Nvidia’s Deep Learning Super Sampling (DLSS) is an advanced image upscaling technology that uses deep learning and AI to boost frame rates in games while maintaining sharp image quality. DLSS leverages a deep neural network trained on high-resolution images to intelligently upscale lower resolution renders in real-time. This allows games to render at a lower target resolution, reducing strain on the GPU, while the DLSS model uses advanced temporal data and AI algorithms to intelligently fill in fine details and edges. The result is significantly faster frame rates with image quality that rivals true native resolution.
How DLSS 2.0 Improves Upon Previous Versions
When Nvidia first introduced DLSS in 2018, it showed promising performance gains but image quality suffered from blurriness and artifacts. With DLSS 2.0 in 2020, Nvidia completely overhauled the AI model and algorithms to deliver a massive leap in image quality. Here’s an overview of the key improvements in DLSS 2.0:
-
New AI Architechture – DLSS 2.0 features a new AI model called the DLSS Image Generator. It’s trained on ultra-high resolution images for greater precision and image quality.
-
Custom Neural Networks Per Game – Individual neural networks are now tuned for each game to maximize performance and image quality.
-
Improved Temporal Feedback – The DLSS model now takes motion data from previous frames into account for sharper details in motion.
-
Greater Performance Gains – DLSS 2.0 provides up to a 3X frame rate boost compared to native resolution, 2X more than the original DLSS.
How DLSS Works to Boost Performance While Maintaining Quality
DLSS provides dramatic performance gains while maintaining crisp image quality by relying on the power of a deep neural network and specialized tensor cores on RTX GPUs. Here is an overview of how it works:
-
Games render normally at native resolution to produce a high-quality image.
-
That image is used to train a deep neural network on Nvidia’s supercomputers.
-
In the game, DLSS renders at a lower target resolution, reducing load on the GPU.
-
The low resolution image passes through the trained DLSS model on the GPU’s tensor cores.
-
The DLSS model uses temporal data and its AI algorithms to intelligently reconstruct high-resolution details and edges.
-
Output resolution is restored to native resolution for a final sharpening pass.
-
The result is an image with quality nearly identical to native resolution, but with significantly higher performance.
DLSS leverages the massive parallel processing power of RTX tensor cores along with the AI model to perform these intelligent upscaling operations in real-time.
Image Quality Comparisons
Multiple comparisons between native resolution, DLSS, and other upscaling methods like temporal anti-aliasing (TAA) show that DLSS 2.0 achieves image quality that is essentially indistinguishable from native resolution renders.
Here is an example comparison between 1440p native resolution and 1440p with DLSS enabled from the game Control:
The level of detail is nearly identical, but DLSS boosts frame rates by over 60% compared to native 1440p in this example.
Other upscaling techniques like TAA exhibit more blurring, edge artifacts, and lost detail compared to DLSS:
Performance Benchmarks
Benchmarks across multiple games demonstrate how DLSS 2.0 effectively increases frame rates while maintaining sharp image quality compared to native resolution rendering.
Here are some examples of the performance gains DLSS enables in popular titles at 1440p resolution:
-
Control (with Max Settings):
-
Native 1440p: 80 FPS
-
DLSS 1440p: 128 FPS (60% increase)
-
Death Stranding (with Max Settings):
-
Native 1440p: 96 FPS
-
DLSS 1440p: 142 FPS (48% increase)
-
Cyberpunk 2077 (with Max Settings):
-
Native 1440p: 48 FPS
-
DLSS 1440p: 88 FPS (83% increase)
These benchmarks demonstrate how DLSS can allow you to maximize graphics settings and maintain smooth high frame rates without sacrificing image quality.
When to Enable DLSS
DLSS provides the biggest performance gains at higher resolutions like 1440p and 4K. It’s less impactful at lower resolutions like 1080p where the GPU is less likely to be the bottleneck.
Ideally, DLSS should be enabled when you want to maximize graphics quality settings but your frame rate is too low at your target resolution. This allows you to increase resolution and detail while still maintaining a steady 60 FPS or higher.
DLSS is also useful if you have a lower-end RTX card but want to play at higher resolutions while keeping frame rates smooth. For example, the RTX 3060 may struggle to run newer games past 60 FPS at 4K. But with DLSS, it can often double frame rates at 4K while still looking great.
Current Limitations
While DLSS 2.0 delivers huge performance and efficiency improvements for most games, it does have some current limitations:
-
Limited game support – DLSS must be specifically integrated into each game by the developer. Support is growing but still somewhat limited.
-
Only works on RTX cards – DLSS requires tensor cores which are only found in RTX 20 and 30 series cards. No support for GTX cards.
-
Resolution constraints – DLSS works best at higher resolutions like 1440p and 4K. Quality gains are less noticeable at 1080p.
-
Texture pop-in – Some users note occasional texture pop-in during fast motion when DLSS is enabled.
Despite these limitations, DLSS 2.0 marks a major step forward for efficient upscaling and anti-aliasing technologies leveraging AI and deep learning.
The Future is Bright for DLSS
DLSS 2.0 showcases the incredible gains that are possible when combining advanced AI algorithms with specialized hardware like tensor cores. With support continuing to grow through integrations into game engines like Unreal Engine 4 and Unity, DLSS adoption is poised to rapidly accelerate.
Nvidia continues to improve the technology too. DLSS 3.0 promises to reduce input lag, enhance image quality further, and boost performance even more by having AI fill in extra frames alongside the traditionally rendered frames.
As GPUs grow more powerful, deep learning super sampling technologies like DLSS will become critical to maximizing both efficiency and graphical quality in next-generation games. I’m excited to see DLSS evolve and help push real-time graphics to new heights with the aid of AI and deep neural networks.
