News Why is AMD's FidelityFX Super Resolution Taking So Long To Develop?

[Admin]

LinusTechTips received some insider knowledge as to why AMD's supersampling tech is taking so long to develop.

Why is AMD's FidelityFX Super Resolution Taking So Long To Develop? : Read more

 

[Kamen Rider Blade]

I honestly would rather AMD take their time to get it right on the first release then do a half assed implementation.

 

[hotaru.hino]

Sometimes having a half-assed implementation but committing to improve it and showing it is better than adding delays with the excuse of "we're making it better."

Until you ship a product, you have no product.

 

[funguseater]

Sometimes having a half-assed implementation but committing to improve it and showing it is better than adding delays with the excuse of "we're making it better."

Until you ship a product, you have no product.

Tell that to CDPR

 

[hotaru.hino]

Tell that to CDPR

I mean sure, you can find plenty of examples where shipping something half-assed wasn't all that great in the end, but there are examples of ho-hum implementations that with time grew better. Such as:

• The original iPhone (though I suppose this counts more as a family of products than the product itself)
• AMD's GCN (if you want to believe it "ages like fine wine")
• The Xbox 360 (half-assed in that the hardware was poorly designed) and funnily enough the PS3
• The Xbox One in a way
• AMD's Bulldozer, though only if you compare it to Sandy Bridge and Ivy Bridge
• I could make a case for Windows Vista, since Windows 7 is the most beloved Windows OS and it's just Vista SP2 with a better task bar.
• Final Fantasy XIV

Again, until you ship a product, you have no product.

 

[TimmyP777]

Lol talk about shooting yourselves in the foot. This is obviously AMD lacking in the RnD for this and a cover up. If they had something to show they would have to compete, instead of appearing essentially worthless.

 

[Makaveli]

Lol talk about shooting yourselves in the foot. This is obviously AMD lacking in the RnD for this and a cover up. If they had something to show they would have to compete, instead of appearing essentially worthless.

And you know this because you work for AMD right?

 

[TimmyP777]

And you know this because you work for AMD right?

25 years of seeing AMD do the same things over and over and over again. Its not that hard. AMD has no answer, that is why they have nothing. It is virtually common sense. You think they wanted to take that beating by Nvidia?

 

[everettfsargent]

25 years of seeing AMD do the same things over and over and over again. Its not that hard. AMD has no answer, that is why they have nothing. It is virtually common sense. You think they wanted to take that beating by Nvidia?

Like Intel doing seven generations at 14nm! The i9-11900 CPU could eat ~400W at maximum OC's. Intel goes from 18-core HEDT to 8-core something or other. That is not treading water but drowning badly.

 

[Makaveli]

25 years of seeing AMD do the same things over and over and over again. Its not that hard. AMD has no answer, that is why they have nothing. It is virtually common sense. You think they wanted to take that beating by Nvidia?

So in other words you got nothing....

 

[chronium]

You're missing the obvious answer it took years for Nvidia to train it's AI to develop DLSS it's going to take AMD awhile to catch up since it's not a fast process.

 

[btmedic04]

You're missing the obvious answer it took years for Nvidia to train it's AI to develop DLSS it's going to take AMD awhile to catch up since it's not a fast process.

AMD has known about DLSS since at least September 2018 just like the rest of us. They should have started development then. Maybe they did, but its been almost 4 months since fidelity fx super resolution was teased as coming Soon™

 

[chronium]

AMD has known about DLSS since at least September 2018 just like the rest of us. They should have started development then. Maybe they did, but its been almost 4 months since fidelity fx super resolution was teased as coming Soon™

Nvidia has been training their AI long before 2018. The techniques to train AI got better but it still takes a long time.

 

[hannibal]

Do you remember how bad original dlls was years!
Better make it somewhat desent than suffer driver fiasko again. Better release it newer than a really buggy solution.... my 5snt...

 

[ottonis]

With resources being limited and preferredly going into Ryzen CPU development (which totally payed off), the graphics card section has certainly been lagging a bit in development.
The challenge with "true to life" - video upscaling is to find a computational solution that will a) deliver visually appealing results and b) doing so without eating up too much of the computational power of the graphics card.
DSS-like results would not be worth their money if they slowed down the frame rates by 50%.

In my opinion, now that AMD is earning good money, they really should employ and hire a few more high-level engineers that will speed up the graphics card R&D.

 

[mitch074]

AMD has known about DLSS since at least September 2018 just like the rest of us. They should have started development then. Maybe they did, but its been almost 4 months since fidelity fx super resolution was teased as coming Soon™

And DLSS 1.0 was so crappy that it took AMD a month and a half to pull out a GPU-agnostic upsampling alorithm that worked better. And Nvidia had to break their DLSS API to actually make it work right in DLSS 2.0 - so yeah, I can understand their R&D team wanting to spend some time to deliver a good first implementation of something that might become a quasi-standard in the industry.
It wouldn't be the first time, either - see Mantle/Vulkan.

 

[Jim90]

I have absolutely no issue waiting till this implementation is ready. If there's one thing we should all have picked up by now, is that QA these days is rather...different! - to how it was done before the internet era, a time many of us will remember. Today, unfortunately, much of that QA has been offloaded to end users.
Let's keep a level head and wait, a bit longer. We'll still be QA-ing, after release of course, though hopefully a fair bit less painfully.

 

[ginthegit]

I've got to give it to AMD here. I hated Microsoft back in the 95 implementation coming out with half Assed and buggy products. As an engineer, if we were to release a hardware product 80% complete, we'd be destroyed by our customers. This Patch Mentality was meant to be a Patch for unseen problems or when new tech needs some kind of patch over (or driver improvements).

I think ever since Ryzen has been smacking Intels Azz, the mentality of AMD has slowed down a bit and they seem to be focused again on Quality. Being as ATi and AMD both sucked with drivers and software, I think they are trying to improve in this too. Remember that AMD does 2 big Hardware products, and has a fraction of R&D for both, so something must give, and it is doing very well against the two bigger giants.
Lisa Su has worked some miracles, and now she wishes to take the bad software out of the equation. Respect the AMD!

 

[TimmyP777]

So in other words you got nothing....

I have common sense. If AMD had a competitive product that could have lessened the beating they are currently taking, they would have for sure released it. Even if its "half baked" because thats what AMD does. Look at OpenGL suppor ton Windows. It is nearly 50 percent faster across the entire line of radeon cards for the past 15 years on Linux . They dont have the manpower to keep up with development for Windows and compete with Nvidia at the same time.

Sorry thats just the way it is, and the way it has always been.

 

[TimmyP777]

And DLSS 1.0 was so crappy that it took AMD a month and a half to pull out a GPU-agnostic upsampling alorithm that worked better. And Nvidia had to break their DLSS API to actually make it work right in DLSS 2.0 - so yeah, I can understand their R&D team wanting to spend some time to deliver a good first implementation of something that might become a quasi-standard in the industry.
It wouldn't be the first time, either - see Mantle/Vulkan.

Yeah, keep spitfiring bs. You have zero idea what you are talking about, so why even try?

*and why are you bringing up Mantle? Its was a failed joke that lead to Vulkan. Vulkan was not developed by AMD.

 

[hotaru.hino]

I get the mentality of "they should release it when it's ready," but the problem is that "ready" is still subjective. Is it ready when it's 100% bug free? Well, you'll be waiting forever because as long as software is written by humans, it'll never be bug free. Will it meet some arbitrary expectation of quality? Well, they better become psychics and test it against what your use cases are now and in the future.

However, if you're trying to make money off your software in some form or fashion, at some point you're going to have to but the brakes on development and just release the damn thing. Lest your software becomes vaporware or you've hyped it up so much that it'll never live up to the expectations people have come up for it. My issue isn't so much if the software is buggy as hell (I mean, depending on what I was trying to use it for, it may be an issue), my issue is if the developer doesn't think it's a problem and doesn't address the issues that people bring up. Software is always going to be a work in progress until the developer no longer maintains it.

At the end of the day, the longer you keep saying "we're making it better!" the more you're compounding the hype. If FFX SR gets released next year, NVIDIA may as well have a DLSS 3.0 that's a driver wide solution and FFX SR no longer looks as good as it did.

 

[Bamda]

"AMD also wants to make Super Resolution GPU agnostic, just like other FidelityFX libraries. That potentially means supporting many generations of GPUs, including Nvidia and possibly even Intel options. " That might be a bit too bold, I hope they know what they are doing.

 

[mitch074]

Yeah, keep spitfiring bs. You have zero idea what you are talking about, so why even try?

*and why are you bringing up Mantle? Its was a failed joke that lead to Vulkan. Vulkan was not developed by AMD.

You are dead wrong on both counts.
AMD gave their Mantle API specs to the Khronos Group that then used it as the basis for Vulkan. It was rewritten top to bottom with input from almost all industry actors - including AMD. As such AMD authored a substantial part of it.
Moreover if Mantle was such a joke, why did Intel publicly asked for the spec so that they could implement it too? https://techreport.com/news/26682/intel-asked-amd-for-mantle-api-spec/
Read it up : https://www.khronos.org/news/press/khronos-releases-vulkan-1-0-specification

 

[mitch074]

I have common sense. If AMD had a competitive product that could have lessened the beating they are currently taking, they would have for sure released it. Even if its "half baked" because thats what AMD does. Look at OpenGL suppor ton Windows. It is nearly 50 percent faster across the entire line of radeon cards for the past 15 years on Linux . They dont have the manpower to keep up with development for Windows and compete with Nvidia at the same time.

Sorry thats just the way it is, and the way it has always been.

AMD have always acknowledged that their official OpenGL driver is only here to manage professional applications and their 20+ years of cumulative quirks - there IS a reason why Nvidia sell Quadro cards with a different driver package (verified with Catia and AutoCAD : AMD mainstream cards work with them, Nvidia don't).
The Linux MesaGL driver, which is mostly compliant with OpenGL with a focus towards being API - correct with only a few disabled by default quirks built in is a collaborative project; for quite a while it was a joke, but since Ati dumped their hardware specs into it, Intel poured resources in it too and finally AMD dedicated a couple developers to it, it got rather good - enough that AMD recommends it for day to day use over their own.
It became fast as heck when Valve decided that they needed to secure themselves an OS where the author can't lock them out and threw a boatload of resources into it.
Remember : make it work, make it right, make it fast.

 

[Conahl]

Lol talk about shooting yourselves in the foot. This is obviously AMD lacking in the RnD for this and a cover up. If they had something to show they would have to compete, instead of appearing essentially worthless.

yea and if they did release it, and it was buggy, or not as good as even dlss 1, you would be complaining that amd should of waited till it was on a better state. so either way, you would be complaining about it.