RTX Mega Geometry
Nvidia recently announced a slew of new hardware and software at the Consumer Electronics Show 2025, and while a lot of people are discussing the new Blackwell 50-series graphics cards and their exclusive Multi-Frame Generation feature, I’ve been interested in some of the free updates coming to all RTX GPUs back to 2018. There’s the across-the-board update to DLSS in the form of a new AI “transformer” model as a heavier but nicer-looking alternative to the older CNN (Convolutional Neural Network) model, but one that’s promised to be more of a flat upgrade is RTX Mega Geometry. Billed as an RTX-exclusive optimisation to games using mesh shaders – a new approach to rendering the various geometry of a game that’s best known by its UE5 implementation, Nanite – Mega Geometry promises to improve performance in games using both ray tracing and mesh shaders by helping the bouncing rays interact with the game worlds. Finer detail explanations of the tech can be found in Nvidia’s reveal video as well as Digital Foundry’s overview of it.
Alan Wake 2’s Implementation
Alan Wake 2 was confirmed as the first game to be updated with RTX Mega Geometry, and as I still had the game installed with a prior update and Nvidia driver, I decided I would compare my current install without updating either (to preserve the old performance not just without Mega Geometry but also without the new driver), to the version following the RTX Mega Geometry update, using the original CNN implementation of DLSS. My PC’s main components are an AMD Ryzen 5800X3D, 32GB of 3600MHz DDR4 RAM, and – crucially for this test – an older RTX 3070 8GB GPU, known for having issues with its small amount of Video RAM. Since RTX Mega Geometry was said to reduce VRAM usage while improving performance, I imagined the difference could be more noticeable. I decided to look at two specific early-game spots – the forest-set crime scene known for being particularly heavy on all machines, and Saga Anderson’s Mind Place, an enclosed location where average performance is much higher. Each spot was photographed 3 times before and after the update – once with no RT at DLSS 4K Performace Mode (internal 1080p resolution), once with RT set to High at the same DLSS 4K Performace Mode, and once with RT set to High at DLSS 4K Ultra Performance Mode (internal 720p resolution), with graphics settings always at the High Preset.
Crime Scene, Pre-Update
Here we see the basics of a pattern you’ll see for the remaining comparisons. The no-RT setting at DLSS Performance Mode runs quite nicely at around 41 frames per second, as would be expected for a decent mid-range GPU like the 3070. Once RT is set to high, activating a significant amount of path tracing and Nvidia’s Ray Reconstruction “denoising” solution (which deploys AI to clean up the issues by using only one or two rays for each pixel, rather than the hundreds used by feature films), the framerate severely tanks down to only 14 fps – and this is following several seconds of much lower numbers, as the PC almost seems to reel from the shock. Next, with a change to DLSS Ultra Performance Mode, which keeps the same output resolution but cuts the internal number of pixels by more than half, the framerate proceeds to rise to about 22fps – not enough to make the game particularly playable in this heavy forest area, but technically a notable upgrade. There’s another major difference to observe here, though – in the second image, showing High RT at Performance Mode, the texture detail suffers heavily, losing a great deal of AW2’s cinematic feel.
Crime Scene, Post-Update
Following the Nvidia driver update to 572.16 and RTX Mega Geometry update, I took these images using the same settings as before, taking care to keep the DLSS in Legacy Mode to avoid the Transformer Model’s higher costs affecting the results. I was rather surprised here – the framerates in the RT modes (second and third images) look practically unchanged, with the PerformancevMode especially having the exact same 14fps. Ultra Performance with RT only sees a 2-frame boost to 24fps – which isn’t nothing, to be fair. The larger framerate gain, interestingly, is seen in the non-RT setting, with a significant jump to around 51 fps. This is interesting, as the Mega Geometry update is supposed to improve the performance of ray tracing specifically – but I’ve been led to believe that even with the RT settings off, AW2 is always using some level of RT, the technology being a fundamental part of the game’s lighting. So this makes sense. But framerate is not the only place for improvement – you may have noticed that the severe texture issues seen in the second image’s RT+Performance Mode setting have largely vanished. So it appears that Mega Geometry may benefit you in different ways depending on your setting choices. The greatest oddity here for me is that VRAM usage has actually seemed to increase, contrary to the claims of Mega Geometry reducing how hard that particular part would need to be working. Let’s see how the Mind Place goes.
Saga’s Mind Place, Pre-Update
These images in Saga’s Mind Place prior to the Nvidia driver and RTX Mega Geometry updates follow the same basic pattern seen in the Crime Scene screenshots – high fps, then low, then back to a decent middle – but with a significantly higher baseline owing to the enclosed space we find ourselves in. So much less for the engine and PC to render translates to higher performance – games are rarely simple on a technical level, but sometimes you get exactly what you would expect. The no-RT Performance Mode gets a less than 50% boost to 54fps while the RT Performance and RT Ultra Performance Modes get 30 and 44 fps respectively – more than double the frames! – we’re likely seeing bottlenecks in the forest sections alleviated in the RT modes. However, RT Performance Mode still suffers from the same texture issues seen earlier – they’re especially blatant on Saga’s sweater, the Christmas-y designs suffering severe blur.
Saga’s Mind Place, Post-Update
And here we see a pretty similar improvement to before – a significant gain by the no-RT Performance mode to 67 fps, while the RT modes increase to 32 and 50 fps respectively. Things seem to be looking up in the Mind Place, though – the RT Ultra Performance Mode goes up by a higher percentage than before, and the RT Performance Mode seeing an increase at all is an improvement over the Crime Scene – and it repeats the feat of fixed texture detail, with Saga’s sweater now looking just as detailed as the other images. But overall it’s certainly the mode without RT that comes out the performance winner here. Interestingly, the overall increase in VRAM usage continues here. I’m hesitant to actually complain, though – while the reduction in usage not panning out seems bad at first, this actually looks more like my 3070’s VRAM is finally being properly used, without actually being maxed out. At least, that’s my hope.
Conclusion
This is just a tiny test on two locations on a single machine, without even video to demonstrate framerate variability – and there were certainly unexpected variables, like the increase in VRAM usage rather than the expected reduction. All this is to say, please look up all the benchmarks you can if you’ve taken an interest in this subject. Different rigs, different settings, and eventually different games will be able to teach us a great deal about RTX Mega Geometry, with my article just being a drop in the bucket. In fact, during the writing of this article, Digital Foundry dropped a complex breakdown of Mega Geometry with a bunch of very interesting results, which you should immediately watch. But I hope it’s been educational for you. And if this of all things has gotten you interested in playing Alan Wake 2, let me say that you’ve got good instincts. It’s a brilliant, beautiful, scary yet amusing experience that more people should be playing.
