Nvidia's RTX 50 series should offer an enormous leap in performance

In addition to the new graphics cards in the RTX 50 series, which promise up to twice the performance, Nvidia is also showing the new version of the DLSS upscaling technology. With DLSS 4, up to eight times the performance of DLSS 3 should be possible.

RTX 5090

There will probably not be more high-end graphics cards in the next two years than the RTX 5090. The card with a recommended retail price of USD 1999 is a beast.

The new Blackwell architecture, on which the RTX 5090 and all cards in the 50 series are based, has fifth-generation tensor cores, new streaming multiprocessor units and fourth-generation ray tracing cores.

The RTX 5090 is based on the GB202 GPU. It has 21,760 computing units activated. The GPU would have a total of 24,576. This gives Nvidia room to possibly release an even more powerful model at a later date. However, due to a lack of competition, the chip giant was already able to dispense with this in the predecessor, which also does not make use of the complete GPU.

Nvidia is not stingy with memory for once: the card offers 32 gigabytes (GB) of the fast GDDR7 type. With a 512 bit interface, the card is up to 30 Gbps fast with a bandwidth of up to 1792 GB/s. The total board power (TBP) is 575 watts. That sounds like a lot, but in reality the power consumption will be significantly lower, for example when gaming. This applies to all cards in the RTX 50 series. The RTX 5090 will support PCIe 5.0 and DisplayPort 2.1b UHBR20 (8K 165Hz).

In terms of performance, Nvidia shows examples with DLSS 4 enabled, with the RTX 5090 offering twice the performance of the RTX 4090 in Cyberpunk 2077, Alan Wake 2 and Black Myth Wukong. Without DLSS, the performance boost in games such as "Far Cry 6" is 30 to 40 per cent.

The card is set to go on sale from 30 January.

RTX 5080

The RTX 5080 utilises the full configuration of the GB203 GPU. It has 10,752 computing cores. Compared to the larger RTX 5090 model, this is 51 per cent less. As an estimate: The difference in computing cores from the RTX 4090 to the RTX 4080 was still around 40 per cent.

In terms of VRAM, the GPU offers 16 GB capacity with a 256-bit bus interface. GDDR7 is used for the modules. With speeds of up to 30 Gbps at 960 GB/s bandwidth, it will offer the fastest memory of its kind. The TBP is 360 watts, which is 12.5 per cent more than the RTX 4080 Super.

Thanks to the new DLSS-4 technology, the RTX 5080 is said to be up to two times faster than the RTX 4080 Super. Nvidia quotes a price of 999 US dollars, which is 200 less than the original RTX 4080. The card is set to go on sale on 30 January.

RTX 5070 Ti

The RTX 5070 Ti is based on the same GB203 GPU, but it is limited. It has 8960 CUDA computing units at its disposal. The memory available is 16 GB GDDR7. Thanks to this, the bandwidth increases by 78 per cent to 896 GB/s compared to the RTX 4070 Ti. The TBP is 300 watts, which corresponds to an increase of just over 10 per cent compared to its predecessor.

In terms of performance, the specifications should double that of the RTX 4070 Ti. Nvidia quotes a price of 749 US dollars, which is also cheaper than the RTX 4070 Ti at launch.

RTX 5070

The RTX 5070 is based on the GB205 GPU. It has 6144 computing units. The VRAM has 12 GB GDDR7 available. The memory speed of up to 28 Gbps at 672 GB/s ensures a 33 per cent increase in bandwidth. The TBP is 250 watts, which is 14 per cent more than the RTX 5070.

Nvidia promises more performance with the RTX 5070 than with the RTX 4090 at a recommended retail price of 549 US dollars for the Founders Edition, which is only available directly from Nvidia.

DLSS 4

The new RTX-50 series also comes with DLSS 4, the most ground-breaking feature of which is Multi Frame Generation. This technology uses generative AI to predict up to three frames before the conventionally rendered frame. The conventionally rendered frame is already based on DLSS super-resolution technology. This makes it possible to up-render one pixel into four. As a result, DLSS 4 has a pixel generation factor of 1:15, which means that 15 out of 16 pixels are generated outside the rendering pipeline.

DLSS 4 launches at the same time as the RTX-50 series at the end of January. The feature will be exclusive to the Blackwell cards. Over 75 games will already support the technology.

If you missed the presentation at CES, you can watch it here: