AMD FidelityFX Super Resolution in Ultra Quality shows visible quality loss

you don't need rt cores to do rt either
see ? gtx1080ti runs it too

and no hardware will make it...... ?

to identical final result and performance ? interesting cause i've never seen any publication comparing hardware and software dlss,but if you claim the results are the same...I mean you gotta have data confirming it,you're just not willing to share it.

 

I simply used a simpletons terminology - you do need certain HW to accelerate the calculations of the matrix multiplication to produce the picture in real time..... sure you can do it using other methods (run a certain algorithm on your "normal" cpu/gpu cores for example) but it will result in the rest of the HW waiting for the solution of such calculation - AMD's "solution" for upscale is nothing more than a gimmick for people captured by hypes and misinformation.

 

I don't think at this point we can say it's a gimmick,we'll see.we don't know how it works yet,maybe it's not just a spatial upscaler like wccf called it and will develop.
but it's funny people seeing one mostly static 20s long scene with 90% being underwater rocks and say nvidia is done with dlss.

how can a person who smells marketing bs everywhere just gobble up whatever amd tell them like a pelican without a thought.

 

Nowhere did I make any claims about performance with respect to speed of execution, but you can continue to fill in the gaps to support your own narrative...

 

then why are you implying that
you're being consciously misleading.
if the results are different in hardware vs software,then spell it out.

 

That's OK buddy, I've been reading this website long enough to know what your commentary looks like... Off you toddle...

 

those are just facts.
they can be confirmed by looking at calendar.
he said it will be hard,not impossible.it's no secret nvidia is longer into ai hw than amd and made more progress there.what good does questioning this fact have ?

how about you stop implying things that are not true if you still wanna contribute anything meaningful here

 

I haven't implied anything that's untrue. As a software and data engineer, I can't be held responsible for your lack of understanding.

Now, I know you like to agitate people, so as I said, unless you have something useful to contribute, off you toddle.

 

fine snowflake,but you're still the one that's deluding yourself.
if you don't wanna hear this maybe block me instead of telling me to scram.

and by the way I'm not questioning what you said,be it true or not.I'm just saying out loud what you intentionally omitted.

 

You really can't help yourself can you?

 

There's no perfect solution available. Nvidia's tensor cores offer better quality, but will be just a waste of silicon if you don't use DLSS. As far as I know, no game studio has used them for anything else (might make the game wholly incompatible for anyone without a minimum amout of tensor cores). AMD's cheaper solution doesn't waste silicon if you don't use it, but it also won't ever reach the same level of quality as DLSS with tensor cores. If miners hadn't destroyed the video card market, one could at least try to ponder if AMD lacking tensor core equivalents made more VRAM available for the same total price, but as things are, talking about prices is meaningless.

 

hmmm well a good upscaler is welcome especially if it works universally .... i mean who ever run 1080p on a 1440p today with out dlss ... knows very well that this is a huge improvement, dlss 1.0 was huge improvement over that too although i believe NATIVE resolution will always be king .... then again ... more options = good in my book ... i will wait for the reviews when it comes out though , literally i had a plan since last spring to snatch a 3060ti or 3060 on lunch date then i saw the prices around here day 1 over 2.5 msrp ...so i decided to wait and sit on my 1060.... fsr atm would have been very welcome XD

 

all you've done is shown you're an atypical ignorant amd consumer with no clue on how image substitution techniques work.

anything using 16bit floats is using tensor cores, but as their intended Tensor matrix function, no, because DirectX doesn't have a way to use it in a build outside of the insider channel.

 

I see you're rude too. I guess if having one AMD video card in the last 30 years, and Nvidia cards all the way back to the Geforce 256, then I guess I'm guilty as charged.

You are correct though, I don't know how image substitution works, but I have spent over 25 years as a software and data engineer, and spent 12 months doing machine learning and training AI models on distributed compute platforms... Unfortunately, all very boring.

 

This. I'm certain it will get better over time.

Besides, if you can't see a difference at a glance and only through nitpicking - then that tech might as well be exactly what I'm looking for.

 

With technologies like these, there is a big subjective part to them. Some people might be perfectly okay with the image quality and welcome the performance boost, others might be way more sensitive. So until we get 3rd party reviews, but more importantly, the ability to try it out ourselves, any discussion on it's merits/demerits is kinda useless.
Given the wide HW support and the fact that you'll get it for free I don't see any point in bashing it. It's just another tool in the toolkit that may or may not turn out to be useful.

In any case, kudos to AMD for making it open. Hopefully, it doesn't take a lot of effort to implement on the game developer side and we'll see a steady flow of titles starting to support it.

 

There was a noticeable image quality loss in the article and more importantly oranges were compared with oranges, both sourced from same youtube extraction method.
IMO it's no biggie at all, i'd like them to up their RT game for next gen, amd cards are known for high frame rates anyways thats the way i see it.

 

I've been using DLSS in Fortnite and Control recently. It's very impressive. It saves a lot of GPU load and the IQ still looks great.

 

You actually don't.