Image Credits: @boliviainteligente via Unsplash
If you have bought a laptop or a PC within the last twenty years or so, chances are you have purchased a machine with parts made by Nvidia. If you have noticed the price of computer parts going up, Nvidia has something to do with that. And if you are wondering why people are talking about something called “DLSS 5,” Nvidia also is involved.
Founded in 1993, the Santa Clara based computer technology company has been at the center of controversy recently among its customers, game developers and the press in general after the unveiling of their new software, Deep Learning Super Sampling 5 (DLSS 5), which some people are alleging undermines the work of artists as the company continues to incorporate AI into its business. This poses a serious question: is DLSS 5 a real advancement in graphical power, or do Nvidia’s critics have a point?
Introduced in February 2019, DLSS is the company’s proprietary suite of machine learning software that is intended to help computer games run faster while maintaining graphical fidelity or better. This works by using AI models to analyze each in-game frame at lower resolutions using Supersampling (a demanding anti-aliasing technique used to make lines and curves on a computer screen less “jagged” albeit a little blurry) and then make more frames per second (FPS) and can enhance the graphical quality of the image.
Unveiled to the world on March 16 at the company’s semi-annual conference in San Jose, California, DLSS 5 made its international debut, displaying the effects of the graphical software when used on games such as Capcom’s “Resident Evil: Requiem,” Bethesda Game Works’ “Starfield,” as well as Avalanche Software’s “Hogwarts Legacy.” Along with their presentation, the company stated on their website that same day: “DLSS 5 introduces a real-time neural rendering model that infuses pixels with photoreal lighting and materials. Bridging the divide between rendering and reality, DLSS 5 empowers game developers to deliver a new level of photoreal computer graphics previously only achieved in Hollywood visual effects.” A similar sentiment was echoed by Todd Howard, Executive Producer and Game Director of Bethesda, who praised DLSS 5 in a short video posted to the studio’s X account, where he praised how DLSS 5 would bring “photorealistic lighting and materials,” to Starfield.
However, most that watched the presentation noticed something strange about the effects on display, most notably with the effect DLSS 5 had on Grace Ashcroft, a character from “Resident Evil: Requiem.” With the software turned on, it seemed like her cheekbones were sharper, her lips fuller and more colorful, her hair more vibrant, her nose slightly upturned and more.
This sequence from the presentation quickly went viral and led to mixed reactions from the rest of the industry and the public to outright scorn. Kyle Orland, writing for website Ars Technica wrote: “Many of the reactions have focused on how DLSS 5 turns in-game faces into overly detailed, uncanny valley versions of the original models…Others have noted how DLSS 5 seems to mangle the intended art direction by dampening shadows in favor of a homogenized look.”
Those criticisms he mentions were voiced by developers within the gaming industry who were concerned with the effect the software has on the work done by programmers and various artists. As pointed out by James Bentley, writing for PC Gamer magazine: “Over on X, New Blood co-founder Dave Oshry shares a meme calling the tech ‘Pure Slopium’, a reference to AI slop.”
“This is NOT the direction games should be going in,” said Jeff Talbot, a Senior Concept Artist with Gunfire Games posting to his BlueSky account, “In every shot the Art Direction was taken away for the senseless addition of ‘details’. Each DLSS 5 shot looked worse and had less character than the original. This is just a garbage AI Filter.”
During a Q&A with the press on the following day of the conference, March 17, Jensen Huang (one of Nvidia’s founders and its CEO) came to his company’s defense in the context of DLSS 5. When asked by editor-in-chief, Paul Alcorn, of website Tom’s Hardware about how he felt about the criticisms leveled at DLSS 5, Huang replied: “Well, first of all, they’re completely wrong…The reason for that is because… DLSS 5 fuses controllability of the geometry and textures and everything about the game with generative AI… it’s not post-processing at the frame level, it’s generative control at the geometry level.”
More concerning though, is how it seems there’s a discrepancy between what Jensen Huang is saying about DLSS 5, and how it actually works. In an article from PC Gamer’s Dave James, during an interview between Nvidia employee Jacob Freeman and YouTuber Daniel Owen on March 19, Freeman explains that DLSS 5 is actually post-processing at a frame level, contradicting Nvidia’s CEO: “As Freeman says: ‘DLSS 5 takes a 2D frame plus motion vectors as input.’ So, unless Freeman is grossly oversimplifying things here, it really is essentially taking a screenshot of a game and applying an AI filter to it…It’s apparently completely limited to screen space and the model has zero awareness of anything that sits outside of the single image it’s working on.”
If that wasn’t enough, it seems as though some of the employees with the game developers that Nvidia partnered with to use DLSS 5, such as Capcom, had no idea about it until the new software was unveiled at the conference. “Developers at CAPCOM tell Insider Gaming that the announcement and the publisher’s involvement were particularly shocking,” writes Tom Henderson for website Insider Gaming, “as CAPCOM has previously been historically very ‘anti-AI’ with projects such as ‘Resident Evil Requiem.’”
As of April 7, according to a poll conducted by PC Gamer, only 10% of 4,484 of voters were sold on DLSS 5, 9% might use it on some games, 37% ethically opposed to it, and 34% decrying it and vowing to never use it.
