Every couple of years, Nvidia unveils a powerhouse graphics card—an exorbitantly priced marvel that propels PC gaming into a new era. The Nvidia GeForce RTX 5090 is precisely that, though its approach to next-generation performance is decidedly unconventional. In many games, the performance boost over the RTX 4090 isn't as dramatic as expected, at least without DLSS Frame Generation. However, the next generation of Nvidia's DLSS, for both upscaling and frame generation, delivers breathtaking leaps in image quality and performance, surpassing typical generational improvements.
The RTX 5090's upgrade value hinges on your gaming habits: the games you play, your monitor's resolution and refresh rate, and your tolerance for AI-generated frames. For those gaming on anything less than a 4K, 240Hz display, this upgrade is likely unnecessary. But for high-end display owners, these AI-generated frames offer a glimpse into the future of gaming.
Nvidia GeForce RTX 5090 – Photos
RTX 5090 – Specs and Features
The Nvidia GeForce RTX 5090 leverages Blackwell, Nvidia's high-end architecture powering data centers and supercomputers behind popular AI models. This hints at the RTX 5090's strengths, but Nvidia didn't neglect the non-AI aspects.
Nvidia managed to increase Streaming Multiprocessors (SMs) within the same Graphics Processing Clusters (GPCs), resulting in a significant boost to CUDA cores—21,760, a 32% increase over the RTX 4090's 16,384. This raw increase in shader cores fuels much of the card's gaming performance.
Each SM boasts four Tensor Cores and one RT Core, mirroring its predecessor. This translates to 680 Tensor Cores and 170 RT cores, compared to the RTX 4090's 512 and 128, respectively. The 5th-generation Tensor Cores are optimized for AI performance, now supporting FP4 operations, reducing VRAM dependency for AI workloads.
This powerful silicon is paired with 32GB of GDDR7 VRAM, a generational leap from the RTX 4090's GDDR6X. GDDR7 offers improved speed and power efficiency, although the RTX 5090's staggering 575W power draw (a substantial increase over the 4090) suggests power efficiency wasn't the primary design goal.
The enhanced Tensor Cores' efficiency allowed Nvidia to shift the DLSS algorithm to a Transformer Neural Network (TNN) from a Convolutional Neural Network (CNN). While not directly impacting frame rates with DLSS enabled, Nvidia claims this improves image quality and reduces artifacts like ghosting.
Beyond the internal DLSS improvements, Nvidia introduces Multi-Frame Generation, an enhanced version of the Frame Gen technology from the RTX 4090. It's more efficient and smoother, generating multiple frames from each rendered image for a dramatic frame rate increase. Similar to its predecessor, it's best enabled when a decent frame rate is already achieved.
Purchasing Guide
The Nvidia GeForce RTX 5090 launched January 30th, starting at $1,999 (Founders Edition). Third-party cards will likely command significantly higher prices.
The Founders Edition
The RTX 5090's 575W power requirement (far exceeding the RTX 4090's 450W) necessitates robust cooling. While expecting a behemoth, Nvidia surprisingly created a smaller, dual-slot card with a dual-fan configuration.
During testing (including standard benchmarks and DLSS 4 with multi-frame generation), temperatures peaked around 86°C at 578W power consumption. While high (higher than the RTX 4090's 80°C), it remained below throttling thresholds.
Nvidia achieved this by centrally positioning a compact PCB, flanked by dual fans and a heatsink spanning the card's width. Air intake is from the bottom, expelled through the top, relying on the PC case's exhaust fans. Unlike previous generations, there are no exhaust vents under the rear output ports.
The design maintains a familiar aesthetic, echoing the RTX 4090's silver 'X' design and gunmetal-gray chassis. The 'GeForce RTX' logo features white LEDs.
The power connector, while similar to the 12VHPWR connector, is a new 12V-2x6 connector, claimed to be more efficient. It's angled towards the back for easier cable connection, appearing more secure.
An included 12V-2x6 adapter uses four 8-pin PCIe power connectors to supply the required 575W. This design allows compatibility with smaller PC cases, unlike previous generations, though third-party cards may be larger.
DLSS 4: Fake Frames?
Nvidia claimed up to 8x performance boosts. While not reaching that level, the RTX 5090 achieves incredibly high frame rates in demanding games, largely thanks to frame generation. While raw rasterization performance improves, the true next-gen benefit lies in AI-generated frames.
DLSS 4's 'Multi-Frame Generation' surpasses its DLSS 3 counterpart. A new AI Management Processor (AMP) core efficiently manages GPU workload distribution, traditionally handled by the CPU. The AMP's on-GPU location improves efficiency.
Nvidia states the AMP and 5th-generation Tensor Cores created a 40% faster frame generation model using 30% less memory. It generates 3 AI frames per rendered frame, mitigating latency with a Flip Metering algorithm for frame pacing. This algorithm is cited as the reason for incompatibility with RTX 4000 cards, as their CPU-based frame pacing introduces more latency.
This isn't a magic bullet; it's most effective with already acceptable frame rates (around 60fps without Frame Gen). Pairing it with DLSS upscaling maximizes performance.
At launch, DLSS 4 supported many games already using DLSS 3 Frame Generation. Testing revealed excellent results in Cyberpunk 2077 and Star Wars Outlaws (beta builds).
In Cyberpunk 2077 at 4K with Ray Tracing Overdrive and DLSS Performance, the RTX 5090 achieved 94fps, increasing to 162fps with DLSS 2x frame generation and 286fps with 4x frame generation (exceeding the display's capabilities). Star Wars Outlaws similarly reached around 300fps with DLSS 4, up from 120fps without frame generation.
Multi-Frame Generation proved effective, with minimal artifacts observed. While needing a high-end 4K display to fully benefit, the technology shows immense potential. However, widespread compatibility across the 75 claimed games remains to be seen.
RTX 5090 – Performance
The RTX 5090 is incredibly powerful, but testing revealed complexities. 3DMark showed generational improvements over the RTX 4090, but real-world game testing revealed CPU bottlenecks, even at 4K with a Ryzen 7 9800X3D. For many with high-end cards, the upgrade offers limited practical benefits, highlighting the need for future games to fully utilize its capabilities. DLSS 4 was disabled for comparative benchmarks.
In 3DMark, the RTX 5090 was up to 42% faster than the RTX 4090 (Speed Way: 14,399 vs 10,130; Port Royal: 36,946 vs 25,997). Compared to the RTX 3090, the performance jump was even more significant (2.5x).
Call of Duty Black Ops 6 demonstrated CPU bottlenecking at 4K Extreme settings (161fps vs RTX 4090's 146fps—a 10% difference). Similar results were seen in Cyberpunk 2077 (125fps vs 112fps—10% difference), with scaling worsening at lower resolutions.
Metro Exodus: Enhanced Edition (DLSS disabled for fair comparison) showed a 25% improvement over the RTX 4090 at 4K Extreme (95fps vs 76fps).
Red Dead Redemption 2 showed a minimal 6% improvement at 4K max settings with DLSS Performance (167fps vs 151fps).
Total War: Warhammer 3 (no ray tracing or upscaling) showed a 35% improvement (147fps vs 107fps), closer to the 3DMark results.
Assassin's Creed Mirage showed anomalous results initially, likely due to a driver bug.
Black Myth: Wukong showed a 20% improvement at 4K (104fps vs 84fps). Forza Horizon 5 showed negligible differences (216fps vs 210fps).
The RTX 5090, while the fastest consumer card, doesn't always deliver a generational leap over the RTX 4090 in current games. Its strength lies in its future-proofing with DLSS 4's AI-powered frame rate boosts. It's ideal for cutting-edge gamers willing to invest in an AI-driven gaming future; for others, the RTX 4090 remains a powerful option.
