DLSS 4.5 testing shows Nvidia’s tech potential

While I remain generally skeptical of many AI applications, one has become an indispensable part of my daily gaming routine: Nvidia’s Deep Learning Super Sampling (DLSS). This technology leverages artificial intelligence to upscale in-game graphics, significantly reducing the workload on compatible graphics cards. The result is that gamers can achieve more impressive visual fidelity and higher frame rates without necessarily requiring the most expensive hardware. For flagship GPUs like the RTX 5090, it enables previously unthinkable feats like smooth 8K gaming. More importantly, it empowers mid-range and budget graphics cards to deliver performance that once demanded a premium price tag.

The core goal of the latest iteration, DLSS 4.5, is to synchronize your GPU’s frame rate with your monitor’s native refresh rate. A perfect match, such as 120 frames per second on a 120Hz display, creates a fluid, responsive experience free from disruptive screen tearing or stuttering. To accomplish this, DLSS 4.5 introduces up to six times frame generation and a new Dynamic Multi Frame Generation (DMFG) feature.

Frame Generation itself, first seen in DLSS 3.0, has been a topic of debate. It works by using AI to create additional frames between those rendered by the GPU, effectively doubling or tripling the perceived frame rate. My initial experience was lukewarm; in titles like Cyberpunk 2077, the generated frames sometimes introduced a soft, blurry quality. The introduction of Multi Frame Generation (MFG) with DLSS 4.0, allowing multiple AI frames per rendered frame, initially raised concerns that these artifacts would become more pronounced.

Instead, I found the opposite to be true. Refinements in the technology made the impact on graphical fidelity far less noticeable, while frame rates saw dramatic gains. A significant advantage of MFG is user control, allowing you to fine-tune the ratio of generated to rendered frames. This let me tailor the experience, for instance, capping Cyberpunk 2077 at a rock-solid 120fps on my 120Hz TV even when my RTX 5090 was capable of much higher numbers.

This evolution has made DLSS an incredibly compelling tool, particularly for gamers without flagship GPUs. Criticism has persisted around the concept of “fake frames,” with some arguing it could lead to poorer native game optimization. That argument has never held much weight for me. The distinction between rendered and AI-generated frames has become increasingly difficult to detect, and developers must still optimize for AMD, Intel, and console hardware regardless. In fact, DLSS’s success has spurred valuable competition, pushing rivals like AMD with FidelityFX Super Resolution (FSR) and Intel with XeSS to innovate.

The launch of DLSS 4.5, with its headline 6X MFG capability, did court controversy. If most displayed frames are AI-generated, would visual quality suffer? The introduction of Dynamic Multi Frame Generation directly addresses this. By automatically adjusting the number of generated frames to maintain a user-set target frame rate, DMFG ensures stability. During intense scenes, it generates more frames; when the action calms, it scales back. This intelligent management aims to deliver smooth performance without unnecessary sacrifices to image quality, automating the manual tuning I previously did.

My hands-on testing involved pairing a PNY RTX 5070, a capable mid-range GPU, with an MSI MPG 322UR QD-OLED X24 gaming monitor, a 4K display with a 240Hz refresh rate. This is a combination that would typically struggle. In Cyberpunk 2077 at 4K with high settings, the RTX 5070 averaged about 51fps natively. Enabling DLSS upscaling and 6X frame generation skyrocketed that average to 140fps, allowing me to even activate demanding path-traced lighting. The game looked and felt phenomenal on the OLED panel, with only minor ghosting and pixelation in reflections as occasional reminders of the heavy lifting being done.

Using DMFG to target the monitor’s 240Hz refresh rate produced more consistent performance and reduced ghosting. Setting a 120fps target for my TV allowed me to switch to a higher DLSS Quality preset while keeping path tracing on, averaging 130fps. While not flawless, with some UI glitches persisting, the feat of 4K path-traced gaming at high frame rates on an RTX 5070 is remarkable. Results varied by title, with Avowed running at a blistering 255fps average, while Hogwarts Legacy exhibited more distracting artifacts, likely a game-specific issue.

For competitive esports where latency is paramount, native rendering at lower resolutions remains the best choice. However, for the vast majority of gamers, DLSS 4.5’s benefits are transformative. It allows mid-range GPUs to fully utilize high-refresh-rate 4K monitors and makes cutting-edge effects like ray tracing accessible to a wider audience.

My concern now is that the solid progress of DLSS 4.5 risks being overshadowed by its announced successor. The preview of DLSS 5 has sparked significant backlash for a fundamental shift in focus. Instead of refining performance and image reconstruction, early examples suggest DLSS 5 may actively alter artistic intent, modifying lighting and even character designs in a manner critics compare to aggressive AI filters. While I will reserve final judgment, DLSS 5’s potential misstep would be causing gamers to overlook the genuine advancements of DLSS 4.5, which has taken meaningful steps toward democratizing high-fidelity gaming. For now, I hope Nvidia continues to refine the performance-driven path it has paved, rather than pursuing a generative AI direction that may compromise the developer’s vision.