Table of Contents
Introduction
Much of the initial buzz around the launch of NVIDIA’s new GeForce RTX 50 Series GPUs was around their AI features, specifically DLSS 4. This suite of features promises faster framerates and higher resolutions without the need for additional hardware. While DLSS stands for “Deep Learning Super Sampling”, one of its most popular features, it has grown to include multiple others that now fall under the DLSS umbrella.
Some readers may have noticed that our testing of the 50 Series so far has not included any of these features. It seems logical that a feature that increases GPU performance would interest us and our customers, but it isn’t as straightforward as it may appear. In this post, we’ll examine all of these features and how they interact with workflows like game development and Virtual Production.
Looking for an Unreal Engine Workstation?
We build computers tailor-made for your workflow.
Don’t know where to start?
We can help!
Get in touch with our technical consultants today.
DLSS Super Resolution
This is the original feature introduced with the RTX 20-series cards. Super Resolution allows a game to render at a lower resolution, such as 720p, but then use machine learning algorithms to upscale that frame to a higher resolution, such as 4k, for the final display. Because 3D rendering is directly impacted by the number of pixels that need to be rendered at run time, the end user could get the performance of a lower resolution with the visual quality of a higher resolution.
In practice, it is not an exact comparison as there is still some overhead, but enabling this will yield a substantial gain. Regarding the visual differences between a native 4K frame and a 720p-upscaled-to-4K frame, the new model in DLSS 4 gets very close. It takes some serious zooming in on side-by-side comparisons to spot the difference and is an easy way for most gamers to get better frame rates without buying a more expensive GPU.
DLSS 4 Multi Frame Generation
Frame generation was introduced with DLSS 3 and the 40-series GPUs. This uses AI to generate entirely new frames between two rendered frames. The aim is that if the GPU can render 30 frames per second, for example, then AI would be able to generate interpolated frames, giving the appearance of 60 frames per second. DLSS 4 introduces Multi Frame Generation, which will generate up to 3 frames between each rendered frame. In the previous hypothetical 30 FPS example, 120 frames would be displayed.
While this is an interesting concept and does work fairly well, there are several downsides. First, there are often graphical errors, such as ghosting, where a faint outline of an object appears on the generated frame where that object was previously rendered. Sometimes, fast-moving objects will appear smudged in the generated frames. In motion, this can be fairly noticeable depending on the exact scenario being rendered. Gamer’s Nexus did a fantastic breakdown of some of these issues with screenshots of the present issues.
Image
Gamers may also not be as excited with Multi Frame Generation due to how game engines work with the game state updates happening with every new frame. For example, if a player presses a button to shoot a gun, that shot will happen in the next frame. If there is travel time on the bullet, the distance traveled, and if it hits an object, that will happen on subsequent frames. If the framerate is low, this may make the game feel sluggish and can lead to missed shots. As a result, gamers, especially those in competitive games, like to adjust settings to get the highest framerate possible. These gamers may have an issue with Frame Generation because the generated frames do not improve this latency. The game is still running at 30 frames per second, even if the GPU is displaying 60 or 120 frames. On the other hand, less action-oriented games may not notice the difference and prefer extra visual smoothness.
DLSS Ray Reconstruction
Ray Reconstruction is a fascinating technique for improving real-time ray-traced reflections and shadows. For ray tracing to work in real-time, the number of samples must be kept fairly low. This then relies on denoisers to smooth out the results, leaving shadows and reflections appearing smudged. DLSS Ray Reconstruction replaces these denoisers with an AI model that can better estimate the desired result. Much like how Super Resolution is able to take a lower-resolution image and upscale it, Ray Reconstruction can take the results of a low ray sample count and output a higher-resolution result.
Ray Reconstruction was added with DLSS 3.5, but with version 4, it moves from the previous Convolutional Neural Network (CNN) to a new Transformer model. There is a lot of very technical jargon to explain the difference, but what really matters is that this new model is much more accurate and temporarily stable than before.
DLSS 4 for Game Developers
The big question is how these features will benefit developers and their day-to-day workflows. On one hand, a developer may want higher-end hardware for their work, on the other, their game might want to support gamers on lower-end hardware. Let’s take the recently released Avowed as an example. The game is made in Unreal Engine 5, and while it uses DLSS 3, it will give us a good example of the wide range of hardware that needs to be supported. While this game uses the latest engine, produces highly detailed graphics, and recommends an RTX 3080, the minimum GPU supported is a GTX 1070. They also support video cards from AMD and Intel that do not use these technologies. That means that whatever they do, they need to ensure the game can run at playable speeds on systems where DLSS 4 is disabled and on a GTX 1070.
Image
Breaking it down further, we can look at specific workflows. A character animator will mainly be working in the DCC of choice, but the game engine will want to see the native frames to ensure everything is working properly. Additional AI-generated frames between their keyframes could cause them to miss issues with their animations. Meanwhile, an environment artist must ensure everything is running smoothly on various platforms.
Beyond that, there is the question of what frame rate is needed to do the tasks. If a developer gets 60 FPS on a GeForce RTX 5090 without any DLSS features enabled, would they see a benefit in enabling DLSS and jumping up to 120 FPS? Also, if they are getting 60 FPS on a 5090, what does that mean for potential customers on a 3090, 20-series card, or AMD?
One important note is that DLSS doesn’t appear in the standard viewport. It only takes effect when the user selects “Play in New Editor Window” or in a compiled exe. Not having the benefit of DLSS in the editor viewport further limits its usefulness during development. This is not to say that Developers won’t care about DLSS. In fact, every developer will be looking into it. The benefit to their customers is great – it just doesn’t help with development at this point.
DLSS 4 for Virtual Production
Another major industry that is keeping an eye on DLSS is virtual production. This industry relies particularly on game engines like Unreal Engine for their real-time graphical needs. One of the headline-grabbing aspects of virtual production is LED walls for ICVFX, and these features seem like a desirable benefit. However, they do not currently work with nDisplay or Mosaic, so these users cannot get the benefit. If NVIDIA updates DLSS to work across multiple sync’d systems, some features, especially Super Resolution and Ray Reconstruction, would be of great interest.
Some DLSS 4 features do work in the Movie Render Queue. Unreal has increasingly been used for film productions in place of, or in addition to, traditional offline rendering pipelines. For example the “Unreal Tournament: Xan” episode of Secret Level on Amazon Prime was made in Unreal Engine. When it comes to these types of productions, graphical quality is paramount. Ray Reconstruction, in particular, will be of great interest. On the other hand, many of these users will go with the much slower Path Tracer, which will not use DLSS. It is a trade-off that is well worth investigating.
DLSS 4 for ArchViz
Architecture visualization is another industry that will benefit from these DLSS 4 advances. Here, graphical quality is also key, but so is real-time playback. Temporal noise isn’t as apparent because there isn’t fact action like a game or a film project. When architects are showing off a project to a client, being able to move about the location and make changes in real-time is key. The more realistic the visuals, the better. D5 Render has become a popular option here, and the new 2.10 update ads DLSS 4 support.
Conclusion
NVIDIA’s continued advancements in DLSS technology showcase the power of AI-driven upscaling, frame generation, and ray reconstruction in improving visual fidelity and performance. While these features undeniably benefit gamers and end users, their impact on professional workflows—such as game development, virtual production, and architectural visualization—remains more nuanced.
For game developers, DLSS 4 presents a compelling way to enhance the final player experience, but it does little to accelerate actual development workflows. Similarly, while virtual production and ArchViz can take advantage of DLSS for final renders and real-time playback, current limitations prevent seamless integration into multi-system pipelines.
Ultimately, DLSS 4 is an impressive evolution of AI-powered graphics technology, but its role in professional content creation is still developing. As NVIDIA refines these tools, we may see broader applications that extend beyond gaming and into the core workflows of creatives and developers. Until then, DLSS remains a powerful tool for delivering better-looking, smoother-running experiences to end users—just not necessarily a game-changer for the professionals building them.
Unreal Engine Workstations
Puget Systems offers a range of powerful and reliable systems that are tailor-made for your unique workflow.
Labs Consultation Service
Our Labs team is available to provide in-depth hardware recommendations based on your workflow.
