Nvidia’s Ada Lovelace GPU generation: $1,599 for RTX 4090, $899 and up for 4080

After weeks of teases, the new computer graphics cards from Nvidia are here. The two new models, the RTX 4090 and the RTX 4080, will be available in two different configurations.

The same high-end category as the 2020 megaton RTX 3090 is occupied by the pricier card. The 4090 will require three slots on your PC build of choice. The 3090's 10,496 CUDAcores and 2.52 GHz of boost clock are indicative of a high-end graphics card. The 3090 still performs within the 450 W power envelope despite the improvements.

TSMC's "4N" process, which is a new generation of 5 nm chips, provides a massive efficiency jump from the previous Ampere generation's 8 nm process, which is one of the reasons for the jump in performance.

In November, there will be two different versions of the RTX 4080, a 192-bit bus model and a 16-bit bus model. Traditionally, this much Nvidia hardware differentiation has come with separate model names, such as the last generation's 3070 and 3080). We don't know if the 4080 models share a chip or not.

The higher of the two will have more CUDA cores (9,728, up from the 3080's 8,702), boost clock (2.51 GHz, up from 1.71 GHz), and power draw (320 W, same as the 3080 but more than the lower). At least for this generation, Nvidia is offering 12GB of memory as a 4080 baseline, which addresses a major criticism with the consistently memory-starved RTX 3000 generation of graphics cards.

Micromap, micromesh

The new models include updates to the proprietary RTX chip elements. Real-time ray tracing in 3D graphics and its Deep Learning Super-Sampling system have been updated by the company. The former will be augmented with two new types of hardware units: an "opacity micromap engine" and a "micromesh engine".

The latter has been bumped to a new version that seems to have an exclusive feature. The system promises to "generate new frames of the game without the need for the game to be played" The frame-by-frame rebuilding of graphics in real time is designed to solve the issues of image reconstruction techniques that can't model the motion of particles in a game. This new method combined with the techniques established by previous generations could be used to reconstruct a huge amount of a scene's pixels. If it works as promised, it will be a huge improvement over the current process used by both Intel andAMD.

A new process called "shader execution reordering" is said to be one of the reasons for the new graphics cards. Huang's description of the system hinges largely on the cost of ray tracing's work. The system will increase the performance of the previous Ampere generation of graphics cards.

Microsoft Flight Simulator 2020 is the biggest title that was presented by Nvidia as part of the announcements. In some of the game's busiest scenes, frame rates were easily reaching the 100s, and raw rasterization of the same scenes on an unnamedGPU and PC showed solid performance as well. The above gallery is captured in 4K, so you may want to click and zoom in to see how the system handles fine details.

Huang tried to curry favor with the PC modding scene by showing how a new Nvidia-developed toolset can be applied to a number of classic games. The best results came from a before-and-after demonstration applied to Elder Scrolls III: Morrowind, where the game's texture and physical material properties were updated with the help of the RTX graphics cards. If you want to see more before-and-after comparisons, go to the site of the graphics card maker. Similar results are expected from a free Portal 1 DLC package that will be released in November.

Ars Technica may make money from links on this post.