NVIDIA DLSS 4.5 test on RTX 50/40/30 and 20 series shows 12% performance loss on Ampe

[harlan4096]

NVIDIA DLSS 4.5 performance impact measured



Two independent reviews from ComputerBase and Hardware Unboxed offer the first detailed look at DLSS 4.5 Super Resolution performance across multiple GPU generations. Both agree on one point: while image quality improves, older GeForce cards take a noticeable performance hit.

According to ComputerBase, the new “Model M” preset in DLSS 4.5 improves several artifacts from DLSS 4, such as noisy shadows, water rendering, and flickering vegetation. The result is cleaner visuals and better temporal stability, though at a small cost in speed. On RTX 5070 Ti and RTX 4080 Super, performance drops by roughly 4 to 5%, while RTX 3090 Ti and RTX 2080 Ti lose about 12% on average. These drops are linked to the lack of FP8 acceleration on older architectures, which DLSS 4.5 now uses extensively.


Source: ComputerBase

Hardware Unboxed focused on raw frame-rate data at 1440p using the quality upscaling mode. The channel found a similar pattern: DLSS 4.5 is about 9% slower than DLSS 4 on mid-range Blackwell GPUs like the RTX 5070, and as much as 20–30% slower on Ampere and Turing models. RTX 40 and RTX 50 cards retain acceptable frame-rate gains over native rendering, but the older series often see reduced or even negative scaling.

Continue Reading...