The Nvidia GeForce RTX 5090: A Next-Gen Leap Fueled by AI

Nvidia's RTX 5090 arrives as the latest high-end graphics card, promising a new era in PC gaming. However, its performance gains are less straightforward than previous generational leaps. While raw performance improvements over the RTX 4090 are noticeable, they're not as dramatic as expected without considering DLSS Frame Generation. The real game-changer is DLSS 4, which significantly enhances both image quality and frame rates.

The RTX 5090's upgrade value depends heavily on your gaming setup and preferences. For those with displays below 4K 240Hz, the upgrade may not be justifiable. But for high-end display owners, DLSS 4's AI-generated frames offer a glimpse into the future of gaming performance.

Nvidia GeForce RTX 5090 – Image Gallery

5 Images

RTX 5090 – Specifications and Features

Built on Nvidia's Blackwell architecture (also powering leading AI models), the RTX 5090 excels in AI-related tasks. However, its non-AI capabilities are equally impressive. The card boasts a 32% increase in shader cores (21,760 vs. 16,384 in the RTX 4090) thanks to increased Streaming Multiprocessors (SMs) within the same Graphics Processing Clusters (GPCs).

Each SM includes four Tensor Cores and one RT Core, resulting in 680 Tensor Cores and 170 RT cores (compared to 512 and 128, respectively, in the RTX 4090). The 5th-generation Tensor Cores support FP4 operations, reducing VRAM dependency for AI workloads.

The 32GB GDDR7 VRAM represents a generational leap from the RTX 4090's GDDR6X memory, offering improved speed and power efficiency. Despite this, the RTX 5090's power consumption is a substantial 575W, significantly higher than the RTX 4090's 450W.

DLSS 4's shift to a Transformer Neural Network (TNN) from a Convolutional Neural Network (CNN) aims to improve image quality and reduce artifacts. Multi-Frame Generation, an evolution of DLSS 3's Frame Generation, generates multiple frames from each rendered image, drastically boosting frame rates, but ideally only when a decent baseline frame rate is already achieved.

Purchasing Guide

The Nvidia GeForce RTX 5090 launched January 30th, with a starting price of $1,999 for the Founders Edition. Third-party cards will likely command higher prices.

The Founders Edition

Despite requiring 575W, the RTX 5090 Founders Edition surprisingly features a smaller, dual-slot design with dual fans. During testing, temperatures peaked around 86°C at 578W power consumption, remaining within safe operating limits. This compact design is achieved through a centrally located PCB and a heatsink spanning the card's width, with fans drawing air from below and expelling it through the top.

The design aesthetics are consistent with previous generations, featuring a silver 'X' design and a 'GeForce RTX' logo with white LEDs. A new, angled 12V-2x6 power connector (with an included adapter for four 8-pin PCIe connectors) is included.

This compact design makes it suitable for smaller PC builds, unlike previous generations. However, third-party versions are expected to be larger.

DLSS 4: AI-Generated Frames

While the RTX 5090 offers raw performance improvements, its true next-gen advantage lies in DLSS 4's Multi-Frame Generation. This utilizes a new AI Management Processor (AMP) core, offloading frame generation tasks from the CPU to the GPU for increased efficiency.

The AMP and 5th-generation Tensor Cores enable a 40% faster frame generation model using 30% less memory, generating three AI frames per rendered frame. A Flip Metering algorithm minimizes input lag. This technology is GPU-dependent and won't function on RTX 4000 cards due to CPU reliance in previous frame generation implementations.

Optimal performance requires a good baseline frame rate (around 60fps without Frame Gen). DLSS 4 was tested in Cyberpunk 2077 and Star Wars Outlaws (beta builds), demonstrating impressive frame rate increases (e.g., up to 286fps in Cyberpunk 2077 at 4K with Ray Tracing Overdrive and DLSS 4x frame generation). While some minor artifacts were observed, the overall performance was remarkably smooth.

RTX 5090 – Performance Benchmarks

Testing revealed a generational leap in raw performance in 3DMark, but real-world gaming showed CPU bottlenecks even at 4K with a Ryzen 7 9800X3D CPU. In many games, the performance gains over the RTX 4090 were modest (around 10%), highlighting the current limitations of game optimization for this level of GPU power. However, significant improvements were observed compared to the RTX 3090.

Test System: AMD Ryzen 7 9800X3D, Asus ROG Crosshair X870E Hero, 32GB G.Skill Trident Z5 Neo @ 6,000MHz, 4TB Samsung 990 Pro, Asus ROG Ryujin III 360

3DMark showed a 42% performance uplift over the RTX 4090. Gaming benchmarks varied significantly. Call of Duty Black Ops 6 and Cyberpunk 2077 showed only around a 10% increase at 4K. Metro Exodus: Enhanced Edition (without DLSS) showed a 25% improvement. Red Dead Redemption 2 showed a minimal 6% improvement. Total War: Warhammer 3 showed a 35% improvement, reflecting raw rasterization performance. Assassin's Creed Mirage results were anomalous, likely due to driver issues. Black Myth: Wukong showed a 20% improvement. Forza Horizon 5 showed negligible differences.

Nvidia GeForce RTX 5090 – Benchmark Charts

14 Images

Conclusion

The RTX 5090 is undeniably the fastest consumer graphics card available, but its performance gains over the RTX 4090 are less dramatic in many current games due to CPU bottlenecks. Its future-proofing lies in DLSS 4's AI-powered frame generation, making it ideal for high-end display users willing to invest in AI-driven gaming performance. For most others, the RTX 4090 remains a powerful and cost-effective option.

Which new graphics card are you planning to buy?

Nvidia GeForce RTX 5090Nvidia GeForce RTX 5080Nvidia GeForce RTX 5070 TiNvidia GeForce RTX 5070I'm waiting to see AMD's new cardsNot planning on upgrading this generationAnswerSee Results