTX, is 2.5 times slower in this regard.

Since its debut in 2018, Nvidia’s deep learning scaling technology (DLSS) has evolved to its fourth version. Its latest iteration switched to a transformer-type AI model, which made it possible to implement a number of new functions, including multi-frame generation (MFG). The latter allows you to create up to three additional frames for each traditionally rendered frame to increase FPS. MFG is exclusive to Blackwell generation video cards, as was the first generation frame generator for GeForce RTX 40 series video cards. To talk about this technology in more detail would require a separate article. Here we just note that at the moment, native support for this function is implemented in only five games. One of them is Cyberpunk 2077.

How ray tracing, scaling, and frame generator work (Nvidia only) on different video cards and the amount of video memory consumed in Cyberpunk 2077

At the start of sales of the RTX 5090, Nvidia promises 75 games with support for this technology. It’s important to note here that Multi Frame Generation is not some kind of miracle. Yes, it significantly increases FPS in games (the function offers several levels of the number of additional frames to be drawn), but at the same time, input lag also increases, since for one frame actually drawn by the video card, a game that supports MFG draws up to three additional frames. As a result, if a graphics card is running at its native 60 frames per second, but the game displays at 200 frames per second, input lag will feel like 60 frames per second rather than 200 frames per second. This feature will likely not appear on graphics cards. previous generations, unlike other DLSS 4 features.

The operation of a multi-frame generator in the game Cyberpunk 2077. Image source: YouTube / Jayztwocents

The transition from the Convolutional Neural Network (CNN) in DLSS 3 to the Transformer AI model in DLSS 4 also had an impact on the image scaling function. The most important thing to know here is that the Transformer AI model provides significantly higher output image quality when upscaled compared to a CNN. According to a TechPowerUp reviewer, this is actually visible to the naked eye. Image quality in DLSS Quality mode is so sharp that in some cases it surpasses native resolution quality. No more flickering or blurry low-resolution textures on the horizon. Fine wires in images become crystal clear, even at native resolutions below 4K. However, all this comes at a price – the performance of DLSS with a transformer AI model is about 10% lower than when using a CNN. But the end result of scaling is worth it.

Energy consumption

Does it make sense to talk about the power consumption of a video card that costs over $2000 and has a stated TDP of 575 W? The target audience of this product is unlikely to pay attention to such a “trifle”. However, the card consumes all the declared power in games. In some cases, power consumption is even higher, with peaks reaching 644 W (data from TechPowerUp). In idle mode, the card’s power consumption is 30 W with one monitor and 39 W with two monitors. Video playback requires 54 watts. In games, power consumption is about 587 W (561 W with RT enabled). The maximum power consumption according to TechPowerUp tests was 601 W. We are talking only about the video card, without taking into account the consumption of the processor, motherboard and other PC elements.

Temperatures and noise of the reference GeForce RTX 5090 Founders Edition

In the reference version, the temperature of the GeForce RTX 5090 GPU, according to tests, was 77 degrees Celsius, and the GDDR7 memory warmed up to 94 degrees. At the same time, the fan noise was 40.1 dBA – it is louder than the reference GeForce RTX 4090 (35.1 dBA) and Radeon RX 7900 XTX (39.2 dBA). Probably, custom versions of the RTX 5090 from Nvidia partners, equipped with proprietary coolers, will offer lower temperatures and noise. Let us remember that the thickness of the reference version of the RTX 5090 is only two expansion slots, and the card is equipped with only two cooling fans. The cooling system also uses a new three-dimensional evaporation chamber and a thermal interface in the form of liquid metal. You can read more about this here.

Image source: NVIDIA

There really are no alternatives for the GeForce RTX 5090. AMD, within the new generation of video cards, decided not to even try to catch up with its competitor and in fact completely lost the senior segment of gaming accelerators to Nvidia solutions. AMD cards lack raw performance and have nothing to match with features like DLSS multi-frame generation and AI transformer scaling model. Nothing is currently known about AMD FSR 4 and its features. On the other hand, the $2000 price tag for the RTX 5090 is extremely high. That’s a $400 increase over the RTX 4090’s MSRP. Even so, there’s no doubt that stock of these cards will sell out—there’s simply nothing better.