As expected, Nvidia CEO Jen-Hsun Huang introduced the next generation of graphics cards during the fall of GTC 2022. Love aka GeForce RTX 4000. A lot of time has also been devoted to topics such as metaverse and industries, but if we focus on graphics cards and point out the two most important points of interest, then it will be a technology called Shader Execution Reordering and DLSS 3.0
Shader Run Reorder (SER)
Shader Execution Reordering is a technology and an architectural feature Lovewhich allows you to modify code processing so that the computing units themselves receive data and instructions in the order that is most beneficial from the point of view of code processing efficiency.
The CEO likened this to the advent of out-of-service computing in processors, which is, of course, a pretty overblown comparison. Maybe not in terms of raytracing though. While for standard code the advantage will be smaller, for raytracing the advantage should be crucial and in the mix of rasterization and raytracing the advantage in terms of overall FPS should be up to about 25%.
DLSS 3.0
If we measure the contribution of new technologies in FPS, DLSS 3.0 is even better. The first generation of DLSS increased the frame rate by internally rendering at a lower resolution when a render request was made, for example at a 4K resolution, thus achieving performance and interpolating the missing details / pixels. The second generation, DLSS 2.0, implemented a time dimension, so that the missing details / pixels were also obtained from the previous images, thus improving and refining the details compared to the first generation.
The third generation, DLSS 3.0, is essentially a superstructure, as it adds a kind of frame rate multiplier to the principles of DLSS 3.0. Inserts images obtained exclusively from AI extrapolation between rendered images (and enlarged using DLSS algorithms).
Here we have a slight paradox: Nvidia – with the justification of greater accuracy and mathematical correctness – has introduced raytracing in the game segment, which is used in place (even by an order of magnitude) of less demanding rasterization effects. However, in order for raytracing games to be playable without problems, half of the image information is “invented” in each odd frame and each even frame is “invented” completely.
DLSS 3.0 will be tied to the GeForce RTX 4000, the 3000 generation still has DLSS 2.x listed as the maximum supported.
The main advantage of DLSS 3.0 is that, being an algorithm that multiplies the FPS, or generates images purely at the graphics card level, these FPS are not limited by the performance of the processor. The main drawback is that the game does not support DLSS 2.0 and must be DLSS 3.0 ready (Nvidia demos were run on an unreleased build that supports DLSS 3.0).
Thanks to this, Jen-Hsun Huang presents the GeForce RTX 4090 as 2x-3x-4x faster than the GeForce RTX 3090 Ti. 2x in normal load, 3x with raytracing (which benefits from the new RT units and Shader Execution Reordering) and 4x with raytracing and DLSS 3.0 (in a nutshell).
GeForce RTX 4090
The flagship built on a 4nm core with 76 billion transistors and over 18,000 stream processors will be the GeForce RTX 4090:
So far, Nvidia hasn’t boasted more detailed specs than Micron’s 24GB GDDR6X presence and a $ 1599 price tag (GeForce RTX 3090 was priced at $ 1499). The card will be available on 12 October.
GeForce RTX 4080 – in two ways
About a month later, the GeForce RTX 4080, or the GeForce RTX 4080, will follow. Unofficial reports have been confirmed that two variants are coming: 12GB (which will cost $ 899) and 16GB (for $ 1199). The current GeForce RTX 3080 was officially priced at $ 699 and was built on the largest core of the generation Ampere. If we believe unofficial sources (which have already been partially confirmed), then the 16GB GeForce RTX 4080 will stand on the second largest core of the generation Love and 12 GB GeForce RTX 4080 on the third largest core of the generation Love (this is the configuration originally intended with the name GeForce RTX 4070).
The unspecified of these two cards should be 2-4x faster than the GeForce RTX 3080 Ti – again, you can expect 2x faster rasterization performance and 4x faster performance when using a combination of raytracing and DLSS 3.0. Both cards are therefore expected to be available in November.
It should be added that the GeForce RTX 4090 achieves the announced double performance in rasterization compared to the GeForce RTX 3090 Ti in only two of the five games tested. In the other three, it is only about 50-60% more potent. The GeForce RTX 4080 12GB is even slower in two games and as powerful as the GeForce RTX 3090 Ti in one. The trio of graphs on the right shows the results with raytraced and DLSS 3.0 with “multiplier” frame rates.
The GeForce RTX 4000 will not be released as a replacement for the GeForce RTX 3000, but rather as an addition to the existing range. Nvidia plans to keep the GeForce RTX 3080 and lower range, with the $ 899 and higher price range being covered by the aforementioned trio of GeForce RTX 4080 12GB / 16GB and GeForce RTX 4090 models.
–
