NVIDIA released its new GPUs in September 2022. Featuring a new graphics processing architecture running on smaller four-nanometer transistors, the new 4000- Series GPUs come with many bells and whistles.
Better still, the new GPUs also come with DLSS 3, an artificial intelligence-powered image upscaling technology that can improve the frame rates on your rig exponentially.
But what is DLSS 3.0, and is it worth the upgrade? Well, let’s find out.
What Is DLSS 3.0?
Short for Deep Learning Super Sampling, DLSS is a neural graphics technology that uses the power of artificial intelligence (AI) to improve the frame rates on your system.
The supersampling in DLSS refers to an anti-aliasing technique used to improve video quality by rendering gaming frames at a higher resolution and then downsampling it—improving video quality by reducing aliasing. That said, rendering frames at higher resolutions is very taxing for your GPU, and using anti-aliasing features usually reduces your FPS. After all, your GPU must process more pixel data and downsample it to your native resolution.
This is where the “Deep Learning” part of DLSS comes into the picture. You see, in traditional anti-aliasing methods, the GPU has to render frames at higher resolutions, but with deep learning, the GPU does not have to do that. Instead, all it has to do is generate the frames at a native resolution, and then the tensor cores on the GPU predict how the frame should look when rendered at a higher resolution.
This approach reduces the computational overhead of rendering frames at a higher resolution due to AI intervention. Therefore, put simply, DLSS renders your games at a higher resolution by using Artificial Intelligence.
DLSS 3.0, on the other hand, is the third iteration of the same technology. It improves on DLSS by predicting complete frames rather than just increasing the frame resolution—improving frame rates exponentially.
Here is how it all works.
How Does DLSS 3 Work?
Before getting into DLSS 3, it’s important to understand how older versions work—and how DLSS 3 builds on it.
As explained earlier, DLSS uses AI to render images at higher resolution. This means that the GPU is not programmed to increase the resolution of the frames. Instead, the GPU is trained by showing lower and higher-resolution images to program itself.
NVIDIA runs a Convolutional Neural Network (CNN) to perform this training on its supercomputers. This network is then shown images of a game running at lower resolutions as the input. Simultaneously, as an output, the network is shown the same images rendered at 64 times the resolution with both anti-aliasing features enabled and disabled.
In addition to the high and low-resolution images, the CNN is also trained using temporal feedback. This feedback provides the network information about how objects in the image move across frames with respect to their native and higher-resolution output. This enables the CNN to predict the appearance of the next frames well in advance—offering better frame rates and image quality.
This constant bombardment of image data on the network trains it, enabling it to upscale the resolution of games instantaneously. Once trained, this network is sent to NVIDIA GPUs through driver updates, enabling them to increase the resolution of images using trained neural networks.
DLSS 3.0, on the contrary, goes a step further and renders complete frames using this methodology. Therefore, not only does DLSS 3 increase the resolution of the games, but it also interleaves AI-generated frames in your gameplay.
Due to this approach, the GPU has to process a lot less data, and according to NVIDIA, with DLSS 3 enabled, the GPU only computes 1/8 of the frame. AI predicts all the rest. It is this increase in AI rendering which enables FPS to be delivered four times faster when compared to traditional rendering methods.
But how does DLSS 3 predict entire frames without using conventional rendering pipelines? Well, it’s all thanks to NVIDIA’s new Ada Lovelace architecture running on new fourth-generation tensor cores, which enables frame generation using AI.
Here is how everything works.
Frame Generation Using AI on DLSS 3
So just like DLSS, DLSS 3 uses tensor cores to increase the resolution of the frames, but it also has special optical flow accelerators which help the GPU predict frames. To predict the frames, the optical flow accelerator gets several high-resolution data frames generated by DLSS. The optical flow accelerator then uses this data to generate the optical flow field.
This optical flow field defines how pixel data changes between two frames, and this data, along with geometric motion vectors, is used to generate AI frames. Therefore using the optical flow, NVIDIA RTX 4000-Series GPUs can place new frames generated using AI in between frames generated using the traditional approach—increasing the FPS.
That said, interleaving AI-generated frames in a game has its challenges, and the biggest one is input lag. After all, the GPU cannot predict the user input on a frame generated using AI.
To solve this problem, NVIDIA uses its Reflex technology.
DLSS 3 and NVIDIA Reflex
Before getting into NVIDIA Reflex, it’s important to understand how your mouse movements reach the GPU. So, when you move the mouse or press a key to move a character in a game, the mouse sends the pointing information to the CPU. Which then processes it and sends it to the render queue. From here, the data is sent to the GPU, which sends your pointing information to the display.
This traditional data input pipeline generates a lot of lag as the user inputs can stay in the render queue for longer, making you miss that headshot. To solve this problem, we have NVIDIA Reflex, a technology that eliminates the rendering queue and sends data directly to the GPU from the CPU—reducing input lag by up to 80 percent.
Can You Use DLSS 3 on Older GPUs?
NVIDIA released DLSS 3 with its RTX 4000-Series GPUs, and if you own an older RTX GPU which supports DLSS, you might be wondering if DLSS 3 will improve your gaming experience.
Most importantly, DLSS on older systems will improve with DLSS 3 as it uses AI, and the neural networks are bound to get better with the new updates. That said, newer frame generation technology on older systems will not be supported as it uses newer fourth-generation tensor cores along with optical flow accelerators, which can only be found on NVIDIA RTX 4000-Series.
That said, according to a Reddit thread, frame generation can be enabled on older RTX systems by making changes to configuration files. However, we’ve not had a chance to test whether this works.
Is DLSS 3 Worth Upgrading For?
DLSS 3 uses artificial intelligence to increase the resolution of the games you play. Not only does this approach offer better framerates, but it also makes gaming at high resolutions possible on lower-end GPUs.
Therefore, if you want to enjoy high FPS while playing demanding games at 4k on a budget, upgrading to DLSS is worth it.