Early DLSS 4 test showcases cleaner images and multiplied framerates

Through the looking glass: Users are eagerly waiting to see whether real-world performance tests back up the big promises Nvidia made when presenting its latest AI-powered game rendering technology at CES. Although those reviews are likely weeks away, Digital Foundry has provided an early third-party trial showcasing how DLSS 4 eliminates upscaling glitches and adds new AI-generated frames.

Eurogamer has released a preliminary demonstration of Nvidia’s newly announced DLSS 4 and multi-frame rendering functionality. The results indicate definitive improvements over what’s currently available on RTX graphics cards, but the full extent of the costs involved remains unclear.

Multi-frame rendering, exclusive to the upcoming RTX 50 series GPUs, generates additional frames to multiply the effect of the frame generation technology Nvidia introduced with the 40 series. By injecting two or three additional frames, multi-frame rendering can triple or quadruple a game’s perceived framerate.

However, frame generation adds input latency, and users are concerned that multi-frame rendering could compound the problem. Digital Foundry’s test shows that, while the added frames aren’t free, they cost less than the initial application of frame generation.

In TechSpot’s benchmark from 2022, introducing frame generation to Cyberpunk 2077 in DLSS performance mode added approximately 10 milliseconds of lag. However, the accompanying 60 percent increase in framerate easily outweighed the delay.

Click to enlarge

Digital Foundry ran an updated build of the game with similar DLSS settings on an engineering sample of the upcoming GeForce RTX 5080 and received a smaller penalty from multi-frame rendering. On top of the prior added latency, the second AI-generated frame introduced around four milliseconds, while the third frame added about two – the price for a framerate increase totaling 71 percent.

Although Eurogamer didn’t discuss how good those “fake” frames looked, multi-frame generation appears to offer a better deal than currently available frame generation tech. Meanwhile, DLSS 4’s other changes promise substantially better image quality.

Transitioning from a convolutional neural network to a vision transformer likely provides the biggest improvement to DLSS super resolution since Nvidia introduced DLSS 2 in 2020. Flaws such as smearing, ghosting, and shimmering are now far less apparent. Ray reconstruction is also notably improved.

Click to enlarge

The vision transformer is likely DLSS 4’s most significant feature because it supports all RTX graphics cards. Furthermore, the Nvidia app will allow users to update older games without waiting for patches from developers.

What isn’t clear is the performance cost. Future benchmarks should extensively test which RTX 20 and 30 series GPUs can handle DLSS 4. Furthermore, the VRAM impact of multi-frame rendering remains undisclosed. The RTX 5070’s ability to implement the new feature with just 12 GB of memory will determine whether it lives up to Nvidia CEO Jensen Huang’s claims that it can match the flagship 4090.

Related Content

Network Traffic Types (With Examples)

Darktrace to acquire Cado Security to strengthen cloud forensic capabilities

Is It Really as Secure as Claimed?

Leave a Comment