The recent launch of Nvidia’s DLSS 4.5 has sparked mixed reactions among users. While it brings notable improvements in image quality, it also has a noticeable impact on performance — especially on older hardware.
DLSS 4.5 introduces the new Preset M, a far more advanced model designed to reduce ghosting and enhance image stability. However, this improvement comes at a cost: it is around five times more computationally demanding than previous versions. On the latest RTX 50 and RTX 40 series, this increased demand is almost negligible — only a 2–3% performance drop — thanks to their native FP8 hardware acceleration. Unfortunately, older GPU architectures aren’t so lucky.
Problems for Ampere and Turing Users
The RTX 30 (Ampere) and RTX 20 (Turing) series lack native support for FP8 precision. As a result, early tests in games like Cyberpunk 2077 show performance drops between 12% and 20% when using DLSS 4.5 compared to DLSS 4.0. In some cases, such as with the RTX 2060, performance can even fall below the levels achieved with traditional anti-aliasing (TAA), making upscaling practically useless.
Apart from the loss in frames per second, DLSS 4.5 also consumes significantly more VRAM. Reports indicate it can require up to 87–103% more video memory on older GPUs, causing stuttering issues on models with just 6GB or 8GB of VRAM.
Nvidia has made it clear that DLSS 4.5 focuses on improving visual quality rather than performance. For RTX 50 users, this upgrade is a welcome enhancement that comes at almost no cost. However, for those using RTX 20 or 30 cards, the recommendation is to stick with Preset K (DLSS 4.0), since the performance trade-off may not justify the modest improvement in image sharpness.
