NVIDIA has finally officially announced the arrival of its new generation of GeForce RTX 50 Series graphics cards, which initially has four models: RTX 5090, RTX 5080, RTX 5070 Ti and RTX 5060. All of these new graphics cards feature the new NVIDIA Blackwell Architecturean extremely important architecture because it integrates a huge amount of new functionality. In this article we explain them all in detail.
We're looking at a GPU architecture, which means it's something quite technical and complex. However, we will try to speak in simple terms (beyond the English names that we are obliged to use because that is how NVIDIA named its technologies) so that everyone can understand it. However, if you have any questions, do not hesitate to tell us in the comments, we will happily expand the article to explain what is needed.
That said, let's go.
NVIDIA Blackwell architecture
According to NVIDIA, one of the main features of the new Blackwell architecture is the introduction of NVIDIA DLSS 4 technology with Multi Frame Generation, which increases gaming performance by up to 8 times and reduces latency by up to 75% thanks to NVIDIA Reflex technology. 2. Additionally, according to the manufacturer, it enables next-generation graphics fidelity with NVIDIA RTX Neural Shaders.
DLSS4: the evolution of neural graphs
We had the first example of neural rendering with DLSS, which uses images rendered at a lower resolution as input to a neural network trained to generate full-resolution images. Since then, DLSS has evolved to the point of generating full frames and understanding the composition of scenes, including shadows or reflections, to produce images that can even improve on native rendering.
Now with the introduction of DLSS 4 with multi-frame generation Working in conjunction with the DLSS suite of technologies, it is possible to increase the frame rate by up to 8x (compared to traditional rendering) and deliver image quality that often exceeds native rendering. But this is just the beginning, as NVIDIA has integrated neural networks into shaders to also create neural shaders, capable of compressing textures up to 7 times (saving a considerable amount of VRAM) while creating Cinema-quality textures with advanced lighting effects.
In addition, technology RTX Neural Faces offers an innovative approach to improve the quality of faces using generative AI. Instead of traditional rendering, Neutral Faces takes a simple rasterized face and 3D position data as input and uses a real-time generative AI model to generate a more natural face.
The DLSS suite also includes Ray Reconstruction, Super Resolution and DLAA which are powered by the graphics industry's first real-time “Transformers” application. These advanced models improve image quality, providing greater temporal stability, fewer artifacts and more detail on moving objects.
How does multi-image generation work?
The DLSS 3 AI model used game data such as motion vectors and optical flow fields to generate an additional image. However, generating multiple images was quite complicated, even prohibitive due to the performance cost. DLSS 4 combines hardware and software innovations to make multi-frame generation a reality. Its new AI model is 40% faster, uses 30% less VRAM, and only needs to run once per rendered frame.
The generation of the optical flow field was also accelerated by replacing its hardware mechanics with an AI model. Together, the computational cost of generating these additional frames is significantly reduced.
To solve the complexities of multi-image generation, Blackwell uses Flip Metering hardware, which moves the logic from the images to the display engine. Blackwell's display engine has also been enhanced with doubled pixel processing capacity to support higher resolutions and refresh rates for hardware Flip Metering with DLSS 4.
Working together, these hardware and software innovations allow DLSS 4 to produce 15 by 16 pixels with exceptional image quality, smoothness and latency.
In games and applications, DLSS 4 with Multi Frame Generation, Ray Reconstruction and Super Resolution increases FPS by up to 8 compared to not using DLSS, i.e. traditional rendering, and up to to 1.7 compared to Frame Generation, the previous technology.
The Transformer model: more quality in games
DLSS 4 also features a major update to Ray Reconstruction, Super Resolution and DLAA, using the first real-time transformer-based model. Previously, DLSS used CNN (Convolutional Neural Network) to generate new pixels by analyzing the context and recording changes in those regions in subsequent frames. According to NVIDIA, after six years, the limit of what this model could offer was reached.
Thus, the new transformer model makes it possible to evaluate the relative importance of each pixel across each image and over several images; Using twice as many parameters as CNN, it is able to understand scenes better. The new model therefore provides greater image stability, reducing ghosting, providing a higher level of detail in moving objects and rounding the edges of image scenes.
The Transformer model for Super Resolution is still in development, but NVIDIA claims that it is showing very promising results and that they will soon release a beta version so that users can take advantage of the improvements it offers.
DLSS replacement
Although there are already 75 games and apps compatible with Multi Frame Generation on day one, there are dozens of titles that are not yet compatible. Therefore, the new NVIDIA app will enable the new feature called DLSS Override which will allow some adjustments via the Graphics -> Program Settings screen:
- DLSS replacement for frame generation– Enables the use of multi-frame generation with RTX 50 series graphics in titles that support single frame generation.
- DLSS Override for Template Presets– Enables the use of the latest Frame Generation model on RTX 50 and RTX 40 graphics, as well as the use of the Transformer model for Super Resolution and Ray Reconstruction for all RTX graphics when DLSS is enabled in-game.
- DLSS replacement for super resolution– Sets the internal rendering resolution for DLSS Super Resolution, enabling DLAA or Ultra Performance mode when Super Resolution is enabled in-game.
NVIDIA Reflex 2
In 2020, NVIDIA Reflex was initially launched, an innovative technology that serves to reduce PC latency and in competitive games improves it by up to 50%. This is achieved by synchronizing the CPU and GPU, and the result is that the player's movements and actions are reflected much faster in the game. Over the past 4 years, NVIDIA has brought this technology to over 100 games, which is no small feat.
Now, NVIDIA boasts that the evolution, Reflex 2, can reduce PC latency by up to 75%, by combining Reflex Low Latency mode with new technology. Frame deformationwhich manages to further reduce latency by updating the image generated by the game with the last mouse action even before it is displayed on the screen.
NVIDIA ACE and Project G-Assist
Originally introduced in 2023, NVIDIA ACE is a suite of RTX-accelerated technologies that seek to bring game characters to life with AI. Now NVIDIA is expanding ACE so that instead of just having NPCs you can talk to, we have real characters who use AI to perceive, plan, and even act as if they were human players.
NVIDIA works closely with developers to integrate ACE into their games, and we can already see it in titles like PUBG: Battlegrounds, InZOI or Naraka: Bladepoint.
On the other hand, we have G-Assist Projectan AI assistant still in the experimental phase. In theory, with a simple text command or even a voice command, we can make G-Assist maximize performance and optimize graphics power efficiency, adjusting game settings and providing real-time diagnostics.
Improvements for content creators
Finally and very quickly, the next generation of NVIDIA graphics also brings many improvements for content creators, such as support for 4:2:2 encoding and decoding for H.264/H.265 videoand demand NVIDIA Broadcast Updated with several improvements, such as:
- Virtual key lightwhich is capable of enhancing the lighting captured by the camera to automatically improve the appearance of the image.
- Studio vocals– A new AI-enhanced microphone effect designed to significantly improve streaming quality. This feature eliminates ambient noise and echo that may be in the room, but it also significantly improves quality by taking samples of your voice and enhancing them with AI, so that even if you're using a microphone poor quality, it looks like you. are in a recording studio.
- New microphone and camera effectslike split camera mode to be able to see live the changes you apply to the recording.
The article NVIDIA Blackwell Architecture: Everything You Need to Know appeared first on HardZone.
Table of Contents
