Earlier this year, Nvidia announced its most powerful graphics card yet. Here’s everything we know about the Nvidia RTX 5090.
The GeForce RTX 5090 sits at the top of Nvidia’s new 50 Series of GPUs based on the brand’s Blackwell architecture. It’s also a direct successor to 2022’s Nvidia 4080 Lovelace card, as well as one of the best graphics cards you can get your hands on.
Read on to discover everything you need to know about the Nvidia RTX 5090, including the release date, pricing and all the important specs and details.
Last Updated: Updated on 12/2 with official launch information.
Nvidia RTX 5090: at a glance
- The Nvidia RTX 5090 launched on January 30
- The GPU starts at $1999/£1939
- The GPU is based on Nvidia’s new Blackwell architecture
- The RTX 5090 is Nvidia’s fastest consumer GPU yet, doubling the speeds of the RTX 4090
- DLSS 4 includes Multi Frame Generation
- FP4 reduces model sizes when tackling generative AI tasks
- Ray tracing has been improved
- The Blackwell architecture supports displays up to 8K/165Hz
- Laptops could see an up to 40% battery life increase
Nvidia RTX 5090 release date & pricing
The Nvidia RTX 5090 was announced on January 6 alongside the RTX 5080 and RTX 5070. The graphics card hit stores three weeks later on January 30, though it quickly sold out. As of February 2, the RTX 5090 is still sold out on the Nvidia website here in the UK.
As far as pricing goes, the RTX 5090 will run you a minimum of $1999/£1939. This is a few hundred dollars/pounds higher than the RTX 4090, which started at $1599/£1679, marking a price increase for the 50 Series. However, the GPUs can run significantly higher than this, depending on the manufacturer, clock speed and other factors, such as demand.
Nvidia RTX 5090 specs
The RTX 5090 is based on Nvidia’s new Blackwell architecture which features up to 92 billion transistors, making it Nvidia’s most powerful consumer GPU yet.
According to the company, the Blackwell streaming multiprocessor features more processing throughput and is more tightly integrated with the Tensor Cores than its predecessor to improve the performance of the neural shaders. The Shader Execution Reordering has also been optimised with doubly efficient reorder logic for better speed and precision, which, in turn, accelerates the performance of the neural shaders.
This Blackwell architecture, combined with DLSS 4 technology, enables the GPU to run twice as fast as the Lovelace-powered RTX 4090 and DLSS 3.5 technology that preceded it. In a video shared by Nvidia, Cyberpunk 2077: Phantom Liberty runs at 234 FPS on the RTX 5090 compared to 109 FPS on the RTX 4090, both with full ray tracing on.
DLSS 4 also introduces Multi Frame Generation, enabling the generation of up to three additional frames for every traditionally rendered frame and unlocking 4K 240Hz/FPS with full ray racing in games like Cyberpunk 2077, Alan Wake 2 and Black Myth: Wukong.
The RTX 5090 offers faster performance in generative AI applications, too. The GPU performs twice as quickly during image generation tasks and requires half the memory thanks to FP4, a method that shrinks model sizes in a manner Nvidia compares to file compression. For example, Black Forest Labs’ FLUX.1 [dev] model uses less VRAM and can generate images in just over five seconds compared to the 15 seconds required by the RTX 4090.
New RT cores offer double the ray triangle intersection rate of those found in the RTX 4090. This, along with a reduced memory footprint, enables the RTX 5090 to ray trace levels of geometry that were previously impossible.
The Blackwell architecture introduces enhanced PCIe Gen5 and DisplayPort 2.1b UHBR20 on the RTX 5090, enabling display support up to 8K/165Hz.
Meanwhile, RTX 50 Series laptops take advantage of new Max-Q technologies, including Advanced Power Gating, Low Latency Sleep and Accelerated Frequency Switching. These can increase the battery life of the device by up to 40% compared to the 40 Series GPUs.
Finally, the RTX 5090 is fitted with up to 30Gbps GDDR7 memory, capable of delivering up to 1.8TB/s of memory bandwidth.
