AI-driven enhancements that make existing hardware work smarter
At Computex 2026, NVIDIA extended its AI-driven rendering ambitions to every GeForce RTX graphics card ever made, releasing DLSS 4.5 with a second-generation transformer-based ray reconstruction system. The move is less a technical announcement than a philosophical one — rather than reserving its most sophisticated visual tools for those who can afford the newest hardware, NVIDIA chose to democratize access, allowing a five-year-old GPU to reason about light the same way a flagship card does. With over a thousand games now participating in this ecosystem, the company is quietly arguing that the future of visual fidelity belongs not to raw computational force, but to machines that have learned to imagine what they cannot yet see.
- NVIDIA's DLSS 4.5 arrives with a second-generation transformer model that teaches GPUs to intelligently reconstruct missing ray-traced pixels rather than brute-force render them.
- The tension between cutting-edge visuals and accessible hardware has long frustrated players with older GPUs — this update collapses that divide by extending full ray reconstruction support to every RTX generation.
- Over 1,000 games and applications now carry the enhanced capability, with titles like Deliver Us The Moon joining the rollout and signaling broad developer adoption.
- The conspicuous absence of DLSS 5 at Computex suggests NVIDIA is deliberately deepening the current generation rather than racing to the next, giving developers a stable, reliable target.
- The trajectory is clear: AI inference is becoming the primary engine of visual progress, with neural reconstruction doing work that traditional rendering pipelines would find prohibitively expensive.
At Computex, NVIDIA unveiled DLSS 4.5, anchored by a second-generation ray reconstruction system built on transformer architecture — the same class of machine learning models that has reshaped fields far beyond graphics. The feature is now available across the entire GeForce RTX lineup, from the latest 50-series cards down to RTX 20-series GPUs that are several years old.
Ray reconstruction addresses a fundamental tension in modern rendering: ray tracing produces stunning, physically accurate lighting, but computing it at full resolution is enormously expensive. NVIDIA's AI system learns to predict what missing pixels should look like based on surrounding visual data, allowing games to run faster without sacrificing the quality players expect. The transformer-based approach improves on the first generation by drawing on more sophisticated pattern recognition during that reconstruction process.
The practical reach of the update is substantial. More than 1,000 games and applications now support DLSS 4.5, including newly added titles like Deliver Us The Moon. Crucially, a player running a five-year-old RTX 2080 can now access the same transformer-based reconstruction as someone with the newest hardware — performance will vary, but the capability itself is no longer gated by GPU generation.
This democratization is a deliberate strategic choice. By flattening access across the RTX ecosystem, NVIDIA lowers the barrier for developers who can now build toward a single, consistent feature set, and for players who no longer need to upgrade to participate. The decision to deepen DLSS 4.5 rather than announce DLSS 5 — which some had expected at Computex — reinforces the message: NVIDIA believes there is still significant value to extract from AI-driven rendering before the next major leap, and it intends to extract it.
NVIDIA announced an upgrade to its DLSS technology at Computex, rolling out version 4.5 with a second-generation ray reconstruction system built on transformer architecture. The new feature is now available across the company's entire lineup of GeForce RTX graphics cards, a significant expansion from previous limitations that confined advanced rendering tools to newer hardware.
Ray reconstruction is the process of filling in visual information that ray tracing—a technique that simulates how light bounces through a scene—would normally compute at full resolution. By using AI to intelligently reconstruct missing pixels, NVIDIA's system allows games to run faster while maintaining visual fidelity. The transformer-based approach represents a step forward from the first generation, using machine learning models trained to better predict what those missing pixels should look like based on surrounding data.
The practical impact is immediate. Over 1,000 games and applications now support DLSS 4.5 with the enhanced ray reconstruction capability. This includes titles like Deliver Us The Moon, which received DLSS and ray tracing support as part of the rollout. For players, this means a broader range of games can now leverage AI-driven rendering improvements, regardless of whether they own a cutting-edge RTX 50-series card or an older RTX 20-series GPU from years past.
The decision to make ray reconstruction available across all RTX generations is a strategic one. It democratizes access to NVIDIA's most advanced rendering features, lowering the barrier for both developers and players. A gamer with a five-year-old RTX 2080 can now access the same transformer-based ray reconstruction as someone with the latest hardware, though performance gains will naturally vary based on the GPU's compute power.
The announcement also signals where NVIDIA sees the future of gaming graphics heading. Rather than chasing raw polygon counts or texture resolution, the company is betting on AI-driven enhancements that make existing hardware work smarter. Ray reconstruction is one piece of this puzzle—using neural networks to do work that would otherwise require brute-force computation. As games become more visually complex and players demand higher frame rates at higher resolutions, these kinds of AI shortcuts become increasingly valuable.
Notably absent from the announcement was any word on DLSS 5, which some in the industry had anticipated might debut at Computex. Instead, NVIDIA chose to focus on deepening the capabilities of DLSS 4.5, suggesting the company sees room to extract more value from the current generation before moving to a major new version. For developers integrating DLSS into their pipelines, this means a stable target to work toward, with the transformer-based ray reconstruction now the standard feature set they can rely on across the RTX ecosystem.
Notable Quotes
Ray reconstruction is the process of filling in visual information that ray tracing would normally compute at full resolution using AI to intelligently reconstruct missing pixels— NVIDIA technical documentation
The Hearth Conversation Another angle on the story
Why does it matter that this works on all RTX cards now, not just the newest ones?
Because it means a developer building a game today doesn't have to choose between using the best rendering tech or supporting a wider audience. They can use transformer ray reconstruction and know it'll work on hardware from 2018 onward.
What's actually different about the transformer approach compared to the first generation?
The first version used simpler AI models to guess what pixels should look like. Transformers are more sophisticated—they can look at larger patterns in the image and make smarter predictions about what's missing. It's like the difference between filling in a puzzle by looking at adjacent pieces versus understanding the whole picture.
Does this mean ray tracing is finally practical for everyone?
More practical, yes. Ray tracing is still expensive computationally. What this does is let you run ray tracing at lower resolution and use AI to reconstruct the full image convincingly. It's a compromise, but a good one—you get the visual benefits without the full performance cost.
Why didn't NVIDIA announce DLSS 5 at the same time?
That's the interesting question. Either they're not ready yet, or they decided DLSS 4.5 still has room to grow. Announcing a major new version would make 4.5 feel obsolete. Instead, they're saying: this is the standard now, it works everywhere, and we're making it smarter.
What does this mean for game developers right now?
It simplifies their decision-making. They can implement DLSS 4.5 with ray reconstruction and know it'll reach a huge installed base. That's 1,000 games already doing it—that's momentum. New developers see that and think, why wouldn't we add this?