So while NBA 2K27 is officially the first game to launch with Nvidia’s newly launched DLSS 5, you’ve probably already seen it pop up in hundreds of more games than that. And I’ll say it again just like I said it before — in photorealistic games, it’s fantastic. But If you venture out of that into anything more stylized, it’s going to look weird.
That’s why it makes the world of sense in 2K27, but during my hands-on testing with my Asus ROG Zephyrus G14 (armed with an RTX 5070 Ti), I noticed something bizarre. The moment I entered a menu or triggered a replay, it suddenly felt like it was trudging through molasses.
I dug into the raw telemetry data to figure out what was going on, and what I found completely changed my understanding of Nvidia’s “3D-guided neural rendering.” That intensive new graphics pipeline of DLSS 5 may indeed be a graphical revolution — but on my humble laptop (and lower end desktops), I get the sense Frame Generation is a requirement to make it playable.
Latest Videos FromTom’s Guide
The laptop I’m playing on
|
CPU |
Intel Core Ultra 9 386H |
|
GPU |
RTX 5070 Ti (115 watts) |
|
RAM |
32GB LPDDR5X |
|
SSD |
1TB PCIe Gen 4 |
I’m seeing a lot of people test it on massive desktop rigs. I’ll of course get round to doing this when I’m not on my honeymoon (my wife will kill me if I started benchmarking DLSS 5 in Cancun), but first I wanted to show you some numbers on something that isn’t really being covered here — my Asus ROG Zephyrus G14 gaming laptop.
Nvidia’s been talking about how the model running this mode of neural rendering has been made super efficient over the past few months. In fact, when it first debuted, it ran on a separate RTX 5090 desktop GPU; whereas now, it’s able to run on a single card with the game.
So let’s give it a go in a notebook with a lower power level running through those graphics.
By the numbers
To figure out why the game was suddenly stuttering, I ran the telemetry on three different game states with DLSS 5 turned on and off. And as you can see, while the gameplay numbers are very close matching, elsewhere, the difference is pretty significant.
|
Scenario |
Setting |
Avg FPS |
1% Low FPS |
PC Latency (ms) |
GPU Util (%) |
|
Menu screen |
DLSS 5 Off |
109.8 |
85.3 |
35.8 |
91.5 |
| Row 2 – Cell 0 |
DLSS 5 On |
30.4 |
27.9 |
102.1 |
96.9 |
|
Gameplay |
DLSS 5 Off |
115.8 |
19.8 |
79.8 |
41 |
| Row 4 – Cell 0 |
DLSS 5 On |
115.2 |
19.8 |
111.2 |
86 |
|
Replay |
DLSS 5 Off |
92.6 |
57 |
56.6 |
92.9 |
| Row 6 – Cell 0 |
DLSS 5 On |
26.9 |
16.2 |
128.8 |
97.7 |
In fact, this kind of lines up with PC Gamer’s report of a 50-60% drop in performance. So what’s going on here? Let’s talk about it.
The heavy cost of the transformer
So why does turning off the new tech triple frame rate in menus and replays? Because DLSS 5 is not just an upscale filter. It is a highly complex transformer model that interacts with the on-screen geometry and injects neural rendering.
Speaking to Nvidia, it’s clear that this AI pipeline is a heavy one, and the answer to why you don’t see that same impact in gameplay is a simple one: Frame Generation.
During live gameplay, I engaged the 6x Frame Generation multiplier to take that slower neurally rendered base and fill the gaps. However, developers typically disable frame gen during static menus, UI-heavy pause screens or camera-cut replays.
You can see how much this first iteration of DLSS 5 is sweating the system by looking at a couple of factors in the gameplay. Yes, the framerate is barely touched, but the 1% Lows takes a dip and the latency is upped quite drastically, as well as the GPU utilization spiking.
In a controller-centric game like NBA 2K27, that’s not necessarily a noticeable hit, but I can imagine that when this in its current form is moved to games that require faster inputs with a keyboard and mouse, it could become more apparent.
Take away the prop of frame generation in menus, and you see the DLSS 5 penalty more realistically on an RTX 5070 Ti laptop GPU.
But it looks incredible
Like I said in my first impressions, when faced with photorealistic graphics, DLSS 5 is undeniably impressive. There is one very specific takeaway when I looked at these “before and after” pairs of screen captures: the core game look is preserved impressively.
DLSS 5 does a flawless job of preserving the game’s visual integrity. The shots of Jalen Brunson and Victor Wembanyama show off the micro-details on the skin and the way the sweat catches the harsh stadium lighting. The upscaler isn’t adding any aggressive smoothing or plastic-looking artifacts — it’s remarkably lifelike.
Then there’s the fabric and material textures. Brunson’s Knicks jersey and the fabric of the couch in the cutscene get an upgrade to show off the stitching more with high-resolution clarity, and without any noticeable AI-induced blurring.
The lighting reacts to on-screen characters more realistically, there isn’t any noticeable AI “fizzing” or ghosting around finer textures like hair — something upscalers have had problems with in the past.
Is this a frame-generated future?
DLSS 5 integration proved to me that the uncanny valley is dead, but it also proves that native frame rates are on their way out with it.
Nvidia themselves keep saying that Moore’s law is dead, and getting crafty with AI is what will bridge that generational gap. That threshold has been crossed — photorealism is getting too heavy for high-end silicon to render natively.
And DLSS 5 proves that future of photorealism isn’t just about rendering better pixels; it’s about accepting that a good chunk of the frames you see on your screen aren’t technically going to be “real” at all.
We’ve been coming to this moment in the PC gaming space for a while now, and it’s also happening on consoles too.
And personally, I have no problem with it if it looks this good — provided it’s used the right way (such as in photorealistic games).
Follow Tom’s Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds.