DLSS 2.0 in Death Stranding - Nvidia 4K Performance Trump Card

An introduction to DLSS 2.0

DLSS 2.0 in Death Stranding - Nvidia Ace in the Hole for 4K performance

DLSS 2.0 in Death Stranding - Nvidia 4K performance Trump Card

Update - Our full Performance Review and Optimisation Guide for Death Stranding is now available to read. 

Death Stranding is the latest game to utilise DLSS 2.0, an AI-powered upscaling technology which allows Nvidia to deliver higher performance levels on its RTX series graphics cards while promising similar, or better, visuals. 

In the early days of DLSS, we were harsh critics of the technology. Thankfully, Nvidia was able to take these criticisms to heart and develop what is now known as DLSS 2.0, an updated technique which now delivers what the company promised with the launch of its RTX series of products in 2018.

We have already looked at Nvidia's DLSS 2.0 technology in games like Minecraft RTX, Control and MechWarrior 5: Mercenaries, each of which highlights the benefits of the technology. With Death Stranding, we will find out how beneficial Nvidia's deep learning upscaling can be, both performance-wise and in terms of visual quality.  

Will DLSS continue to impress, or will it have too many downsides to be worth considering? Let's find out. 

Contents

- DLSS 2.0 - Image Quality Compared
- DLSS VS AMD FidelityFX CAS
- DLSS Performance at 1080p
- DLSS Performance at 1440p
- DLSS Performance at 4K
- Conclusion


What is DLSS? 


Nvidia's Deep Learning Super Sampling (DLSS) technology is a bold attempt to utilise AI hardware to make PC games run faster and potentially look better. Outside of raytracing, DLSS was set to be one of Nvidia's killer RTX series features, using the power of Nvidia's Tensor cores to increase the performance of supported PC releases. 

In basic terms, DLSS uses the powers of AI and Machine Learning to create higher resolution versions of rendered images, creating high-resolution images from lower resolution renders through Nvidia's DLSS algorithm. Nvidia's DLSS algorithm promised to increase game performance without any notable reductions in graphical quality, giving gamers more performance headroom for raytracing and other advanced graphical effects. 

Sadly, early implementations of DLSS failed to live up to Nvidia's claims, creating a lot of ill will towards the company, RTX series graphics cards and AI-enhanced gaming as a whole. But thankfully, that wasn't the end of DLSS. 

Nvidia has taken its lumps and has developed DLSS 2.0, a "new and improved deep learning neural network" that promises to give gamers higher levels of image quality and enhanced game performance. Better still, DLSS 2.0 is said to be easier to integrate into games, which is great news for users of RTX series graphics cards.  

Now before we move onto DLSS 2.0, we need to look at where it all started, and where the controversy surrounding Nvidia's DLSS technology began. 

Early DLSS - Extra Performance & Extra Blur


Our first encounter with DLSS was in Metro Exodus, which released last year as one of Nvidia's first RTX titles. Alongside Raytraced Global Illumination, DLSS was one of Nvidia's highlight features within Metro Exodus, promising to help gamers reach higher framerates when the game's raytraced global illumination was enabled. 

Below is an image comparison from our day-1 PC Performance Analysis of Metro Exodus, highlighting what would later become known as "DLSS Blur" by much of the PC gaming community. While the game's DLSS implementation was later improved through game updates (more information here), the obvious downsides of DLSS remained. In short, DLSS wasn't the free performance upgrade that Nvidia initially marketed, the performance benefits of DLSS came with graphical downsides. 

In many games, the graphical/performance trade-off of DLSS was worth it, but for many others, the feature only generated disappointment for early RTX adopters. Nvidia needed to improve its DLSS technology to deliver what they initially promised, and that's where DLSS 2.0 comes in.  




DLSS 2.0 - The Next Generation of DLSS 

With DLSS 2.0, Nvidia promises to deliver sharper, more detailed images and more choice than ever before. For starters, all future DLSS games will feature several DLSS quality settings, Performance, Balanced and Quality, each of which is designed to deliver compelling visuals. 

Another change is that Nvidia's DLSS algorithm is now game-wide, not game-specific. This change will allow Nvidia to improve its algorithm to deliver further quality enhancements moving forward. This means that all DLSS 2.0 games can benefit from Nvidia's continued improvements in their DLSS algorithm, which means that DLSS' performance and quality should get better over time. 

Nvidia's DLSS algorithm has also been modified to remove many of the unintended artefacts which were present in earlier DLSS implementations. This change will make the change to DLSS less noticeable, and make DLSS a lot more seamless than before. 

How is this possible? One of the key factors of DLSS 2.0 is Nvidia's improved hardware efficiency. Nvidia can now get 2x more performance from its Tensor cores than before, allowing DLSS to work much harder than before. This change will help DLSS deliver higher framerates, and has allowed Nvidia to achieve better visuals by using more complex algorithms. 

Please head over to the next page to see DLSS 2.0 in action within Death Stranding. 

 

«Prev 1 2 3 4 5 6 7 Next»

Most Recent Comments

15-07-2020, 16:44:37

AlienALX
I didn't think much of it, until I saw that chick's face. Her eyebrows look so much better, just so much more detail there.

I'm glad this is becoming a used thing now.Quote

15-07-2020, 17:30:17

WYP
Quote:
Originally Posted by AlienALX View Post
I didn't think much of it, until I saw that chick's face. Her eyebrows look so much better, just so much more detail there.

I'm glad this is becoming a used thing now.
Yeah, in many scenes both look near-identical, but when TAA artefacts come into play, DLSS 2.0 wins hard. I didn't believe how good DLSS looked initially.

TBH, while I love DLSS 2.0, I really want to see a 3rd party alternative that can work on both AMD and Nvidia cards. That's surely coming, as console makers will want to see this kind of upsampling, I want to see this kind of thing widely adopted on both the software and hardware side.Quote

15-07-2020, 17:47:17

AlienALX
Oh I don't care I have three RTX cards

haha jokes aside yeah this is cool.Quote

16-07-2020, 02:57:19

Avet
Quote:
Originally Posted by WYP View Post
Yeah, in many scenes both look near-identical, but when TAA artefacts come into play, DLSS 2.0 wins hard. I didn't believe how good DLSS looked initially.

TBH, while I love DLSS 2.0, I really want to see a 3rd party alternative that can work on both AMD and Nvidia cards. That's surely coming, as console makers will want to see this kind of upsampling, I want to see this kind of thing widely adopted on both the software and hardware side.
It will be hard to make 3rd party software that does it on both AMD and Nvidia cards. It will be either-or. This is a hardware-specific feature. Tensor cores are doing all the AI stuff locally on GPU die. AMD cards won't have that on their dies. And I don't think that stream cores can do all that math efficiently.

Consoles would kill for this feature, but we need to see what will AMD do. They don't have hundreds of thousands of Tesla GPUs in the basement doing all AI training, neither they have years of AI development like Nvidia has.

The console market is absurdly large so there may be an AMD alternative. Will it be this good remains to be seen.Quote

16-07-2020, 03:33:03

WYP
Quote:
Originally Posted by Avet View Post
It will be hard to make 3rd party software that does it on both AMD and Nvidia cards. It will be either-or. This is a hardware-specific feature. Tensor cores are doing all the AI stuff locally on GPU die. AMD cards won't have that on their dies. And I don't think that stream cores can do all that math efficiently.

Consoles would kill for this feature, but we need to see what will AMD do. They don't have hundreds of thousands of Tesla GPUs in the basement doing all AI training, neither they have years of AI development like Nvidia has.

The console market is absurdly large so there may be an AMD alternative. Will it be this good remains to be seen.
There has been a version of DLSS which didn't use Tensor cores, I believe the original release of Control used that version. There is definitely a way that similar feats to this can be done with more generic compute, and you can be sure that RDNA 2 will contain some new AI-ish features. The Xbox Series X does IIRC.Quote
Reply
x

Register for the OC3D Newsletter

Subscribing to the OC3D newsletter will keep you up-to-date on the latest technology reviews, competitions and goings-on at Overclock3D. We won't share your email address with ANYONE, and we will only email you with updates on site news, reviews, and competitions and you can unsubscribe easily at any time.

Simply enter your name and email address into the box below and be sure to click on the links in the confirmation emails that will arrive in your e-mail shortly after to complete the registration.

If you run into any problems, just drop us a message on the forums.