AMD Radeon RX 5700 Series image sharpening (RIS) examined

Quote:

---

And Radeon Anti-Lag (RAL) comes to all GCN-based or newer GPUs in Adrenalin Edition 19.7.1.

---

Read more.

Quote:

---

Originally Posted by Article

Remember, RAS only works on RDNA GPUs at this time.

---

Slight mnemonic hiccup there. I guess it's meant to be RIS. ;)

So we spend loads of GPU power on motion blur and out-of-focal-plane effects, only to throw more at it to sharpen it again? :p

Its AMD's typical commitment to keep improving older products is the reason I went for an RX480. Nice to see they don't disappoint here.

Quote:

---

Originally Posted by kalniel

So we spend loads of GPU power on motion blur and out-of-focal-plane effects, only to throw more at it to sharpen it again? :p

---

Motion blur is the first option I disable in games...

Quote:

---

Originally Posted by GinoLatino

Motion blur is the first option I disable in games...

---

It's meant to give the sense of speed when moving around but it never ceases to anger/frustrate me. They put it in because everything is blurred when you move your eyes quickly because you refocus while moving. However, from what i understand, your brain deletes the junk data while moving that's why when you look from one thing to another you don't remember the images between the object you were looking at to the object you're now looking towards.

If i can't turn it off, i struggle to continue playing the game.

The interesting thing with RIS is how it can be used to sharpen upscaled games, going on a chat EposVox had with Wendell (YouTube) it seems to do a rather good job and they speculate that AMD have drawn on their experience with upscaling consoles games to 4K.

It got me wondering why Nvidia have gone down the DLSS and machine learning route when decent scaling options are available, Intel seems to be joining AMD by going down that route with integer upscaling, maybe I'm missing something though and console like upscaling isn't suitable for PCs.

Radeon antilag tech is a very good ideia)Does another team has something like this?

Nice, although I'll stick with Corsire ;)

Quote:

---

Originally Posted by Tabbykatze

It's meant to give the sense of speed when moving around but it never ceases to anger/frustrate me. They put it in because everything is blurred when you move your eyes quickly because you refocus while moving. However, from what i understand, your brain deletes the junk data while moving that's why when you look from one thing to another you don't remember the images between the object you were looking at to the object you're now looking towards.

If i can't turn it off, i struggle to continue playing the game.

---

I nearly always turn it off. However, used right, it can be great.

In film, you want each frame to have motion blur equivalent to the frame time, so that an object moving moves continuously without interruption or stuttering. This conveniently happens automatically if you keep the frame exposed for the same length of time that it will be displayed.

In games, ideally you want the same thing. So if I'm running at 60fps then I want to render my scene as if I'd left my camera exposure open for 16.7ms. This is complex, and I don't think anyone does it in realtime (though I wonder, we're happy to draw frames in advance so can you draw a frame, draw a new frame with exactly 16.7ms time difference, interpolate the differences or blur between the two frame positions and draw the result?)
Consoles and fixed framerate games have it somewhat easier - if you know the framerate in advance then you can apply some kind of blur filter based on how quickly objects are moving relative to the camera.

Quote:

---

Originally Posted by Corky34

The interesting thing with RIS is how it can be used to sharpen upscaled games, going on a chat EposVox had with Wendell (YouTube) it seems to do a rather good job and they speculate that AMD have drawn on their experience with upscaling consoles games to 4K.

It got me wondering why Nvidia have gone down the DLSS and machine learning route when decent scaling options are available, Intel seems to be joining AMD by going down that route with integer upscaling, maybe I'm missing something though and console like upscaling isn't suitable for PCs.

---

How else is Nvidia supposed to get a return on their AI-focussed IP? :p

Quote:

---

Originally Posted by kalniel

I nearly always turn it off. However, used right, it can be great.

In film, you want each frame to have motion blur equivalent to the frame time, so that an object moving moves continuously without interruption or stuttering. This conveniently happens automatically if you keep the frame exposed for the same length of time that it will be displayed.

In games, ideally you want the same thing. So if I'm running at 60fps then I want to render my scene as if I'd left my camera exposure open for 16.7ms. This is complex, and I don't think anyone does it in realtime (though I wonder, we're happy to draw frames in advance so can you draw a frame, draw a new frame with exactly 16.7ms time difference, interpolate the differences or blur between the two frame positions and draw the result?)
Consoles and fixed framerate games have it somewhat easier - if you know the framerate in advance then you can apply some kind of blur filter based on how quickly objects are moving relative to the camera.

---

Why not just interpolate between the frame you just generated and the last frame shown to the user? It would give a slight (1/2th frame) "lag" between the centre of the blur and the real position, but that can be sorted if you track movement of parts of the scene (same way video gets compressed? Spotting blocks of pixels moving in unison?) and put half the blur behind the object and half in front

Quote:

---

Originally Posted by Corky34

The interesting thing with RIS is how it can be used to sharpen upscaled games, going on a chat EposVox had with Wendell (YouTube) it seems to do a rather good job and they speculate that AMD have drawn on their experience with upscaling consoles games to 4K.

It got me wondering why Nvidia have gone down the DLSS and machine learning route when decent scaling options are available, Intel seems to be joining AMD by going down that route with integer upscaling, maybe I'm missing something though and console like upscaling isn't suitable for PCs.

---

Because that isn't what DLSS is doing.
What Nvidia don't tell you, is that the game is actually being rendered at a lower resolution (for better performance) than what you select when you use RTX, and that DLSS is upscaling to try and disguise that.
That is why you can't have DLSS on its own, because its purpose is purely to make RTX look less bad for performance than it is.
Since it has to be trained on samples at a given resolution, you will find that uncommon resolutions can't use RTX, because DLSS has not been trained to fake it.
Nvidia even said themselves, they will later release a DLSS without the upscaling, which tells you what you need to know about what it is doing in its current form.

I say RIS is used to sharpen upscaled games, you say DLSS does do that and then say it does, I'm confused. :confused: