When it first debuted, DLSS was a promising technology. While far from perfect, it was a groundbreaking feature that helped RTX cards deliver more performance while handling demanding features like Ray Tracing. Fast-forward 7 years and 5 iterations, and DLSS is in a completely different place: a real, improved way to boost performance on older hardware, even though Nvidia’s intent has strayed too far.

Behind this tech, AMD and then Intel have struggled to match DLSS’s feature set, but after all these years, they’ve finally caught up. FSR and XeSS offer the best parts of DLSS’s stack, even though they’ve gone far beyond simply competing with DLSS.

FSR and XeSS have finally closed the upscaling gap

AMD needed four tries, and Intel needed a wider net

DLSS started as a game-specific technology, but with the debut of DLSS 2, it became a generalized AI reconstruction system applicable to a wider range of games.

While AMD’s FSR debuted in 2021, two years after DLSS, it was really just AMD playing catch-up with Nvidia. The first iteration was underwhelming because it was only a spatial, algorithm-based upscaler with no machine learning.

FSR 2, however, became a temporal upscaler using information from prior frames, roughly matching DLSS 2, though it arrived later and delivered lower image quality.

With FSR 4, AMD finally implemented machine learning, introducing neural rendering that more closely matches DLSS. FSR 4.1 was just AMD fine-tuning the ML-based neural rendering approach with better dynamic lighting and temporal stability.

Similarly, Intel wasn’t about to let itself fall behind AMD or Nvidia, so it brought its own upscaling tech, XeSS, with the release of the first Arc GPUs in 2022. Although XeSS started with AI upscaling and wasn’t algorithm-based like the first FSR implementations, it arrived too late.

In contrast, XeSS never really became a solid option because it was adopted later than Nvidia and AMD, the image quality wasn’t great, and it was more demanding than the other options. Moreover, game support for XeSS is quite limited, even though it works across all GPUs (higher versions explicitly require Intel’s XMX cores in Arc GPUs).

The Frame Generation race that Nvidia still leads

FSR and XeSS are closing in, and AMD might be cooking up something even better

DLSS 3 introduced Frame Generation, a feature that adds AI-generated frames between rendered frames to boost overall FPS, smooth motion, and become Nvidia’s real selling point. DLSS 4 took it a level further with Multi-Frame Generation, a frame-generation multiplier that, instead of generating a single AI-generated frame, generates up to 5 extra frames (with DLSS 4.5) to boost FPS further.

Nvidia practically changed how the industry measures performance, and raw figures don’t really matter anymore. While public opinion on AI-generated frames is still split, Frame Generation eases the weight of Ray Tracing and Path Tracing.

AMD and XeSS once again fell behind Nvidia, and while both tech giants have their own versions of Frame Generation, FSR 4.1 can only do 2X Frame Generation, and Intel can go up to 4X with XeSS 3. However, AMD is testing FSR Multi Frame Generation of up to 8X in newer drivers for RDNA 4 GPUs (via VideoCardz), and that might arrive sooner than expected and give it an edge over DLSS 4.5.

AMD denoisers finally have the answer to the best part about DLSS’s stack

Ray Regeneration and Radiance Caching are promising technologies for Radeon GPUs

Beyond upscaling and Frame Generation, Nvidia has added more features to the DLSS stack with newer iterations. DLSS 3.5 introduced Ray Reconstruction, which replaced traditional denoisers that clean up grain, speckles, and visual static in ray-traced samples.

Ray Reconstruction works with reflections, shadows, global illumination, fine lighting details, and temporal stability and vastly improves quality for upscaled images with little to no performance downside.

AMD’s answer was Ray Regeneration, a ray-tracing denoiser meant to be the equivalent of Nvidia’s Ray Reconstruction, but it works differently by generating new replacement rays to produce more accurate lighting, reflections, shadows, and illumination.

On the other hand, Radiance Caching predicts how light will bounce around a scene instead of calculating every ray from scratch. It is AMD’s answer to Nvidia’s Neural Radiance Cache, which serves the same purpose on the DLSS side.

These feature sets are packaged together as FSR Redstone and are exclusive to newer RDNA 4 GPUs.

Where Nvidia still wins and where it doesn’t

DLSS 5 is where Nvidia leads, but it’s not about performance anymore

Nvidia has always been one step ahead of AMD and Intel with DLSS, mainly because they pioneered the upscaling technology, but DLSS’s evolution hasn’t been very steady.

DLSS 5 is Nvidia losing the plot, undermining the very reason DLSS exists in the first place, since it uses generative AI to repaint lighting, materials, and textures in a game.

However, DLSS 5 isn’t what anyone had expected and does far more than what Nvidia claims. It completely alters a video game’s intended art style. So far, it is a photorealism filter with a steep performance cost; it uses AI features on the newer RTX 50 Series cards to reduce overall consumption, but the same AI features are resource-intensive, so it works counterintuitively.

Right now, no DLSS-exclusive feature makes a performance difference that AMD or Intel doesn’t offer. Both FSR and XeSS have multi-Frame Generation, and AMD’s FSR Redstone has Ray Regeneration and Radiance Caching to match Nvidia’s Ray Reconstruction and Neural Radiance Caching. However, Intel has nothing to compete with in this department.

If you put FSR 4.1 against DLSS 4.5, both have their own strengths. DLSS 4.5 delivers supreme image quality, but it falls short on performance; AMD has the edge with FSR 4.1, which produces a softer, more natural look by comparison, but they’ve now come on par.

AMD is already lining up its answer to DLSS 5

Beyond software, AMD has finally gained an edge over Nvidia with the latest RDNA 4 GPUs, and now they’re gearing up to release the RDNA 5/UDNA lineup next year, which could possibly mark their return to flagship GPUs.

Leaks suggest Team Red is also working on a new neural rendering tech that takes the same approach as DLSS 5 (via VideoCardz), enhancing lighting, and it could potentially have greater access to 3D rendering primitives to output a better image.

With the best parts of the DLSS stack, i.e., Super Resolution and Frame Generation, already in FSR and XeSS (denoiser only in FSR), and finally, the upscaled image quality doesn’t cost much performance, it will be exciting to see what AMD and Intel will have to offer next against DLSS 5.