[ TPU ] AMD FSR 2.0

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.


Aug 14, 2000

Comparable IQ and performance to DLSS 2.3, proving you never needed Tensor cores for this sort of thing. DLSS is nothing more than vendor lock-in.
Last edited:


Senior member
Mar 21, 2022
This is actually quite good for AMD, as it means that their new lower end cards do better than older mid/high-end cards. If only they had the production capacity...


Senior member
Dec 31, 2016
Someone vaguely paying attention?

DF's video's on the topic have been full of what can only be described as either sabotage or mind numbingly stupid technical errors. All of which conventiently harm AMD.

And they have been doing things like that for a long time:

You cite a source known for spreading misinformation for over a decade. You discredited whatever you are trying to say.

It is not just AMD either. Look up their game reviews. Everyone knows DF is for sale to the highest bidder.
No body is perfect and everyone makes mistakes. Honestly, sometimes when different sites (and YouTube channels) are pushing out reviews and videos things may get overlooked because they want to be first. Their FSR 2.0 review was very through and covered everything well - watch and make your own conclusions. You seem hold grudge against them - like they hurt your child or something. There's no reason to take it so personally.

About FSR 1.0... Well, it did perform pretty much as well as any real-time spatial scaling could but some rabid fans overhyped that to the moon. It doesn't hold a candle to FSR 2.0 or any other proper temporal upscaling solution (like in UE5). It's better than nothing if you have 4K screen but it just doesn't look good enough at 1440p. Good that it's history now (or soon) that FSR 2.0 is out.

But enough about this. I hope AMD keeps working on this. This seems to be in a better shape than DLSS 2.0 (especially in Control). They are currently sacrificing motion clarity (or no ghosting) with reconstruction accuracy (and antialiasing quality) in motion. When performance mode is selected it will start to show. NVIDIA clearly tries to handle that situation but it may result in ghosting and/or artifacts. Judging by Intel's Dolmen videos XeSS might try DLSS like approach judging by some ghosting. It's hard to judge based on YouTube video though but I guess we'll see this Friday.
  • Like
Reactions: Tlh97 and psolord


Diamond Member
Jul 16, 2001
Digital Foundry:

Not bad for first attempt. Sharpening filter set to max can make comparison a bit hard but anyway. In static scenes quality is good (that being said, that's not hard to achieve with any temporal aa solution). It's not quite on DLSS level which is to be excepted, since FSR 2.0 is new.

In motion it has issues though.
View attachment 61489
Here you can also see how DLSS handles the signal tower thingy much better. Also the default sharpness in that game is too high for FSR 2.0 in my opinion. It causes sharpening artifacts (but you can adjust it with slider, which is nice).

Interesting to see how Intel's XeSS compares. Dolmen comes out next week and supports all three. Sadly FSR 1.0 out of the box though. Also it looks like XeSS will not need DP4a or XMX instructions support but instead will benefit from those.

So I noticed that was performance mode. I'd expect that to look like butt, even though DLSS looks ok here. What about balanced or quality? Everything I've seen/read shows that it was on par both static and in motion on balanced or quality settings.
  • Like
Reactions: Leeea


Diamond Member
Nov 16, 2006
TPU has run a series of reviews now that have FSR 2.0 hacked into DLSS.

Biggest issue seems to be some tweaking needed to get some of the ghosting under control on the part of an actual developer, but it goes to show that getting FSR up and running should be a very simple task. Getting it running well might be a bit more complicated.
  • Like
Reactions: Tlh97 and Krteq