This is some old news, it happened during the CES 2026. NVIDIA has announced the new AI Upscaler DLSS 4.5.

It supports all RTX graphics cards, and I'm not sure if it will come to consoles too.

Here are some videos showing off how it works.

If you have a 50 series card, and thus FP8 on the gpu, its not that much more demanding, compared to the improvements to image quality you get.

I saw this video (from Daniel Owen) and it surprised me:

https://www.youtube.com/watch?v=7x7goOdiaSU



Seems well worth using for 1440p.

I like DLSS, I do 4k output with 1440l rendering quite often, gives me quite a few more frames.

Good tech. Still loathe frame gen, ghosting and latency.

Just when AMD had started seriously closing the gap. Are people gonna have to wait for UDNA for the next significant improvement on their side? Cause if so that really won't be ideal especially considering that DLSS 5 a year from now will be even better than this. This is gonna get real wild then 240p upscaled to 4K starts looking good cause the performance improvements will be insane. Like in this video it's already nuts just how usable 240p upscaled to 720p is when the base image is so abysmal so I can't wait to see how much better this gets a few years from now. They're gonna need to introduce a preset below ultra performance before much longer at this rate.

FSR redstone is still a tiny bit behinde the DLSS 4.0 Transformer model.

So its like  dlss 4.5 > dlss 4.0 > FSR redstone > dlss 3.5 > FSR 4.0 > dlss 3.0 > Older FSR models.

I think Intel's XeSS is abit behinde both FSR redstone and FSR4... but ahead of some of the really old FSR models.

Anyways if your a nvidia user your well off with DLSS.
*IF* you have a new AMD card your also well off.....   If your on a intel card.... you are currently the worst off, of the 3.

But like how many people have intel cards?
Nvidia like 85%-90% market share of gaming gpu's.
Then AMD is like 10-15% or so at most.
Intel is basically margin of error... like ~1% at most.

Its brutal how dominant Nvidia is in the discrete gpu market.

AMD's upscaling has gotten good enough that being behind isn't as bad now though it'll be hard for me to recommend an AMD GPU going forward unless it has a great price and the long term support gets a lot better. RDNA 3 isn't as bad as RDNA 1 but that lineup is gonna age notably worse than the RTX 40 series. RDNA 4 is a step in the right direction though so hopefully they have a Ryzen moment with their GPUs in the coming years.

DLSS 4.5 > DLSS 4 > FSR4 > XeSS 2 > XeSS 1 > FSR3

By the way, FSR4 -is- Redstone, just a change of branding that includes other tech in the package like MFG.
Things will get changed up with FSR5 and XeSS3.

Fact is, nVidia pioneered this technology, it's what is giving them industry mindshare which is translating directly into sales, so they aren't willing to give up their leadership so easily.

Intel XeSS will run on everything.

However... Despite Intel only having a rounding error in terms of discreet GPU's (1%) they do hold the majority of graphics marketshare in the PC space thanks to their Integrated graphics which currently sits at around 60-68% of the market.

Intel is huge in this space.

And coming from an Arc notebook to nVidia RTX, it's surprising how capable their integrated graphics have become... To the point where I no longer call them Decelerators.
Even something like the Intel N300 7w TDP "netbook" chip can manage modern esport games just fine.

UDNA is coming... Rolling the best parts of RDNA4 and CDNA into one architecture instead of a bifurcated approach should mean GPU hardware development resources can be focused on a single uArch now... Will still take a few years to see the true benefits of that however.

The issue for AMD is actually their mobile hardware... Still stuck on RDNA 2/3 isn't doing them any favours. - And a big reason for that is simply the transistors.

The RDNA4 RX 9060XT is at 29.7~ billion transistors.
The RDNA3 RX 7600XT is at 13~ billion transistors.
The RDNA2 RX 6600XT is at 11~ billion transistors.

They all have the same 2048 shaders each.

So an increase of 2.5x - 3x the transistors... But only a rough doubling of performance isn't great for mobile chips that are sensitive to chip sizes.

I wouldn't be surprised if AMD goes straight for UDNA and skip RDNA4 completely for mobile hardware...
Intel Panther Lake is set to upset this market a little as well, Intel has done a brilliant job maximizing their use of transistors for their mobile graphics solutions and is bringing technology to reduce shader compilation stutter by pre-compiling shaders server-side, dynamic multi-frame gen and low latency upscaling.

That dude is fooling no one with that hair, holy shit! HAHAHA!

It's a shame Nintendo didn't go all in on DLSS. Would have liked to have seen what could be achieved with these features on it but as far as I know Switch 2 hasn't used the DLSS4 transformers model at all yet, just doesn't have enough tensor cores for more demanding DLSS features.