We analyze the specifications of the Geforce RTX 5090 compared to the RTX 4090.
At the Consumer Electronics Show 2025 (CES 2025) in Las Vegas, Nvidia presented its new flagship graphics card in the RTX 50 series, along with other models. The focus was not only on raytracing performance, but also on the performance of the latest DLSS version.
Thanks to Multi Frame Generation, DLSS 4 is said to be able to generate up to eight times more images per second than would be possible natively.
But how well does the RTX 5090 perform in terms of rasterization – i.e. without raytracing and DLSS?
RTX 5090 vs. RTX 4090: The specifications
| RTX 5090 | RTX 4090 | |
|---|---|---|
| Graphics chip | GB202-300 | AD102-300 |
| Cores | 21,760 | 16,384 |
| Clock speed (base/boos) | 2,017/2,407 MHz | 2,235/2,520 MHz |
| Memory | 32 GB GDDR7 (512-bit) | 24 GB GDDR6X (384-bit) |
| Memory Transfer Rate | 28 Gbps | 21 Gbps |
| memory bandwidth | 1,792 GB/s | 1,008 GB/s |
| Power Consumption | 575 Watt | 450 Watt |
RTX 5090 vs. RTX 4090: Theoretical Computing Power
The number of cores
To get an idea of how much more powerful the new flagship is compared to its predecessor, a comparison of the number of cores is usually a good indicator:
The RTX 5090 has 21,760 computing units, around 32.8 percent more than the RTX 4090.
The raster performance should therefore make a similar leap.
The TFLOPS
It is even more precise to also include the clock rate.This results in the so-called teraflops (TFLOPS), i.e. how many floating point operations per second (additions or multiplications) can be performed.
This is the formula for it: Teraflops (TFLOPS) = Shader Compute Units x Clock Frequency (MHz) x Operations per Clock / 1,000,000
Using the RTX 5090 as an example: 21,760 cores x 2,407 MHz x 2 / 1,000,000 = 104.8 TFLOPS
Compared to the RTX 4090 (82.6 TFLOPS), the RTX 5090 has 26.9 percent more (theoretical) computing power.
Comparing the teraflops of two graphics cards from different series provides only limited information about performance in games, because, for example, the architecture can be more efficient.
However, if the models are directly successive and probably partly identical in structure, they are still a useful indicator.
RTX 5090 vs. RTX 4090: How big will the performance difference be in games?
In our estimation, the RTX 5090 should be roughly 20 to 30 percent more powerful than the RTX 4090 – in terms of raster performance.
When raytracing and especially DLSS 4 come into play, however, things look different. Multi Frame Generation is able to generate three intermediate images instead of just one (as is the case with DLSS 3 and simple frame generation), which significantly increases the frame rate.
However, this results in a certain input delay, and it remains to be seen how well the new form of frame generation ultimately affects the decisive feel for the game.
Does the faster graphics memory affect performance?
Not directly. The speed or rather the bandwidth is basically only there to ensure that the graphics chip is always supplied with sufficient data (textures, models, etc.) to ensure that it always exploits its full performance potential.
However, if bottlenecks occur because the memory cannot keep up and becomes a bottleneck, faster memory can indeed lead to more FPS.
This is especially relevant at high resolution levels such as 4K (or even 8K), when particularly high-resolution textures and models as well as ray tracing are used.
