RTX 5090 vs RTX 4090: See the full comparison between GPUs

Nvidia introduced the powerful GeForce RTX 5090 with an impressive announcement at CES 2025. The model is coming to be the most powerful graphics card in the world of home computers and is focused on offering more performance and much more video memory for AI, but how different is it from this model? Is this card from its predecessor RTX 4090?

Targeting the high-end segment, both the RTX 4090 and RTX 5090 are technically highly advanced cards with an emphasis on delivering maximum performance in games or professional applications.

Powerful chips

As expected from this announcement, the GeForce RTX 5090 offers a much more powerful chip than the previous model. However, while the new generation uses the Blackwell architecture that Nvidia used in its artificial intelligence accelerators a few months ago, as well as billions of transistors, the RTX 40 generation uses the highly praised Ada Lovelace architecture.

The differences here translate into improvements and how the manufacturer polishes this new chip. Naturally, there is a very noticeable generation jump between the models, which we will explain with numbers and percentages in the following paragraphs.

The main way to tell both cards apart is to look at CUDA core count, which is GeForce’s primary unit of measurement. While the RTX 4090 has 16,384 of these units, the RTX 5090 increases this value by 32% by offering 21,760 CUDA cores optimized with the new architecture.

This doesn’t necessarily mean that the 5090 will deliver 30% more performance, but it’s a clear indication of the power of the new silicon. The same goes for Ray Tracing’s core count, which jumped from 191 to 318 in the third generation of these accelerators to work with realistic lighting in games and 3D projects.

However, when we talk about artificial intelligence, what seems most impressive is this. Nvidia’s own conference was entirely focused on this topic, with the company’s CEO Jensen Huang announcing the RTX 50 while introducing the next step in AI for the future.

What attracts the most attention is the amount of AI TOPS found in the RTX 5090, that is, the unit of measurement used to measure the power of the chip and perform artificial intelligence-related tasks. The new GPU has 3,300 TOPS; This represents a 150% increase compared to the previous generation’s 1,321.

more memory

In addition to the chip, the RTX 5090 also brings a very nice feature in this new hardware era: GDDR7 memories. This new protocol, which has been on the market for a short time, will work even faster in information transfer and will be necessary for the GPU to draw frames faster.

For this graphics card, Nvidia installed 32 GB of VRAM via a 512-bit memory interface, creating 28 Gbps speeds and 1.792 GB/s bandwidth. In this category, there are again increases of more than 30% compared to the RTX 4090, which has 24 GB and a total bandwidth of 1 TB/s on the GDDR6X protocol with a 384-bit bus; This is a huge number.

This increase in memory is not all a coincidence, as the demand for components more compatible with AI inference requires a certain amount of VRAM. In games, this memory needs to be used by technologies such as DLSS 4 and Multi Frame Generation, although it seems like something much more idealized for professional workflows.

Performance

The most important issue in graphics cards is performance, and Nvidia promises serious performance breakthroughs for the new generation. In the graphics published by the company, RTX 5090 has twice the performance compared to RTX 4090 in games such as Alan Wake 2, Cyberpunk 2077 and Black Myth Wukong; the latter runs at an incredible 240 FPS.

However, you need to be careful when analyzing such charts. All three of these games achieve this frame rate with the help of DLSS 4 and Multi Frame Generation, even with Ray Tracing enabled. Therefore, this is a performance made with the help of artificial intelligence and does not represent the raw performance of the GPU, that is, running natively.

Design and construction

Another striking element of the RTX 5090 compared to the RTX 4090 is its size. Although it maintains the length of its predecessor, Nvidia has managed to reduce the thickness of the graphics card enough to make it a two-slot model, meaning it is much thinner.

Measuring 304 mm (L) and 137 mm (W), the new Founders Edition model has a redesigned look and looks more minimalist without the rough look of the previous generation. A significant change has been made to the power connector, which is now in an angled position to prevent the user from incorrectly plugging it in.

power consumption

The RTX 4090 caused controversy at launch when it required users to have an 850W power supply to use the GPU. Things are even more complicated with the RTX 5090, because the card increases energy consumption by 27% and draws 575W from the outlet. Therefore, Nvidia recommends combining the product with sources with a power of at least 1,000W.

availability

GeForce RTX 5090 is launching on January 30, but its price is the biggest question mark right now. The model will be priced at $1,999 in the Founders Edition, which is a 25% increase over the $1,599 charged for the RTX 4090 at the time of the card’s launch.

Nvidia stated that the recommended price of the RTX 5090 in Brazil will be up to R$ 20,000, but this value may change due to fluctuations in the dollar and special conditions at major retailers. In the national market, the GPU price will be 33% higher than the recommended R$ 14,999 for the RTX 4090 in 2022.

KaBuM Coupon!

Looking for discounted Nivia GPUs? Take advantage of the KaBuM coupon! and guarantee special prices to upgrade your PC!