Nvidia DLSS 4 transformer model exits beta, set to bring improved graphics to more games

[Daniel Sims]

Why it matters: Most people think of multi-frame generation when they hear about Nvidia DLSS 4, but the transformer model upgrade in DLSS Super Resolution might be the update's most consequential upgrade. Many games can already benefit from the feature, and it's likely to become the standard across upcoming releases.

The latest version of Nvidia's DLSS Super Resolution and Ray Reconstruction SDK, released on Wednesday, brings the transformer model out of beta. Promoting the upscaling technology into DLSS's stable branch will likely broaden its adoption in upcoming PC games.

First introduced in January alongside Nvidia's RTX 50 series graphics cards, the vision transformer model replaces the long-standing convolutional neural network previously used in DLSS. This marks the most substantial improvement to Nvidia's image reconstruction method since DLSS 2 popularized the technology in 2020.

Also read: DLSS 4 Upscaling at 4K is Actually Pretty Amazing

Although DLSS has earned praise for using machine learning to scale images to higher resolutions with good image quality over the past several years, the process isn't without visual flaws. Issues that often occur include ghosting, shimmering, and smearing. Ray Reconstruction, which combines denoising for ray tracing with the DLSS algorithm, can also sometimes worsen the problem.

DLSS 4's transformer model alters the algorithm to minimize these blemishes, resulting in a noticeably cleaner image, even when upscaling from half resolution in performance mode.

Our testing shows that, while AMD's competing FSR 4 method beats DLSS 3, DLSS 4's transformer model takes the lead. While Nvidia's marketing for DLSS 4 mostly focuses on multi-frame generation, which is exclusive to the RTX 50 series GPUs, the transformer model which improves upscaling quality on games supports RTX 20 series and newer.

Also check out: DLSS 4 Ray Reconstruction Analysis: Fixing Ugly Ray Tracing Noise

While many prominent games such as Doom: The Dark Ages, Dune: Awakening, Stellar Blade, and F1 25 have adopted the beta version of DLSS 4 since January, some notable recent titles still default to DLSS 3, such as The Alters, Oblivion Remastered, Expedition 33, and Assassin's Creed Shadows.

The next major game to upgrade to DLSS 4 will likely be Diablo IV when the Season 9 update launches on July 1.

Users can also force DLSS 4's transformer model into many older DLSS-compatible titles in the Nvidia App by navigating to Graphics > selecting the desired game > clicking on DLSS Override Model Presets under Driver Settings > and selecting Latest in the drop-down menu.

You can also try third-party apps like DLSS Swapper, DLSS Updater, or Optiscaler.

Permalink to story:

Nvidia DLSS 4 transformer model exits beta, set to bring improved graphics to more games

 

[godrilla]

I wonder if Nvidia is working on adaptive frame generation like $7 app Lossless scaling 3.0 has featured for the past 6 months or so? Another feature that is a no brainer advantage in the frame generation arena is improved latency from secondary gpus ( mix and match gpu vendors). People have been pairing Nvidia with amd gpus to improve latency from the secondary gpu offloading the frame generation rendering.

Circling back to the article and dlss transformer model the biggest winners are the 2000 series like 2080ti. Pair that with Lossless scaling app by allowing another entry level gpu to improve latency.
Here is some more information about dual gpu features from Google.

Ai Overview
Lossless Scaling (LS) with a dual-GPU setup for frame generation requires a secondary GPU powerful enough to handle the frame generation process, but not necessarily as powerful as the primary gaming GPU. The secondary GPU's performance directly impacts the final frame rate achieved with LS. A good starting point is a GTX 1060 or similar, but a stronger card like a GTX 1660, RTX 2060, or even an RTX 3060 or RX 6600 can offer better results, especially at higher resolutions.
Here's a more detailed breakdown:
1. Secondary GPU Capabilities:
Not necessarily the same level as the main GPU:
While a more powerful secondary GPU will generally lead to better performance with LS, it doesn't need to match the primary GPU's power.
Impact on frame generation:
The secondary GPU's processing power directly affects how many frames it can generate and how smoothly the final output appears.
Resolution and settings:
Higher resolutions and demanding LS settings will require a more powerful secondary GPU to maintain performance.
Example cards:
Some users have had success with cards like the GTX 1050 Ti, GTX 1060, GTX 1660, RX 6400, RX 6600, RTX 2060, and even RTX 3060 as secondary GPUs.
2. Dual GPU Considerations:
Power supply:
Ensure your power supply can handle both GPUs, especially if they are high-end.
Compatibility:
Not all games support GPU passthrough or dual GPU setups for LS.
Performance issues:
Some users have reported weird issues with dual GPUs, so it's crucial to test and configure your setup properly.
iGPU usage:
Laptops with integrated graphics (iGPUs) can be used as the secondary GPU, but performance can vary.
Connection and configuration:
Connect the monitor to the secondary GPU (or iGPU) and configure Windows to use the primary GPU for rendering the game.
3. Specific Recommendations:
Laptops:
For laptops with a discrete GPU and iGPU, consider using the iGPU for LS if it's powerful enough (e.g., Radeon 780M or better).
Desktop PCs:
For desktops, a GTX 1060, RTX 2060, or similar cards can be good starting points, with more powerful options like RTX 3060 or RX 6600 offering better performance.
VRAM:
Unless you're working with very high resolutions (above 4K), 4GB of VRAM on the secondary GPU should be sufficient.
4. Important Notes:
LS Discord:
Join the official Lossless Scaling Discord server for documentation and support.
LSFG version:
Use the latest version of Lossless Scaling (LSFG) for optimal dual-GPU support.
Testing is key:
Experiment with different configurations and secondary GPUs to find the best setup for your system and games.
Potential performance bottlenecks:
Be aware that the secondary GPU's PCIe slot bandwidth and other factors can potentially cause bottlenecks.
........................
Update : New Transformer model and Lossless scaling app are contributing more to mitigate e waste.

 

[Jarenien]

I swapped out all the DLLs in Spider-man 2 and hfs, it runs amazingly now. I have nearly everything on highest settings (including all Ray Tracing settings) and still get like 80-90fps on my 4070 and the game just looks freaking beautiful!