Nvidia just announced a game-streaming service, Grid, that promises to stream top-tier titles up to 1080p and 60 frames per second.
Nvidia DLSS 3.7 is the latest update to the long-running AI upscaling technology, and it further shows native performance doesn't matter.
I think hardware development is at a point where they need to figure out how to draw less power, These beefy high end cards eat wattage, and I'm curious if using DLSS & AI in general will lower the power draw. It would seem like the days of just adding more VRAM & horsepower is over. Law of diminishing returns. Pretty soon DLSS/FSR will be incorporated into everything, and eventually the tech will be good enough to hardly notice a difference if at all. AI is the future and it would be foolish to turn around and not incorporate it at all. Reliance on AI is only going to pick up more & more.
DLSS certainly lowers power consumption. Also, the numbers such as the 4090 at 450W does not tell you everything, most of the time the GPU stays between 200-350W in gameplay, which is not too different from the highest end GPU of 10 years ago. Plus, today you can undervolt + OC GPUs by a good margin to keep stock performance while utilizing 80% of the power limit.
You can make the 4090 extremely power efficient and keep 90% of its performance at 320W.
However, in today's world the chip manufacturing is limited by physics and we will have power increases in the next 5-10 years at the very least to keep the technology moving forward at a pace that satisfies both businesses and consumers.
Maybe in 10 years we have new tech coming to the markets which we are yet to invent or perhaps we can solve existing technologies problems with manufacturing or cost of production.
On the other hand, if we were to solve the energy problem on earth by utilizing fusion and solar etc. it would not matter how much these chips require. That being said, in the next 30-40 years that is a pipedream.
I don't think fusion is the way forward. It will mosy likely be too late when it's finally ready, meaning it will probably never be ready. Something else might arrive before though and then it becomes viable.
We need to stop the smear campaign on nuclear energy.
We could power everything forever if we wanted too.
PS4 Pro had dedicated hardware in it for supporting checkerboard rendering that was used significantly in PS4 first party titles, so you don't need to look to PC or even modern PC gaming. The first RTX cards released nearly 6 years ago, so how many nails does this coffin need?
I'm not sure what the point you're trying to make is, but PS4 Pro was before DLSS and FSR, and it still provides one of the highest performance uplifts while maintaining good image quality.
DLSS is it's own thing but checkerboarding om PS5 still is a rival to the likes of FSR2.
Um. That is my point. That there have been so many nails in this "native performance" coffin and they've been getting hammered in for years, even on PS4 Pro before DLSS was even a thing.
Don't know what's OP's point is either but ... checkerboard rendering was good enough for its time but in terms of image quality its wayy behind what's DLSS 3 or FSR 3 is currently offering.
The main point of the article and what OP missed here is that DLSS 3.7 is soo good that its nearly undisguisable from native rendering and basically throws the "its still blurry and inferior to native rendering" debacle, (that's been going around in PC community since 2019), right out of the window.
RaidenBlack
DLSS is as i said a different thing from FSR and checkerboard.
But you're talking about FSR 3 which probably is better than checkerboard, but FSR 3 has only started to get games this year, so checkerboard which was the first hardware upscaling solution was and is still one of the best upscaling solutions.
Give credit where credit is due, PlayStation was first and they got it right from the get go, and PSSR will almost certainly be better than it will be given credit for, heck digital foundry is already spreading misinformation about the Pro.
Tacoboto
Yes... Its amazing how many talekd about KZ2 deferred rendering, pointing out the explosions were lower res than the frame itself..
And of course, Then the idea of checkerboard rendering, not being native....
For sure, maybe this tech makes it minimal while pixel counting, but alas, seems performance and close enough , and not native now matters.....
I want to see it run native without DLSS.. why not?
Almost deaf person:
- lightweight portable 5$, speakers of 0,5cm diameter are the final nail in coffin of Hi-Fi audio!
Some people in 2010:
- smartphones are the final nain in the console gaming's coffin!
This is just the same.
AI upscalling is complete dogshit in terms of motion quality. The fact that someone is not aware of it (look at the deaf guy example) doesn't mean the flaws are not there. They are. And all it takes to see them, is to use a display that handles motion well, so either gets true 500fps at 500Hz LCD TN or OLED (or faster tech) or uses low persistence mode (check blurbusters.com if you don't know what it means) also known as Black Frame Insertion or backlight strobing.
Also, image ruined by any type of TAA is just as "native image" as chineese 0,5$ screwdriver is "high quality, heavy duty, for professional use". It's nowhere near it. But if you're an ignorant "journalist", you will publish crap like this article, just to flow with the current.
There's no coffin to native res quality and there never will be. Eventually, we'll have enough performance in rasterization to drive 500fps, which will be a game changer for motion quality while also adding other benefit - lower latency.
And at 500fps, the amount of time required for upscalling makes it completely useless.
This crap is only usable for cinematic stuff, like cutscenes and such. Not for gaming. Beware of ignorants on the internet. The TAA is not "native" and the shitty look of the modern games when you disable any TAA, is not "native" either as it's ruined by the developer's design choice - you can cheat by rendering every 4th pixel when you plan to put a smeary TAA pass on it later on. When you disable it, you will see a ruined image, horrible pixellation and other visual "glitches" but it is NOT what native would've looked like if you'd like to honestly compare the two.
Stay informed.
Main point of the article is how far DLSS has come with v3.7 since 2018. If this is what we're getting already, then imagine what we'll get within next ~3 years. Yes parity will obviously be there compared to the then native rendering tech but it'll slowly narrow down to the point it'll be indistinguishable.
Something similar is like the genAI Sora ... AI generative videos were turd back when they were introduced (the infamous Will Smith eating video) ... but now look at Sora, generating videos that just looks like real life.
You can improve quality but you will never be able to reach native quality in motion. The biggest part of why these upscallers are so praised is because they use previous frame data. You cannot do that without degrading latency and/or hurting the motion quality. If you put another flaw on top of it, coming from sample and hold method of displaying image, or coming from low framerate, sure, the difference between "screwed up image" vs. "image screwed up even more" may seem small or non-existent. But if you talk about gaming, not interactive movies, the upscallers are overhyped and harfmul tech for the gamers and the whole gaming industry. For example, a game designed around screwed up motion, like the TAA enabled games, will never be played with improved quality even 100 years later when hardware allows for native 16K res. The motion quality will be broken and even if you disable the AA pass, you will still get the broken image, cause the devs were designing their effects with smeary filter in mind - this is why you can disable TAA in some games today, manually, with some tinkering, but you get 1 to 16 understampled crap.
It's annoying that nobody seems to understand the serious drawbacks of AI assisted upscallers. Everyone just praises it and calling it a great revolution. Don't get me wrong. AI has its place in rendering. But NOT in gaming.
How much VRAM is standard today? My laptop has a 1080p QLED display but only an Intel Iris Xe with 128MB of VRAM. I currently do all my gaming on it but certain titles do suffer because of it. I plan on getting a Steam Deck OLED soon to play the newer and more demanding titles.
Maybe better to get a budget gaming laptop and link a dualsense to it
= Portable console with far better graphics than a steam deck! + bigger screen and able to use it for work / etc
Aleksha writes: "Nvidia has established itself as a dominant force in the world of AI, but I can't shake the worry of what this means for the RTX 50 series."
Echo sentiment here - I think the way GPUs are going, gaming could be secondary to deep learning. Wonder if the 40 series was the last true generation of GPUs?
No.. Jensen believes GPUs should stay expensive. Those wanting a top end GPU will have to splash out for it, or play at just 1080p and 60fps or something if you can only afford a low end option.
On the other hand if you don't care about RT or AI performance then there's always AMD that are doing ok at the mid range.
***or play at just 1080p and 60fps or something***
My over 2-year-old laptop GPU still runs fine. I think this is more a reason why GPUs are going to other things in priority, because the market reach for new users is shrinking as more PC gamers focus less on replacing older and still working parts that run RT/AI fine enough as it is. Not to say there aren't people who still do it, but I think the market is shrinking for having the latest and greatest like it has been the past two decades. Problem is we aren't growing things at a rate as we were, we're reaching the the flattening of that exponential curve in regards to advancement. We need another major technological advancement to restart that curve.
You also need to consider that NVIDIA are heavily invested in cloud gaming. So they are likely going to make moves to push you into yet another life subscription service.
NVIDIA will never change their price point until AMD or intel makes a GPU that is comparable and cheaper than them .
it happend before in the days of gtx280 which they changed the price from 650$ to 450$ in a matter of 2 weeks because of rx4870 which is being sold at 380$.
Nvidia presented Covert Protocol, a tech demo aiming to showcase the "power" of the Nvidia Ace technology applied to video game characters.
They look like they are in pain. Almost begging to be put down. It was uncomfortable to watch.
I don’t know why people keep thinking of it as AI vs no AI.
A much more likely scenario is the use of AI alongside human work.
Eg. AI voices used during side quests or banter to boost the lines of dialog.
AI generating additional pre determined branches in dialog tree options for more freedom in conversations with NPCs
"AI generating additional pre determined branches in dialog tree options for more freedom in conversations with NPCs"
I'm wondering about that last one. Will that make a game more fun or more immersive? In the end, how can it possibly be more than filler content and then if it is filler content how much do I really want to engage with conversing with it if I know it will lead no where?
It's one of those things that sounds cool on paper. But will probably get old fast.
The tech is now available, and it is up to creators to create something unique with it.
The biggest thing to talk about here is that every interaction requires communication to inworld servers so there's three big impacts here
1) games are always online no question about it
2) delays in processing on inworld servers, outages or unexpected load as a result of some astronomically popular game will cause real time game delays ever wait for a chat got response this will be similar as the context must be pulled via the llm.
Now as for the other impact the artistic one no I don't think writers can be replaced I've mentioned before often AI generated writing is word soup I still standby that it's also evident in the video to.
AI can not convery accurately human emotions and I don't think ever will.
I know publishers are looking to cut down on development costs but what happens when inworld decide to charge per interaction or update their pricing a year after your game goes live you have no choice but pay it or shutter it.
Ive felt for a while that we are heading towards this place of games being disposable entertainment and now it's feeling more and more accurate
The PS Now competitor... I wonder what the prices are going to be. I still hate playing streamed games though. I'd rather run them on the hardware itself with the lowest latency possible.
pretty impressed with what they have achieved here. my only issue is if it were my main device and i had a internet service interruption i wouldn't be able to do any gaming at all.
I rather download the game. To many risks with streaming games.
I rather download from Steam, it's usually cheap enough as it is, I don't know, maybe it's cool for those who has internet fast enough to do this, but mine can hardly handle Ps4 share play, so I'm presume this is a no go.
The tech behind this is interesting, It makes me wonder how the XB1 or PS4 would of turned out if Nvidia was given the contracts.