NVIDIA CEO, Jensen Huang, delivered his keynote speech at CES Las Vegas on the 6th January and confirmed the much rumoured launch of the Nvidia GeForce RTX 50-Series Graphics Cards. We now know for certain when the first of the new GPUs will be available, with the GeForce RTX 5090 releasing on the 30th January starting at £1,939 and the RTX 5080 also launching on the 30th January, starting at £979, which is towards the higher-end of the $2000-$2500 estimate that was rumoured for the 4090. You can read the full announcement from Nvidia here. Read on below for our summary of the main announcements and what they mean for you below.
Built on Blackwell architecture
Neural rendering and AI enhanced graphics are the big focus for the 50 Series, and that is why the NVIDIA RTX Blackwell architecture has been built specifically optimised for this. Blackwell offers unprecedented power and efficiency, pushing the boundaries of computational graphics and AI workloads.
At the heart of Blackwell is its incredible 92 billion transistors, marking it as the most powerful consumer GPU ever built. The redesigned Blackwell Streaming Multiprocessor (SM) integrates tightly with Tensor Cores, enhancing throughput and optimizing the performance of neural shaders. Additionally, Blackwell introduces advanced Shader Execution Reordering, doubling its efficiency to accelerate the speed and precision of shader processing.
The architecture’s Tensor Cores, equipped with cutting-edge AI processing capabilities, support FP4 precision models. This enables faster model processing while significantly reducing the memory footprint, critical for memory-intensive workloads like machine learning and neural network applications.
A major advancement in Blackwell is its dedicated hardware for next-gen DLSS Multi-Frame Generation, with precision hardware flip metering that ensures smooth, high-quality frame pacing. For ray tracing, the architecture includes new RT Cores that deliver twice the ray-triangle intersection rate of prior generations, alongside advanced compression techniques to manage larger, more complex geometry without taxing memory.
Blackwell also integrates PCIe Gen5 and DisplayPort 2.1b UHBR20 for next-level display performance, supporting resolutions up to 8K at 165Hz. For mobile applications, the GeForce RTX 50 Series laptops feature Max-Q technologies, which include Advanced Power Gating, Low Latency Sleep, and Accelerated Frequency Switching. These innovations boost battery life by up to 40%, enhancing portability without compromising performance.
To power these advancements, Blackwell is equipped with the latest GDDR7 memory, achieving speeds up to 30Gbps. This enables up to 1.8TB/s of memory bandwidth, providing the necessary throughput for the most demanding computational tasks, whether for complex simulations, high-resolution rendering, or deep learning applications.
In summary, the RTX Blackwell architecture sets a new benchmark for GPU performance, combining AI acceleration, ray tracing, and memory efficiency to cater to the most specialized needs of industries at the forefront of technology.
What does this mean for your customised PC?
If you’re looking to build a system utilising the latest generation graphics cards once they become available, you are probably going to want to consider a latest generation motherboard with support for PCIe 5.0, especially if you’re using your system for non-gaming or multi-GPU workloads.
Welcome to the future with Neural Rendering
For PC enthusiasts focused on performance-driven computing, particularly those in fields like gaming, photography, videography, architecture, and AI development, neural rendering represents a transformative leap in graphics technology. This next-generation rendering process integrates neural networks directly into the graphics pipeline, enabling breakthroughs in performance, image quality, and interactivity, providing unparalleled levels of realism and immersion.
DLSS (Deep Learning Super Sampling) was the pioneering example of neural rendering, utilising lower-resolution frames as input for a neural network, which then upscales them to full resolution. This technology has since evolved significantly, with DLSS 4 with Multi-Frame Generation amplifying frame rates by up to 8X compared to traditional rendering methods while enhancing image quality to exceed native resolution. This leap not only boosts graphical performance but enables smoother, more responsive experiences without sacrificing visual fidelity.
Beyond DLSS, RTX Neural Shaders are now pushing the boundaries of graphical innovations. These programmable shaders harness the power of neural networks to deliver impressive results, such as compressing textures by up to 7X, significantly reducing memory usage, and enabling cinematic-quality textures and advanced lighting effects. This will have a major impact on memory-intensive applications, allowing for more complex scenes without overburdening system resources.
An exciting advancement is RTX Neural Faces, which uses generative AI to enhance facial rendering in real time. By taking a basic rasterized face and 3D pose data, Neural Faces can produce more natural and detailed facial expressions, revolutionizing character rendering for games, films, and virtual environments.
For enthusiasts, this shift means that investing in high-performance GPUs that support neural rendering, like those based on the NVIDIA RTX Blackwell architecture, will offer tangible benefits: more efficient rendering, better texture quality, and smoother frame rates. For industries relying on visual realism, neural rendering will dramatically improve workflows, whether for game development, 3D modelling, or AI-based visual tasks.
GeForce RTX 5090
The lead selling point of the RTX 5090 is that thanks to Blackwell architecture and innovations in DLSS 4, it should outperform the RTX 4090 by 2X.
The 5090 will also boast an enormous 32GB of latest generation GDDR7 memory, compared to the 4090’s 24GB of GDDR6X. The full specs of the 5090 can be found here but the key takeway for gamers is that you should be able to play some of the most popular recent games at 4K 240Hz/FPS will full ray tracing and max settings.
For none-gaming applications, the 5090 makes some equally impressive boasts, particularly around Generative AI, where Nvidia claims it is also 2X faster at image generation whilst using half the memory. This is a significant speed increase for anyone working in AI development.
GeForce RTX 5080
The RTX 5080 is also reported to be twice as fast as the previous generation 4080, with DLSS 4, 5th Gen Tensor Cores and 4th Gen Tensor Cores. The VRAM is still 16GB, but it’s 16GB of GDDR7 compared to the GDDR6X of the 4080.
RTX 5070 Ti and 5070
Both the RTX 5070 Ti and 5070 are claimed to be twice as fast as their 40-Series equivalents. The 5070 Ti will start at £729 and the 5070 will start at £539.
NVIDIA DLSS 4
Nvidia DLSS (Deep Learning Super Sampling) is the name given to the neural rendering technologies that boost frame rates. DLSS renders frames at a lower resolution than displayed and then uses deep-learning and AI to upscale the frame so they look sharp and well defined at the native resolution.
Nvidia has announced a new advancement in the form of DLSS 4 which will feature Multi Frame Generation, generating up to three additional frames per traditionally rendered frame, multiplying frame rates by up to 8X over traditional brute-force rendering.
NVIDIA REFLEX 2
For anyone serious about gaming, latency can be the difference between winning and losing. Nvidia Reflex was introduced to reduce latency by synchronising CPU and GPU work, reducing input times, making games more responsive and reducing latency by an average of 50%. Nvidia Reflex 2 boasts a latency reduction of up to 75% which is significant.
