I tested DLSS on a mid-range GPU, and native resolution suddenly felt pointless
Nvidia’s push for AI has pervaded every aspect of gaming. Over the years, I’ve seen the pitch change from it being an “acceptable trade-off” against performance uplift to the bold claim that DLSS now matches or beats native resolution and hands you a stack of free frames at it. It’s always been the kind of claim that’s easy to repeat and harder to verify, so I decided to stop taking it on faith. With my RTX 4070 Ti Super and a 1440p OLED display, I ran three visually demanding AAA games through native rendering and every DLSS preset from Quality down to Ultra Performance, comparing them in the same scenes. I must confess, I expected to walk away from this experiment validating my hypothesis that native is still king, but instead, by the end, I found myself struggling to justify why I had been rendering my games the harder way, all while losing some free performance along the way.
Related
DLSS 4.5 is expensive, but it’s the best thing to happen to my aging RTX PC
The upside is remarkable, but the cost isn’t equal for all.
Spider-Man 2 made the strongest case against native
The city looked the same, but the swinging felt easier
I started the experiment with Marvel’s Spider-Man 2 for the simple reason that it’s the kind of game that makes it easy to call out any deficiencies in any upscaler. Manhattan stretches out all the way over to the horizon in the distance, and every cubic meter is packed with skyscrapers, bridges, traffic, pedestrians, and just enough fine geometry to make reconstruction algorithms break. Perched on top of the Empire State Building, I switched from native to DLSS-Quality preset while running DLSS 4.5 Preset M and watched the frame rate climb substantially in the same scene, while the 4070 Ti S ran cooler and pulled less power to do it. In terms of visual quality, there were almost no legible sacrifices. The lattice of the Avengers Tower in the distance, bridges spanning the river and the grid of windows all held together perfectly well under reconstruction. At Quality preset, hunting for any differences at all feels like a pointless chore with no payoff. On the other hand, dropping to Balanced and Performance presets tends to soften the fine detail to push some extra frames that I don’t particularly care for.
Preset
FPS
1% low
GPU %
Temperature
Power
PC latency
Native
54
40
99%
80°C
248W
51.2 ms
DLSS Quality
81
68
85%
75°C
223W
33.4 ms
DLSS Balanced
81
66
73%
74°C
205W
30.5 ms
DLSS Performance
82
71
68%
68°C
174W
31.4 ms
DLSS Ultra Performance
83
62
53%
67°C
150W
29.4 ms
DLSS 4.5 makes the Wizarding World more immersive
Hogwarts Legacy is where I started seriously doubting my assumptions
At this point in the test, Spider-Man 2 had already made a rather decisive frame rate and efficiency case for Nvidia’s second-gen transformer-based upscaler. The next title I tested had a reputation for traversal hitches and micro-stutters that could break immersion in a game world purpose-built for an escape from reality. For this run, I maxed out every possible graphical preset (as in all the other tests) and discovered that the DLSS Quality preset both lifted the frame rate in scene and tightened the frame consistency at the same time, acting as the de-facto antidote to the hitching problem the game shipped with. The familiar stutters became far less prominent. This is something average frame rate figures won’t convey well. A game can theoretically report a perfectly respectable average and yet feel inconsistent if frame times are all over the place. Native rendering on Hogwarts Legacy suffers from the worst of this problem, whereas toggling DLSS Quality delivered an experience that felt just right. Visually, I wasn’t left wanting for more either. Similar to the verdict on Spider-Man 2, stepping down to Balanced or Performance fetched more frames, but not in a way that changed how the game felt. Perhaps the effects would be more pronounced for anyone who prefers to game in “hard” mode where the added fluidity matters. It’s certainly one of the games where DLSS on Quality feels better than native.
Preset
FPS
1% low
GPU %
Temperature
Power
PC latency
Native
64
35
92%
76°C
247W
58 ms
DLSS Quality
85
76
85%
72°C
217W
51 ms
DLSS Balanced
86
76
62%
65°C
217W
51 ms
DLSS Performance
86
68
75%
71°C
195W
52.9 ms
DLSS Ultra Performance
89
79
87%
76°C
159W
49.7ms
Forza Horizon 6 showed where reconstruction gets harder
Yet the difference from native was indistinguishable
Forza Horizon 6 was the one test that I found myself looking forward to, especially since the alpine scenery, bright daylight and particle effects that make up the atmosphere of the game stand to benefit the most from Nvidia’s upgraded second-gen transformer. Nvidia specifically markets DLSS 4.5 on the basis that it improves temporal stability in games heavily reliant on smooth motion, and I couldn’t think of a title better suited for this test than the most beloved racer in the gaming industry at the moment.
At the Quality preset, the reconstruction simply left no room for complaint. The snow retained its texture, and the orange safety fencing lining on the off-road track next to the danger sign stayed crisp. Distant tree lines also seemed to hold their definition against the mountains. I would argue that it’s this specific instance that convinced me that I would certainly fail a blind test between native rendering and DLSS Quality, even after years of insisting that native rendering was the benchmark every upscaler should aspire to.
Preset
FPS
1% low
GPU %
Temperature
Power
PC latency
Native
202
166
98%
72°C
217W
47 ms
DLSS Quality
205
161
98%
76°C
234W
52 ms
DLSS Balanced
209
163
98%
72°C
215W
45 ms
DLSS Performance
225
166
98%
71°C
204W
41 ms
DLSS Ultra Performance
252
175
97%
69°C
189W
38 ms
Related
I tried running games at native resolution again, but DLSS 4.5 changed my mind
Even pixel peepers would struggle to justify native rendering at this point
Native resolution isn’t dead, but it’s no longer the default either
DLSS looking and “feeling” better than native rendering used to be an edge case, and often the exception you’d point to in a favorable game that was optimized well for it. Across these three tests, it’s starting to look like the rule. At Quality, on a mid-range card from two years ago, I got performance I could measure for the price of a compromise I struggled to see. That’s enough to make a decision for me.
Diterbitkan : 2026-07-10 23:00:00
sumber : www.xda-developers.com
