Nvidia GeForce RTX: RTX 4090 and RTX 4080 with DLSS 3 officially presented

As expected, the premium model of the RTX 40 generation will be the RTX 4090. According to the official announcement, this will be launched with a massive increase in performance. The RT cores are said to be able to deliver up to 191 raytracing teraflops (2.8 times the previous generation) and the tensor cores even 1.32 petaflops (approx. 5 times the previous generation). Streaming multiprocessors are said to deliver twice the performance of a 3090 TI at 83 teraflops.

The third-generation RT cores should be able to calculate triangle intersections up to twice as fast as before. A micro-mesh engine should also be able to calculate new micro-meshes in the rendering process that offer more geometry details. This should increase geometry complexity without bringing the corresponding performance costs.

The performance figures of the RTX 4090

The RTX 4090 has 76 billion transistors, 16384 CUDA cores and 24GB of GDDR6X memory, which Nvidida claims should reliably deliver an average of 100FPS at 4K. The power consumption should not increase: Nvidia specifies 450W as the TDP of the RTX 4090. The RTX 4090 will be available October 12 for $1599.

RTX 4080 in two variants

Nvidia also officially presented the 80s model, which is probably of great interest to many gamers. The RTX 4080 will appear in two variants with 12 or 16 GB GDDR6X VRAM, which will also differ in terms of performance. The 16GB version has 9728 CUDA cores, the 12GB version only 7680 CUDA cores. Nevertheless, the 12GB version should also outperform the 3090TI using DLSS 3.

Both RTX 4080 models should be available in November. The 16GB variant will cost $1199 and the 12GB variant will cost $899. It can be assumed that the TDP will be in similar ranges as with the predecessor cards. Both cards should be available in Founders Editions as well as from third-party providers.

DLSS 3 as a potential performance game changer

In addition to the hardware, Nvidia has also announced a significant leap in its Deep Learning Super Sampling (DLSS) technology. This uses – in a much simplified way – deep learning routines to scale lower resolutions without loss of quality. New technologies in the Ada Lovelace architecture of the RTX 4090 and 4080 should bring a significant boost in performance with even less loss of image quality. Nvidia explains the technology in detail in their blog entry.

With DLSS 3 it should also be possible to calculate entire frames on the graphics card side without having to wait for the CPU. This means that the performance, which is limited by the CPU, can almost double. Microsoft Flight Simulator, for example, which is notoriously slowed down by CPU bottlenecks, should, according to Nvidia, achieve double the frames thanks to DLSS 3.

DLSS will be supported at launch for 35 games including A Plague Tale: Requiem, Dying Light 2, Microsoft Flight Simulator and Warhammer 40,000 Darktide. The Unity, Frostbite and Unreal engines will also natively support DLSS 3 in the future. According to Nvidia, the technology should be able to almost quadruple the frame rate in supported games.