Today the Blackwell architecture for home use is officially presented, giving life to NVIDIA’s new RTX 50 range. We were lucky enough to attend its official presentation and learn all the details during the last CES, as you could see in our news coverage.
This new architecture has some important pillars, in addition to features that we have all read and heard about these days. Here I will try to summarize what I saw in person during the “Editor’s Day” that NVIDIA held at CES. These are just a few strokes of what we will later see in practice in our next reviews of these new GPUs, which are marking an important turning point with respect to what we understand by GPU.
Blackwell architecture for RTX 50 GPUs
This new architecture has significant milestones that define it as the first GPU designed specifically to work with workloads typical of AI transformation models. Its objective is to offer immense AI processing capacity to give life to a new way of understanding a GPU, based on this technology, which allows performance leaps that until now depended more on classic hardware dedicated to rasterization.
The new GeForce RTX 5000 are based on an architecture with 92 billion transistors manufactured using a TSMC 4N process (versus the previous generation’s 5N process). They include new generations of Tensor Cores, now in their fifth generation, and also a fourth generation of RT Cores. Both engines offer significant advances in AI-powered image rendering, including new “Neural Shaders” and technologies aimed at geometry, lighting and digital humanity.
The epitome of this new generation will be the NVIDIA GeForce RTX 5090based on a Blackwell chip with a power of up to 3,352 TOPS (teraoperations per second). Combined with the new DLSS 4 technology, it manages to double the performance of the powerful NVIDIA GeForce RTX 4090, the best graphics card of the previous generation.
This new generation is also the first in the GPU market to use high-speed, more energy-efficient GDDR7 memory, with up to 32 GB available on the GeForce RTX 5090. This memory not only consumes half the power of GDDR6, but also offers twice the bandwidth. In the case of the RTX 5080, the memory capacity is reduced by half, but this generation optimizes memory consumption in various important processes.
Blackwell also comes to laptops
In this generation, the line between laptop and desktop GPUs is further blurred, as every GPU in this generation will feature NVIDIA’s Max-Q technology. Even desktop GPUs will support deeper sleep states, with options to turn off sections of the GPU that are not in use at a given time, as well as make dramatic frequency changes in fractions of a second, adapting nimbly to each situation.
For laptops, we will begin to see these new graphics in April, with prices that will range between 1,500 and 3,500 euros, depending on the integrated graphics. There will be options ranging from entry-level RTX 5070 models to RTX 5090 laptop models.
This new generation of laptops, which will come from the main manufacturers on the market and with both Intel and AMD processors, will enjoy more autonomy not only on the part of the GPU, but NVIDIA has built a complete platform where we can expect up to 30% more autonomy while browsing or playing video than in previous generations. This is achieved both with new generation CPUs, better energy profiles in the operating system and also more efficient storage units.
DLSS 4: the star of performance
DLSS 4 will be exclusive to this new range. This version not only improves super-resolution and ray reconstruction, but also expands frame generation, allowing up to three artificially generated frames to be configured using AI algorithms run on the new fifth-generation Tensor Cores. This technique doubles the performance of DLSS 3.5 on supported cards and improves up to eight times the performance of traditional raster rendering.
To counteract the latencies generated by artificial frames, NVIDIA has developed NVIDIA Reflex 2which introduces Frame Warp technology to reduce latency by up to 75%. In tests carried out, with prepared games, latency even seems to be reduced compared to native rendering.
NVIDIA hopes that from launch day DLSS 4 It is available in up to 75 next-generation games and some very recognized and played titles such as Microsoft Flight Simulator 2024, Cyberpunk 2077, STALKER 2: Heart of Chornobyl or Starship Troopers: Extermination. You have the complete list here: NVIDIA.com
NVIDIA not only achieves surprising results in terms of performance, but this generation also allows additional frames to be generated in a single pass for each real frame. This allows up to three frames to be generated where previously it was only possible to create one additional frame.
Possible visual defects are corrected thanks to technology Blackwell Flip Meteringin charge of managing the presentation of the images. Furthermore, this process is carried out with an improvement in efficiency, being up to 10% more efficient than the previous generation and consuming up to 400 MB less memory by generating 4K images with the highest quality. In total, 15 of every 16 pixels generated by DLSS 4 are generated by AI, awesome.
Previously, DLSS used Convolutional Neural Networks (s) to generate new pixels by analyzing the localized context and tracking changes in those regions over successive frames. After six years of continuous improvements, we have reached the limits of what is possible with the DLSS architecture.
The new model of DLSS transformers uses AI-based transformerwhich allows “self-understanding” operations to be performed to evaluate the relative importance of each pixel in the entire frame and across multiple frames. With twice the parameters of the model, this new approach achieves a deeper understanding of scenes, generating pixels with greater stability, reduced ghosting, greater level of detail in movement and smoother edges within a scene by eliminating flickering. typical of these algorithms with image generation and ray reconstruction.
In ray-traced intensive content, the Transformers model for Ray Reconstruction offers a notable jump in image quality, especially in scenes with challenging lighting conditions. Stability is improved on complex details such as fence mesh, ghosting on fan blades is reduced, and flickering on electrical cables is eliminated, significantly improving the player’s immersive experience in this third-person game. .
New technologies and tools for creators
The RTX 5000 present the new RTX Neural Shaderswhich use transformation algorithms to improve the performance of traditional shaders. These improvements include RTX Neural Faces, which offer more natural gestures in characters, better light transmission in materials such as skin, and massive geometry (NVIDIA RTX Mega Geometry) with billions of triangles on screen.
Besides, NVIDIA ACE introduces more autonomous AI-assisted characters, with enemies that adapt better to situations and improved management of NPC behavior. NVIDIA also presents RTX AI PCwhich allows you to create AI agents and assistants on computers with multi-generation RTX graphics, standardizing access to advanced AI models such as FLUX, Meta and Stability AI.
The GPUs also include a new dual video processing engine capable of handling advanced modes such as AV1 in 4:2:2 configuration, and new interfaces such as DisplayPort 2.1, which supports 8K monitors at 240 Hz or 1080p monitors at 960 Hz.
Availability and prices
The new series NVIDIA GeForce RTX 50 It will be available in stores starting January 30, with prices in Spain that will range between 599 euros for the RTX 5070 and the 2,500 euros for the RTX 5090. There will be two intermediate models: the RTX 5070 Ti, for about 800 euros, and the RTX 5080, which surpasses the RTX 4090 in performance, with an approximate price of 1,300 euros.
The new reference models of the brand, fundamentally RTX 5080 and RTX 5090feature a new ultra-compact PCB design that is centered on the card. Both axial fans, this generation located in the same direction, have free crossing throughout the card. In reality, these new models are made up of three independent PCBs: the main one, connected by a vapor chamber to the aluminum radiator, a second PCB with the PCI Express connector and a third for front graphics connectivity.
| GPU | GeForce RTX 5090 | GeForce RTX 5080 |
|---|---|---|
| Shader Cores | Blackwell | Blackwell |
| Tensor Cores (AI) | 5th Generation – 3352 AI TOPS | 5th Generation – 1801 AI TOPS |
| Ray Tracing Cores | 4th Generation – 318 TFLOPS | 4th Generation – 171 TFLOPS |
| Standard Memory Config | 32GB GDDR7 | 16GB GDDR7 |
| Display Support | 4K at 480 Hz or 8K at 120 Hz with DSC | 4K at 480 Hz or 8K at 120 Hz with DSC |
The first laptops with these GPUs will not be available until March or April. We will have to wait to see the renewed consumption performance in demanding environments in terms of temperatures and autonomy.
End of Article. Tell us something in the Comments!
Add Comment