JVMag – DLSS, FSR, XeSS, PSSR. It’s what ? How does it work?

Who could have believed, more than ten years ago, that scaling, or upscalingwould one day be a workhorse in the world of technology. With ever more demanding games and ever crazier resolutions, graphics card manufacturers had to find a solution. Today, in this big ring, we find Nvidia with its DLSS, which incidentally got the ball rolling, closely followed by AMD and its FSR as well as Intel which brings up the rear with its XeSS. We also find the PSSR which is reserved for Playstation consoles.

If in broad terms these technologies have the same goal, it was important for me to give you a little overview of how they work and how they have developed. Here is a short summary of the different technologies.

DLSS (Deep Learning Super Sampling)

• Functioning : DLSS is based on the use of artificial intelligence. NVIDIA drives networks neuronal on high-resolution and low-resolution images so that the model learns to reconstruct missing details. To put it simply, the AI ​​will learn the scenes by heart in all resolutions to be able to restore them at the necessary time depending on the need. All this thanks to the Tensor Cores of RTX GPUs (from the RTX 20XX generation). Here the calculations are carried out in real time, which allows you to display an image almost identical to native resolution.

• Technical characteristics :
  • The latest version of DLSS, namely 3.0, also introduces Frame Generation, which creates intermediate images to improve fluidity. For example, as you can see in the image above, for an image calculated in FullHD (1080p), the card is capable of displaying two images in 4K thanks to AI.
  • Compatible only with NVIDIA RTX cards, which limits its accessibility. Worse still, Nvidia has the unfortunate tendency to offer the latest version of its tool only with the latest ranges of its cards.

FSR (FidelityFX Super Resolution)

• Functioning : Unlike DLSS, FSR does not require dedicated hardware. Indeed, it relies on scaling (FSR 1.0) or temporal (FSR 2.0 and 3.0) algorithms to reconstruct images using data from previous and current images. Its mode of operation is therefore completely different from DLSS, all the understanding work is carried out in real time by software. This openness and extensive compatibility which does not stop due to the hardware currently suffers from a slight loss of visual quality from which Nvidia no longer suffers. However, AMD is making great strides and we may see a small revolution in the coming weeks with the new generation of graphics cards.
• Technical characteristics
  • Open-source, which allows its implementation in a large number of games and on various GPUs, including NVIDIA and Intel.
  • Version 3.0 also introduces frame generation. Enough to further reduce the GPU load.

XeSS (Xe Super Sampling)

• Functioning : The XeSS uses different artificial intelligence models to reconstruct higher resolution images. The idea is to restore a clean image inspired by the previous image, but also by absorbing existing pixels. The technology is optimized for Intel GPUs with XMX (Xe Matrix Extensions) accelerators. It remains functional on other GPUs thanks to the use of shaders.
• Technical characteristics
  • Xe Matrix Extensions (XMX) hardware accelerators built into Intel Arc GPUs perform calculations matrices necessary for AI. These accelerators are comparable to Tensor Cores from NVIDIA. Support based on standard shaders allows XeSS to run on non-Intel GPUs. This flexibility ensures great compatibility, although quality and performance are optimal on Intel hardware.
• In terms of quality, the Intel XeSS is close to what Nvidia offers with its DLSS 2.0.

Intel is still young in the graphics card market. However, the manufacturer has more than one trick up its sleeve and risks doing poorly in the entry-level and mid-range segment in the years to come. It targets a more down-to-earth market even if it means being a generation behind its two major competitors.

PSSR (PlayStation Super Sampling Reconstruction)

• Functioning : PSSR is a technology developed specifically for Sony consoles. It uses a temporal reconstruction pipeline that analyzes motion and data from previous frames to produce sharper, smoother images. As we have mentioned in the past, it was important for the manufacturer to release a machine like the Pro to be able to study, work and improve its technology in order to prepare the future generation of machines. Unfortunately, Sony isn’t very talkative about its said technology. We just know that it works via 44 machine learning algorithms that take an image at a lower resolution and upscale it to a higher resolution. This is very similar to what AMD is offering with its FSR, the manufacturer which also produces chips for Playstation consoles.
• Technical characteristics :
  • Optimized for PS5 Pro’s modified RDNA 3 hardware.
  • Great integration with exclusive PlayStation game engines, maximizing their visual potential. Optimization will be the key to success for Playstation once again.

As you can see, there is no perfect technology at the moment. If we limit ourselves to the hardware, which offers the best quality, we lose the “universal” aspect with technology that must be validated in games. Conversely, in scenarios with free technology that works everywhere, we still find losses in overall quality.

Either way, the years when only CUDA Cores were important are over.