sortiwa
Something to look forward to: According to critics, the PlayStation 5 Pro’s new machine learning implementation delivers noticeable improvements over previous console upscaling methods but still leaves room for growth. Thanks to Sony’s partnership with AMD, which recently made significant advancements with FSR4, further improvements could begin materializing as early as 2026.
PlayStation console architect Mark Cerny recently told Eurogamer that Sony plans to integrate the neural network behind AMD’s FSR4 upscaler into the PlayStation 5 Pro’s PSSR technique starting in 2026. This revelation comes just a day after AMD confirmed its collaboration with Sony on FSR4.
AMD’s newly launched Radeon RX 9000 graphics cards introduced the company’s first machine learning-based upscaler, FSR4. Our analysis shows that this new method delivers significantly better image quality than AMD’s previous spatial and temporal upscaling techniques, which are now widely used in console games.
The PlayStation 5 Pro incorporates elements of the Radeon RX 9000’s RDNA 4 architecture for its machine learning-based PlayStation Spectral Super Resolution (PSSR). However, its results don’t yet match the quality of AMD’s FSR4 or Nvidia’s DLSS. Cerny explained that PSSR will more closely resemble FSR4 starting next year, as Sony collaborates with developers to integrate AMD’s neural network.
This transition will take time, as RDNA 4 and the PlayStation 5 Pro are not identical. The first RDNA 4 GPUs – the RX 9070 and RX 9070 XT – belong to a higher performance class than the PS5 Pro, which also lacks support for sparsity. Sparsity is a key technique that helps RDNA 4 and Nvidia hardware optimize upscaling workloads. However, Cerny believes the PS5 Pro can still achieve strong performance without it.
Cerny describes the PlayStation 5 Pro as an early step in AMD and Sony’s collaboration to refine machine learning-based upscaling for future products, likely including the PlayStation 6. Sony benefits from AMD’s advancements in AI, while AMD gains from Sony’s expertise in game optimization.
AMD has provided the architecture for both the PlayStation 4 and PlayStation 5, and the PlayStation 6 is expected to continue this trend. Recent reports suggest that Sony’s next-generation console could arrive in 2027 or 2028, featuring a Zen 6 CPU and a GPU based on RDNA’s successor, reportedly named UDNA.
Meanwhile, the Nintendo Switch 2 – expected to launch in the first half of this year – could provide another compelling example of ML-based upscaling on consoles. Utilizing Nvidia’s Ampere architecture from the RTX 30 series, the upcoming device will be the first handheld to employ DLSS, or possibly a more lightweight variation of it.
Source link