Nvidia DLSS 3.0 and 3.5 impressions — Lights, tracing, action!

To someone not familiar with the technology, DLSS feels like magic. Just turn it on in the settings and boom, your game both looks and runs better. Version 3.5 of NVIDIA’s Deep Learning Super Sampling has premiered alongside the release of Cyberpunk 2077: Phantom Liberty. So let’s put it through its paces and see what kind of difference it really makes along with version 3.0.

Currently, only two games support DLSS 3.5, being Cyberpunk 2077 and Portal RTX, with Alan Wake II joining that list later this month. These are perfect examples, however, as both struggle to be playable even on an RTX 4060 without DLSS. Without the technology, Cyberpunk runs at a whopping… 2 frames per second with everything maxed out. In the same area, with settings maxed out and DLSS on Balanced mode, the game runs at a buttery smooth 60 fps. The enhanced version of Portal yields somewhat similar results, given that it was literally made to show off ray tracing with DLSS. Without the technology, I get about 20 fps in early test chambers, but with it on I can play at 80.

This large gap even with ray tracing on is thanks to the Path Tracing technology included in version 3.5. You can read all about the tech on NVIDIA’s website here, or why our resident tech head and Editor in Chief Ron Burke loves DLSS here, but for laymen like me here’s a simplified explanation. Computing ray tracing for every single pixel on the screen in relation to every other pixel is super resource intensive, so most games don’t do that. Path Tracing takes the data from ray tracing, looks at the environment currently on screen, and calculates how the light would bounce off of the environment. This leads to more accurate reflections and lights. The best example is one NVIDIA provides themselves, involving headlights on cars in Cyberpunk, which you can see above. Without path tracing, the headlights move through and under your car, but with it on you get a more realistic flood straight ahead. It’s something you only notice is missing in comparison, but still very neat regardless.

Plenty of other games support DLSS 3.0, and while I will miss path tracing, let’s check them out. First off, Hitman 3. Using the Dartmoor benchmark, which is laughably intense compared to anything you’ll actually encounter in the game, DLSS on reaches an average of 30 fps while off goes down to 29. Not a big difference, so let’s test again in-game. The starting area of Miami is perfect for this, since it has a huge crowd of people lining up to enter the race track. With the game installed on my hard drive instead of an SSD for this test, the initial load was stuttery, but with DLSS on it achieved around 50 fps on average after calming down. Swapping DLSS off hovered at about 40 fps. So an increase of 10 or so in one of the game’s most crowded areas, not too shabby.

Marvel’s Spider-Man: Miles Morales is another more minor increase. Swinging around NYC without DLSS still nets an impressive mid 20s to 30s, while turning on Balanced mode cranks that up to be in the range of 45 to 60. Ultra Performance gets a more stable 50 to 60, though the quality of effects takes a noticeable hit. This is one case where I would rather stick to console, however. No shade on Nixxes, their PC ports are amazing, but Insomniac is in the business of leveraging PlayStation hardware to its limits. Still, put the settings on anything but maximum and you’re bound to see greater visual quality and performance than the PS5 is capable of outputting.

In my testing, I ran into a few weird examples, namely with Hogwarts Legacy (trans rights, by the way) and Star Wars Jedi: Survivor. In HogLeg, starting the game with DLSS enabled would net around 30 fps in Hogsmeade, while turning it off in-game would go down to 20. Starting the game again without DLSS would yield the same result. However, enabling it again without restarting would have the game remain around 20. This is a more intense example framerate wise, but a similar issue happened wherever I tested. No idea what restarting the game has to do with the technology, because every other game can turn it on and off no problem. Jedi: Survivor fared much better, though we’ll likewise use an intense area for testing. The landing combat area of Coruscant plays at about 30 to 40 fps normally, but with DLSS it ranges from 70 to 80. That’s not at all weird, but my initial test during a boss fight firmly stuck to 80 fps with or without DLSS. It’s worth enabling to keep performance consistent, but it does show that without optimization DLSS isn’t a blanket solution for every game’s performance problems.

When it works well, DLSS can feel like a miracle of technology. As games get more and more visually sophisticated, upscaling is becoming a necessity to even run on bleeding edge hardware. I am in two minds on this. On one hand, it lets me play gorgeous games at frame rates I never could have dreamed of when I started PC gaming on an old Acer Notebook of all things. On the other, it shouldn’t be relied on to fix every performance problem a game can have. Hogwarts Legacy and Jedi: Survivor both still experience issues even with DLSS. Jedi has the excuse that it didn’t ship with DLSS, only AMD’s comparable FSR technology, but neither company should be paying off games to exclude the other’s tech. Even if I find DLSS superior in every way, it’s only supported by NVIDIA’s graphics cards whereas FSR will work on anything. Again, the technology is rad and I hope to see it come to consoles such as the successor to the Switch and the next iteration of the Steam Deck (which already has FSR), but ideally it should make everyone’s experience better, not only some.

See below for our list of partners and affiliates: