Rumor: AMD FidelityFX Super Resolution 2.0 is scheduled to be introduced

Personally if the minimum fps drop below 60 fps too often i consider this not playable or at least not enjoyable. Can't play games anymore if i don't have a stable 60+ fps which obviously wont happen if the average is around 60.

 

Thats why you have adaptive sync monitor not to notice those drops. Personally i think 50-60fps is perfectly playable in that case.

 

Oh good -- so he balances you out:

__

I'm the same way. Honestly it's gotten to the point where 60fps probably isn't enough. I can't stand even using 60hz monitors at work anymore.. this especially became problematic when my phone shipped with a 90hz screen (4XL).

 

impossible to say that is why its better to rely on the calculations I posted above since they compare non-rt/tensor vs rt-tensor cores by nvidia themselves

if rt+tensor is the unidentified grey part it fits ~15.5-16.5x on the chip area, 1/16th is 6.25%,how that 6-7% is distributed for tensor and rt cores is unknowable
problem is we don't know


so under 15% could still be right but for Turing only.

sorry,but that is not an equivalent comparison
aufkrawall is pretty knoweldgeable
it's fine if sb hates nv,the problem is when their input ends at that.useless.

the most important part is still frame delivery
stable 55 fps plays a lot better than choppy 65
I can get used to smooth 45,I can't get used to stuttering/hitching no matter how high the framerate
so the cpu frametime delivery is very underestimated imo in reviews

 

Not sure about die space but Ampere has 4 Tensor cores (3rd generation) per SM, Volta and Turing had 8 (2nd gen) per SM.
The Tensor core is within the streaming multiprocessors.

Spoiler

 

I have low expectations of anything of quality coming out of this.

 

Its still an upscaler,no image reconstruction.doesnt mean it cant be good.

 

I tried TAAU in Watch Dogs Legion and Metro Exodus Enhanced and they look pretty bad compare to DLSS, for example TAAU at 70% looks noticeably worse than DLSS Quality (around same FPS)

 

Sunday small celebration ... They say in my village be patient and we will see in 2 days how bad or how good is what AMD will present.

 

If this comes with a slider and preset percentages... I 'd love it. As I could then I get more flexibility in tweaking settings. Allowing me to max things a little more potentially and use just a dash of res scaling. Maybe bring it down to 90% or 95% res scaling to keep me just outta the LFC zone while using Freesync. That sorta flexibility would be awesome

 

So I could not hold my hand and got elden ring . I was originally planning to wait till I get a new card. I run a 1440p monitor so I play with settings and mediu-high - max a soup of those .... The game looks fantastic but it was running 43-50 fps range ..... Not ideal for my freesync window 48-75 for the first time I went into Nvidia panel and used 70% native and 30% sharpening ..... I got in the game and I thought .... Crap it did not worked , then I saw my fps 55++ (55-60), in game I could not really tell em apart from native...... Till I saw some branches over water they had all over sharpening halos around em . I guess the difference will be Really apparent on other places too,on still images 400% zoom. My attitude toward these technologies is ... Cool more options =better , and if I can not tell em apart during gameplay .... They did their job just fine!

 

Block diagrams like that are typically meant to show hierarchy within the chip and are not representative of how the chip is physically laid out.

In the picture Martin linked though, the left side labeled "GPC" looks identical to the right side just mirrored. It labels one of those blocks as a GPC. The picture is of GA102 which has 7 GPCs but the left side only has six. So I'm not even sure what the picture is supposed to be.

 

True but a SM diagram is still a SM, but I'm probably wrong

*Edit* I just realised I was looking at Malkins img, not Martins.

 

Radeon Super Resolution
AMD Link, for streaming games
Privacy View, for those things you shouldn't be watching at work

 

If it's doing inference, then it can be better than native, because it literally dreams parts of the image that "fill it", so it puts things that there is no pixel information for.

 

@Denial Funny thing you stalked me that hard, and thats rather creepy.
Everything still stands. Im not even a FSR praiser either.

And Im not in a constantance like that pr**k is. He has been called out by quite a few. But what ever.

On topic.

Now that my 3440x1440 100hz monitor is broken and now am on a 2560x1080 60hz, I find myself downsampling 3440x1440 and 3840x1620 more then I am using 2560x1080.
Only game I use 2560x1080 in is Cyberpunk instead of using FSR. I still a believer that Native+Reshade is the way to go over RIS/Nvidias methoad with FSR/DLSS.

Im not against the DLSS/FSR as options. Its the fact these people act like DLSS is the lord and savior of everything. And now its leading developers to rely on it to make a game playable. And this is annoying like the crappy PC ports of 2006-2011.

 

that is why what I linked about tu106 and tu116 is the only thing that actually matters.the actual die size difference.
don't know what that other dude is suggesting from that die shot,and he doesn't know either.

what we do know is the difference between rt and non-rt die would be "under 15%" for turing,so that's obviously rt+tensor die space.what we know is that ampere has reduced the number of tensor cores by half,by giving 3rd gen tensor cores (1st gen was on volta) wider set of features,so they could take half the spce on ampere as they did on turing.



or maybe they just realized that the amount of tensor cores on turing just wasn't being utilized,so they could just cut it in half.3090 has about the same amount of tensor cores as 2070 Super.

 

It is more like that until now DLSS is the best alternative to get more fps without sacrificing to much image quality.
Previous FSR wasn't really a good alternative, FSR 2.0 will hopefully change that.
No end user with their mind set to the right places likes when something is propriety, but that is a complete different subject from preferring a rendering technique over another biased on performance and image quality alone.

 

So it's probably going to be similar to UE5's Temporal image reconstruction would be my bet. Since it doesn't use a deep learning component/tensor core like accelerants. My "bet" then would be DLSS > AMD Temporal FSR, but we'll have to wait and see.

Does this mean it can only be plugged into games that already support Temporal Anti-Aliasing? Weirdly iirc there was one game to use DLSS that did not have temporal anti-aliasing (Nioh 2 or something? Can't recall, it was one of those souls-like Japanese games and otherwise use SMAA iirc). Maybe I'm mistaken on that front, just thinking back to the DF videos from the time.

 

So here is a quick test to show how each solution deals with reconstruction in movement.
Same settings as before, but moving forward and taking a screenshot.
I tried lining up as best as I could, but it's not perfect.
One thing I have to emphasize is how goof TAA Gen5 is. It makes TAA less blurrier and it also reduces TAA ghosting.
But it takes around 10% of performance, on my 2070S at 1440p.

https://cdn.knightlab.com/libs/juxt...html?uid=d8f08aca-a543-11ec-b5bb-6595d9b17862

DLSS


TAA Gen5 + TAAU