Open Source AMD FidelityFX Super Resolution Impresses In PC Game Tests (hothardware.com) 35
MojoKid writes: AMD's FidelityFX Super Resolution (FSR) PC graphics up-scaling technology is ready for prime-time and the company has allowed members of the press to showcase performance and visuals of the tech in action with a number of game engines. AMD FidelityFX Super Resolution is vendor-agnostic and doesn't require specialized hardware to function like NVIDIA DLSS, which relies on Tensor cores on-board NVIDIA Turing or Ampere GPUs to accelerate neural network models that have been specifically trained on game engines. In contrast, AMD FSR utilizes more traditional spatial upscaling to create a super resolution image from a single input frame, not multiple frames. AMD FSR then employs a library of open-source algorithms that work on sharpening both image edge and texture detail. In game testing at HotHardware, frame rates can jump dramatically with little to no perceptible reduction in image quality, and the technology even works on many NVIDIA GPUs as well. There are currently 19 titles that are available or planned with support for AMD FSR, but with the open nature of the technology and cross-GPU compatibility, game developers theoretically should have significant incentive for adoption to breath new performance into their game titles.
Re: (Score:2)
Much like Freesync [wikipedia.org] vs G-sync.
Re:AMD vs NVDA (Score:5, Insightful)
Re: (Score:2)
Re: (Score:2)
Did FXAA help nvidia stock?
Because we're looking at a shader that is universally supportable on modern GPUs that appears to perform better than alternatives. And the main opponent is massively invested in expensive, trainable AI that only works on its most expensive hardware and needs to be carefully adapted for every new game (read: costs money to the company to make work).
Re:AMD vs NVDA (Score:4, Informative)
DLSS 2.0 does not need to be trained on every game, like DLSS 1.0. The difference between Nvidia's and AMD's approach is that DLSS uses spacial and temporal data, while AMD's FSR uses spacial data only.
https://www.nvidia.com/en-us/g... [nvidia.com]
Re: (Score:2)
FSR if you read the description is an open shader like FXAA. It's very much a drop in. That's the entire purpose of it, and it appears to be replacing the previous AMD's take on a similar tech, but much more advanced.
Re: (Score:2)
I got the impression that FSR also uses temporal data, it just doesnt know how detail should look like as it is general scaler, not game refined AI.
Re: (Score:2)
I don't know, but here's what the article says:
"NVIDIA relied on Turing and Ampere's tensor cores to accelerate an AI model using temporal and spatial data to weave in detail where it isn't. The effect is pretty convincing, but it absolutely requires NVIDIA's proprietary core technology. That by itself has limited adoption. Instead, AMD uses just spatial data, meaning FSR relies solely on information related to the current frame instead of trying to blend multiple frames together as NVIDIA does. "
It comes a
Re: (Score:1)
[FSR] appears to perform better than alternatives
the more technically-inclined sites/parties, like digital foundry, are saying the current implementation of FSR is a bit shit compared to alternatives, including checkerboard upscaling, with regard to image quality.
time will tell with FSR, but it's nice more upscaling techniques are being developed. increasing the native res can be extremely expensive, and native 4k isnt nearly enough to resolve jaggies and other IQ issues.
Re: (Score:2)
*depends on game and implementation
**depends on capabilities of hardware
Re: (Score:3)
This should help AMD stock, because giving away stuff for free is exactly what excites Wall Street.
This isn't about giving stuff away for free, this is about neutralising a key advantage of the main competitor so they can sell more of their cards rather than having buyers think that games perform and look better on Nvidia due to DLSS.
It's a shader. (Score:4, Informative)
AMD says FSR is easy to integrate, as the company intends to make the shader source code available as part of the GPUOpen project under the permissive MIT open-source software license. The company expects source to be published sometime in mid-July. While AMD points out that FSR doesn't use temporal data or specialized hardware, there's no per-game training required
So it's a sophisticated shader and thus should work for any game that employs it's use. Quake II is gonna be lit. ;)
Re: (Score:2)
It has been speculated that it could be retrofitted to games by modders, even if they don't have the original source code.
Essentially it needs two things. First it has to be inserted into the render pipeline at the right point, after the main scene is finished but before the HUD is generated because the HUD is rendered at full resolution for clarity.
Second it needs to be tuned to the specific game, something that is likely to be quite easy. So it's really just the insertion that might be tricky.
Upscalers suck (Score:1)
I don't understand the point of rendering at low resolution and using upscalers. If there is a time budget wouldn't it be better to intelligently work out what is important when rendering to meet targets instead of this crap?
Re: Upscalers suck (Score:4, Informative)
Re: (Score:2)
What is important? The pixels showing the ennemy, the mosaic on the wall I like to discover, the shadow that change with the folliage of trees and perhaps the ennemy comming in my back? Even if I take the example of the shadow that in the beginning was not important and for that badly renderend to save some time budget and because an ennemy is comming in my back the sahdow became very important for the clue that there is something/someone in my back, the change in the quality of the shadow would kill any su
Re: (Score:2)
What's not to understand? You get a bunch of budget back from rendering a at a lower resolution and present an image that is 98% as good as rendering at a higher resolution. That's budget you just don't have on a reference spec at that higher resolution.
Think of it in reverse. Say you target 55fps: you can't get anywhere near rendering at resolution A on a certain hw spec, but you can render at B, so you render B - a very small number and use that small number to upscale to A. Even if you used that time to
Oh yeah! Woo-hoo! (Score:4, Insightful)
I can't wait to play Leisure Suit Larry 3 in super high resolution!
Re: (Score:2)
I'd rather play Leisure Suit Larry 4 in ANY resolution!
Re: (Score:2)
I wouldn't try to find LSL4 if I were you, to this date there is no known fix for the Vohaul virus.
What would be really cool (Score:2)
Is it you could upscale and old game with this
Re: (Score:2)
The fact that neither vendor will let you take control of your own framebuffer, not letting you use your own post-process shaders for _everything_... its rather disappointing.
Sure you can (sometimes) use things like ReShade, but thats only for specific DirectX/OpenGL applications... ReShade doesnt work with Windows Media Player even though it too uses DirectX, yeah? The operating system is also growing more and more hostile to reshade, blocking some shader injections.
The point of course is that
Re: (Score:2)
Re: (Score:1)
So instead of people cheating in games, we'll have people cheating in games?
Re: (Score:2)
Is it you could upscale and old game with this
This could make Pong SO AWESOME.
Loved the news! (Score:1)