Nvidia GeForce RTX 5090 Founders Edition Review

Today is the day when we can finally lift the lid on all the juicy benchmarks for the new Blackwell-based RTX 5090. This is a card that has been hyped up to new levels even before its official announcement at NVIDIA’s keynote at CES at the start of the month. As the new flagship, with some bold claims as to its performance, are we going to see a huge uplift in rasterisation and ray tracing performance, or are we solely going to have to rely on upscaling through the way of DLSS? Well, that’s what we’re going to find out!
Nvidia RTX 5090 Founders Edition
I won’t go into the packaging and design of the card here or what’s included with a Founders card, as we have an unboxing that went live just a few days ago showing that all off. If you’ve not seen that yet, then definitely head over and check it out. I will go over the aesthetics, at least somewhat, of the card and the specs because on paper, things look pretty tasty.
So let’s talk about some of the important specs quickly as being a new architecture, it’s not so much an apples-to-apples comparison, because while we do see a 33% increase in CUDA cores, now at 21,760, and the same in Texture Mapping Units, now up at 680, the render output units see a smaller 9% increase up to 192. Now where things get somewhat interesting is that the base clock actually comes in slower on the RTX 5090 by almost 10%, now sitting at 2017MHz, and the same with the boost clock, which sees a 4% hit, now at 2407MHz.
Key Features
- Blackwell Architecture: The new Blackwell architecture boasts significant improvements in performance and efficiency.
- AI-Powered Rendering: Features like neural shaders and DLSS 4 leverage AI for groundbreaking graphics rendering.
- DLSS 4: This technology uses AI to generate more frames, leading to up to 8x performance boosts compared to traditional rendering.
- RTX Neural Shaders: Introduce AI networks into programmable shaders, enabling film-quality materials and lighting in games.
- RTX Neural Faces: Renders high-quality digital faces in real-time using AI.
- Reflex 2 with Frame Warp: Reduces latency in games by up to 75% for a more responsive experience.
- Autonomous Game Characters: New NVIDIA ACE technologies allow for AI-powered non-player characters (NPCs) that can perceive, plan, and act more like human players.
- AI Foundation Models: Support for NVIDIA NIM microservices enables users to build AI agents and assistants on their RTX AI PCs.
- FP4 Precision: Boosts AI image generation performance and allows for running generative AI models locally.
- Enhanced NVIDIA Broadcast App: Includes new AI-powered features like Studio Voice and Virtual Key Light for improved live streaming.

GDDR7
One area that NVIDIA has focussed on and sees a huge increase is in memory. Not only do we see a 33% increase, now up to 32GB, but we’re now sporting GDDR7, compared to GDDR6X on the Ada Lovelace-based flagship RTX 4090 and now using a 512-bit bus, which is quite a step up from the 384-bit interface seen on the 4090.
Nvidia
Also being a new architecture, we also see an improved fabrication process, now using TSMC’s 4NP process node, which allows for a die size increase over the 4090 of 22% at 744mm2, which also means another 21% more transistors can be packed in, now coming in at a staggering 92.2 billion, making up the GB202-400-A1 core and coming in as NVIDIA’s first consumer-grade GPU that utilises the PCI-Express 5.0 x16 interface.
We also now have the next generation of both RT cores and Tensor cores to give us the very best raytracing performance, but the ability to utilise the latest DLSS4 technology, of course, in games that support it. With AMD seemingly leaving the high-end market, it does look like NVIDIA may have this market completely sewn up with the RTX 5090, especially if the specs we’ve just gone through translate to real-world performance, which is why we’re all here.
These specs, while they do sound impressive, do also come at a cost, as the RTX 5090 sees an MSRP of $1999, so understandably won’t be for everyone, but with the RTX 5080, 5070 Ti and 5070 on the horizon, at least this is a taste of what’s to come.