I saw DLSS 5 running across multiple games. It's not a face filter.

The guys with the slide rule built the F-1 engine in the Saturrn V rocket. The guys with the calculators, PC's, scanners, and supercomputers can't replicate it. Better tools does NOT equal a better outcome.
Eh? Compared to the F-1 on the Saturn V's, Starship's Raptor 3 engine has a higher specific impulse, far higher thrust/weight ratio, and is good for 50+ flights rather than's the F-1's single flight -- yet it was designed and built with 1/20 the staff, in 1/3 the time, and for less than 1/1,000 the cost. Calculators and computers really are better than slide rules.
 
The AI model powering DLSS 5 is a single unified model. Same model for every game. It's not trained per-title, per-face, or per-object type. It takes the raw color buffer and motion vectors as input, analyzes the scene semantics from that single frame, and enhances the lighting and material response while staying anchored to the original 3D content. It recognizes the difference between skin and metal and water and stone and foliage, and it processes each of those materials differently based on how light should interact with them.
Not sure how you can "frame" this as any different than a filter. The fact that it's not trained on the assets it's changing it's a dead give away that it's practically acting like a filter. You mention how the hardware isn't capable of producing that level of detail, well the AI isn't magic, it's stealing assets from what it's learned. It's processing the rendering into individual pieces and taking what it knows as normal assets and applying lighting and shading to it. What results is in something photorealistic because it's just spitting out what it's learned. This is why we see facial features like lips being changed. It's not upgrading the lighting and shading on the original texture it's changing the entire thing to conform to what it thinks is "real", which is most likely real photographs. The fact of the matter is "photorealism" when it's not actually real is just art imitating life. If you've trained a model on real assets then it's just slopping those assets into place. The "Huge Step" this is taking is the massive amount of time it takes to take an drawing of what an artist wants someone to look like and redrawing it to photorealistic levels. Only the artist can make that jump because then who is filling the gaps?
"They get spatial masking, so they can set the water enhancement to 100%, wood to 30%, characters to 120%"
That doesn't sound like the developers who are creating assets get a say, just the directors aka the business.
I am truly curious which developers responded well to this technology? Most likely not any of the artists dealing with lighting, shading, textures. Probably the same type of people who "responded" well to autonomous generative AI creating programs in tech companies. Most likely "Lead" developers.

Do better techspot. this reads like propaganda.
 
Eh? Compared to the F-1 on the Saturn V's, Starship's Raptor 3 engine has a higher specific impulse, far higher thrust/weight ratio, and is good for 50+ flights rather than's the F-1's single flight -- yet it was designed and built with 1/20 the staff, in 1/3 the time, and for less than 1/1,000 the cost. Calculators and computers really are better than slide rules.

At 1.5 million pound-force of thrust, the F-1 is almost 3 times the thrust as the Raptor 3 at 617,000 pound force. That's why the starship needs 33 engines to get of the grround while the Saturn V only needed 5 engines. I'm not sure where your going with costs, or how you broke out how much Rocketdyne's costs and manpower for the engine were from the entire budget for the Saturn V and NASA. The point is moot anyway.....There is simply no single chamber engine more powerful. So much so that Jeff Bezos went to the trouble of recovering several of them from the ocean to see how they worked.

Apparently even Elon knows it's easier to make 33 smaller engines work together than 5 huge ones. Aside from the fact that the technology to build them is lost.

Considreing the Apollo flight computer was not much bigger or more powerful than a modern calculator and was built in the 1960's Space-X had a substancial computational advantage.
 
Last edited:
I had a bunch of company directed AI training sessions and they wanted us to do an hour + of prep reading.

So I took the prep documentation, ran it through co-pilot and asked for an executive summary.

Boom.
 
I like the idea of ​​DLSS 5 for the visual appeal. These are video games—the "Insta filter" effect on fictional characters doesn't bother me. What bothers me is the order of priorities: we still have games running at 20 fps with path tracing on high-end hardware. If Nvidia first figured out how to make ray tracing efficient and accessible, and then added the aesthetic layer on top, it would be a completely different story.
 
Nobody's forcing people to use it if you don't like it turn it off. All the complaining and not picking is childish. Just like all the people who complain about DLSS and frame generation.
 
I really liked some of the environmental upgrades they showed. Some of the faces also looked pretty good.
As long as the developers are the ones putting the final touches on how it looks - then they decide if the result is in line with what they intended or not.
I am concerned we’ll see «neural rendering» addon cards coming as yet another expensive piece of hardwarw though - that was a bleak time in gaming history
 
The Nshitification of Video Games :

The AI model powering DLSS 5 is a single unified model. Same model for every game. It's not trained per-title, per-face, or per-object type [....] And it's deterministic and consistent
 
And yet, in that last year, the economy continues to grow healthily ...
Which one?

Before the Industrial revolution, essentially 100% of adults and children age 12 and up worked long hours every day. Today, a full 40% of adults (and nearly all children) don't participate in the work force at all ...
Before Stone Age human species spent 4 hours a day in average to get food.
Now some have to take 2 jobs to pay the bills.

When food, housing, and even complex medical care can be cheaply provided via a robotic workforce ...
For those who can afford that.
And those who can .. will use the cheap human slaves.
 
Last edited:
The better question is "What is photorealism?" Does anyone actually know or see it? According to common photography beliefs, the IPhone can replace the DSLR or medium format cameras. How is this possible, you may ask? Because the digital enhancement produces a high resolution picture that is pleasing to the eye.
Exactly.
People do not insist on it being real. They want to be pleased.

Jensen is the P.T. Barnum of the 21st century.
I doubt some youngsters know the reference without asking ChatGPT (or other A.I. bot).

 
To me with dlss5 applied It looks like oversaturated images with too much light, sometimes the light color changes. So I do not understand what is the benefit? If the developers wanted the light to be a color or another they could have done that, they do not need postprocessing. Even the samurai in the last image is left red-faced by the dlss5 processing.
 
The people on this website:

"Ai is taking away jobs day by day"...
people's response:
"Well it's just a generational shift."

"NVIDIA is changing the way games look"
People's response:
"OMG, how dare they do such thing"

Obviously people on here have the wrong priorities on what to worry about.
This is hilarious levels of missing the point.

As a software developer/architect/consultant/nerd, with years and years of experience, I am not worried that AI is so competent that my job will be gone.

I am worried that some VP has their annual bonus tied to adopting AI and getting enough productivity to warrant a 40% headcount reduction... where they decide to start with headcount reduction, and expect the remaining devs to make magic happen.

I get there's a really, really subtle difference there, that may be hard for some people to perceive. The AI doesn't need to do my job to replace me. All that needs to happen is for a salesman to convince a director that it *could*. Neither the salesman, nor the VP have any understanding of what the job entails, or what the product or the teams need... they just want to hear that KPIs will be hit, and they'll make their bonus. If it all falls apart in Q3, that's fine, as long as they maximize net until the wheels fall off.

Both are cases of the VC-backed leadership telling the experts in the field that "this is the future".
 
It looks interesting, but let's see how it works in reality at home and at what price range. I am always a bit sceptical about the unusual cases... games that have very strange environments, a bit like cyberpunk, horizon zero dawn. The second thing I am worried about is how the artistry of the game studios will develop, are we going to end up with draft looking games being filled up by AI? I still think artistry and composition need human thoughts, things that make games beautiful and unique. The probabilistic engines can fill some detail carving but the intention should be there...
 
I do not quite agree with you. If I was a artist in the digital realm I would:

1) Engage and understand how my artistic skills can be used by this model to drive cost down & efficiency up for game development.
2) Point one makes the artist a partner and not a grudge for the developer. If not, you're just a cost to the developer that needs to be managed in line with available budget.
3) If quality is the artists concern then work with the technology team for quality application outcomes and better efficiency.

Taking a hardline approach will lead to artist alienation, technology will not stop to evolve. If you do not like Nvidia then partner with AMD or Intel for a better solution to drive competition which benefits consumers in the end. Nvidia is very dominant and it's because they are giving people the technology they wanted. The only complaints now is that they're pivoting to AI too fast and alienating their core past customer base. But that opens up opportunity for other players in the GPU space who can't compete with the AI game.

Just My2c.

This is a diffusion model layered into the upscaling to give the illusion of better graphics.

If that isn't sad enough on its own, just consider how it uses the generic looking faces you see in every diffusion model to conpletely butcher the original look made by the artist - assuming the studio still has enough integrity to hire artists in the first place.

This is a low-IQ dull sledgehammer for a fabricated "problem". Like a texture pack for your game made by Shutterstock. It must be very cost effective, because the end result is completely lacking anything that could even be confused for quality.
 
Before Stone Age human species spent 4 hours a day in average [sic] to get food.
Now some have to take 2 jobs to pay the bills.
And many of those human species hunted other humans, a time-saving shortcut to finding meat. Only about 57% survived even to age 15. They also had essentially no possessions, no medical or dental care; housing was a cold dank cave without furniture or lighting, and the usual night's entertainment was watching your father impregnate your sister. I'll stick with the modern lifestyle.
 
DLSS 5 have cached lot of attention.
At least the mass will see it live (implemented) in the news in real time.

So, you can't afford new GPU to see it in your PC setup?
No problem, go buy (still) new cheap big DumbTV!
You are again on the roof with the tide!

/s
 
That's a bit like saying "With self driving cars you wont need wheels or gas anymore, they'll just drive themselves!".

You still need ROPs and Texture units and Shader cores to render what the AI infers. The Tensor cores themselves cant make the video happen.
This isn't true. AI image and video generation works through diffusion models; the output is finished pixels, not 3D frames which need to be textured and shaded. An AI-only graphics card is certainly possible (though far beyond our computational means at present).
 
Last edited:
Shrout has form for not declaring conflicts of interests, not declaring that he and others at PCper were working for Intel as 'private consultants' at the same time as running a tech review site that was reviewing those products, a tech site that just happend to heavily favour Intel/Nvidia in its output.
Some basic due diligence, would show that. Guess people have short memories.
 
Back