Support NeoGAF

[Kezen]

http://gpuopen.com/

An AMD initiative designed to enable developers to create ground-breaking PC games, computer generated imagery and GPU computing applications for great performance and lifelike experiences using no cost and open development tools and software.

I have met enough game developers in my professional life to know that these guys are among the smartest people on the planet. Those particular individuals will go to extreme lengths to extract the last bit of performance or functionality from a computer system. Graphics engineers in particular are on a quest to bend the hardware to their will. Their passion, smarts and dedication has enabled successive waves of real-time graphics technology breakthroughs in console and PC game titles. To achieve these results, developers need to gain information about the hardware, use specific tools to analyze it and write efficient code to control it. More often than not, innovative results are only possible via the exchange of knowledge that happens within the game development community. While whole conferences are dedicated to this information sharing, it is often in more modest settings that inspiration takes form. Dinner conversations, plan files, developer forums or chats are common catalysts to graphics greatness.

Yet there are hurdles getting in the way of productivity and innovation. It can be difficult for developers to leverage their R&D investment on both consoles and PC because of the disparity between the two platforms. Console games often tap into low-level GPU features that may not be exposed on PC at the same level of functionality, causing different — and usually less efficient — code paths to be implemented on PC instead. Worse, proprietary libraries or tools chains with “black box” APIs prevent developers from accessing the code for maintenance, porting or optimizations purposes. Game development on PC needs to scale to multiple quality levels, including vastly different screen resolutions. Triple monitor setups, 4K support or dual renders for VR rendering require vast amounts of GPU processing power yet brute force rendering only gets you so far. There is still a vast amount of graphics performance still untapped, and it’s time to explore smarter ways to intelligently render those increasing numbers of pixels. “Opening up” the GPU is how we solve this.

GPUOpen marks the beginning of a new philosophy at AMD. It is the continuation of the initiative we started with Mantle where we embarked upon a journey to give game developers more efficient ways to drive the GPU. A couple of years later and the game development world is now rejoicing in the advent of standard and explicit graphics APIs such as Microsoft’s DirectX® 12 and Khronos’ Vulkan™. Now is time to do even more for developers.

 

[QaaQer]

Curious to see how this gets spiun into something bad. Go go Nvidia!

 

[TouchMyBox]

I'm guessing it will probably be at least a year before we start to see any benefits from this initiative?

Hopefully this makes Nvidia be a little bit more of a team player.

 

[michaelius]

So are they actually doing something or is it more vague promises of great things to come in not defined future ?

I've seen this patter so many times from AMD already that I won't fall for it again until we see some finished products.

 

[Tarin02543]

Good news for the Personal Console

 

[Kezen]

Curious to see how this gets spiun into something bad. Go go Nvidia!

What a strange post.

 

[thematic]

I pray for this initiative to be successful
so everybody wins!

 

[LordOcidax]

Curious to see how this gets spiun into something bad. Go go Nvidia!

Ehhhhh.... Ok...

 

[digitaldevmike]

Have the good folks at WB Games seen this?

 

[Danlord]

So are they actually doing something or is it more vague promises of great things to come in not defined future ?

I've seen this patter so many times from AMD already that I won't fall for it again until we see some finished products.

Well there's all the source code for things like TressFX (used in Tomb Raider 2013 for example), a lot of SDK stuff like LiquidVR, tools that will help graphics programmers and developers with graphics optimisations, the professional applications like the HSA Compute kernal driver and also compilers, but I know little for that.

Short term, there might not be any noticeable uses but long-term, having source code for these available like TressFX could find its way into games and the tools developers use for making their games run on AMD based hardware.

This is just the start, and I'm quite impressed with how much is available straight away on Github for it. I would like to see something come about this for games and developers also, it's a nice start I think.

 

[QaaQer]

What a strange post.

Spoiler

I like to see how Ayzenberg-types shape stories, that's all.

I'm not sure if this is actually something that would drive sales, so I might be waiting in vain.

 

[Kezen]

Spoiler

I like to see how Ayzenberg-types shape stories, that's all.

I'm not sure if this is actually something that would drive sales, so I might be waiting in vain.

So scepticism in regards AMD plans and promises are the act of viral marketers ?

 

[QaaQer]

So scepticism in regards AMD plans and promises are the act of viral marketers ?

No. Next question.

 

[Kezen]

No. Next question.

Good.

Because your post was really strange, like the only reaction one should to AMD's way of doing things is unbounded admiration.

I do not suscribe to the view that AMD are complete strangers to their current situation, mindshare (or lack thereof).

Brand recognition does not build itself.

 

[ethomaz]

AMD create or try a lot of initiatives but the sensations is that it never works...

Maybe cool ideias on paper bad execution.

 

[Buggy Loop]

AMD create or try a lot of initiatives but the sensations is that it never works...

Maybe cool ideias on paper bad execution.

Uh? Care to give an example? Mantle, although they stopped working on it with the announcement of DX12 (which is mostly because of mantle..), was badass from the go.

 

[QaaQer]

...like the only reaction one should to AMD's way of doing things is unbounded admiration.

Oh God no.

Brand recognition does not build itself.

Ain't that the truth.

 

[Danlord]

Uh? Care to give an example? Mantle, although they stopped working on it with the announcement of DX12 (which is mostly because of mantle..), was badass from the go.

and as far as I'm aware, Mantle was used as the foundation which has become Vulkan, the next initiative by the Kronos group succeeding OpenGL.

 

[Datschge]

Public ISA documentation of GPUs so that they can be directly used is long overdue. Hope AMD expands this beyond GCN3 and others follow suit for their hardware.

The site is also overdue as a central place for all the different efforts that previously where scattered around rather randomly.

 

[ethomaz]

Uh? Care to give an example? Mantle, although they stopped working on it with the announcement of DX12 (which is mostly because of mantle..), was badass from the go.

Mantle is a good example... AMD do that every time since old days of ATI.

Good ideia... poor execution... it is dead.

 

[Nostremitus]

Mantle is a good example... AMD do that every time since old days of ATI.

Good ideia... poor execution... it is dead.

It became Vulkan, how is that dead?

Instead of finishing it on their own, they gave it to Kronos group. Vulkan is Mantle+.

Not to mention that Mantle forced Microsoft's hand and brought us DX12.

 

[LCGeek]

AMD create or try a lot of initiatives but the sensations is that it never works...

Maybe cool ideias on paper bad execution.

The problem with AMD time and time again is execution. Not going back till their cards have features I can use in most of my DX11 or DX9 games like Nvidia. Sad fact is this won't happen and I see them sabotaging anything good that comes of DX12.

Not to mention that Mantle forced Microsoft's hand and brought us DX12.

Which too many forget.

 

[ethomaz]

It became Vulkan, how is that dead?

Instead of finishing it on their own, they gave it to Kronos group. Vulkan is Mantle+.

Not to mention that Mantle forced Microsoft's hand and brought us DX12.

That is exactly what I call poor execution like every new initiative from AMD/ATI.

 

[Nostremitus]

That is exactly what I call poor execution like every new initiative from AMD/ATI.

How do you equate becoming a new standard from Kronos group as poor execution or failure?

How does bringing us DX12 equal a failure? AMD even stated when launching Mantle that they were doing so because Microsoft were refusing to create a low level API... It did exactly what it was designed to do.


Also, adaptive sync became a VESA standard. Intel will begin supporting it soon, and it works great with AMD cards. The only holdout is Nvidia.

 

[Kezen]

That is exactly what I call poor execution like every new initiative from AMD/ATI.

Mantle had decent uptake for a vendor specific API. Many AMD technologies have also been met with success.

 

[Rushersauce]

Curious to see how this gets spiun into something bad. Go go Nvidia!

What a weird post.

 

[Durante]

Neat, I'll try to check out and build the AOFX sample now. I've admittedly never been a huge fan of the HDAO effect (too much haloing, not enough grounded contact shadowing), but I'd like to see what you can do to tweak it.

It became Vulkan, how is that dead?

Vulkan is not as alive as I'd love it to be :/
I hope we get a full spec release soon.

Mantle had decent uptake for a vendor specific API.

I really can't agree with that part of your post. It was used in a handful or two of shipping games, and in some of those you could get worse performance than with DX with the wrong CPU/card/driver combinations.

Curious to see how this gets spiun into something bad.

Persecution complex much?

 

[clav]

Seems like a AMD PR site.

If anyone has been following the Linux Kernel scene closely, you would know AMD support is better with 4.4 and and even better on 4.5.

Expect to hear more when Ubuntu 16.04 releases.

 

[LordOcidax]

What a weird post.

Looks like a post that comes directly from the Wccftech comment section.

 

[Nostremitus]

Vulkan is not as alive as I'd love it to be :/
I hope we get a full spec release soon.

Truth. Hopefully something is released soon.
Maybe they'll have something to coincide with the launch of the new generation of GPU architectures.

(Hopefully Polaris and not Pascal. Don't want to wait that long.)

 

[Bolivar687]

The problem with AMD time and time again is execution. Not going back till their cards have features I can use in most of my DX11 or DX9 games like Nvidia. Sad fact is this won't happen and I see them sabotaging anything good that comes of DX12.

Please tell me this is a joke.

How do you equate becoming a new standard from Kronos group as poor execution or failure?

How does bringing us DX12 equal a failure? AMD even stated when launching Mantle that they were doing so because Microsoft were refusing to create a low level API... It did exactly what it was designed to do.


Also, adaptive sync became a VESA standard. Intel will begin supporting it soon, and it works great with AMD cards. The only holdout is Nvidia.

It reminds me of when posters chastise Sony for not having real exclusives, as if something is only worth celebrating if it excludes others. It's completely beyond me how anyone could call it a failure when a company invests in something great that later proliferates and benefits everyone. Hopefully the same will happen with this GPUOpen.

It's sad when the first post becomes true and yet people facetiously pretend it was "strange."

 

[Durante]

Nice, built without any issue by simply cloning the repo and loading up the solution. That's a rare treat

(Could you quote all those full-size images at least? Makes the thread hard to read)

It's sad when the first post becomes true and yet people facetiously pretend it was "strange."

Oh puh-lease.

 

[Durante]

So, the first thing I saw launching the demo application was the typical behaviour I associate with HDAO, that is very small-scale AO which overdarkens tiny crevices and most crucially introduces a lot of haloing that I find really ugly.

However, the good news is that playing with the settings in the sample (which is really convenient, I wish I had a setup like that for injecting AO!) I arrived at this:

I think that's quite a visually pleasing result. Not quite HBAO+ level, still too much overdarkening in some spots and not enough coverage in others for that, but ~2.6 ms (it actually averaged lower than what you see in the shot) on a 970 for 1080p isn't really such a high asking price for quality AO.

One issue is that I really could only get satisfactory results with Normals, which makes it much harder to use in an injection scenario.

 

[Danlord]

Nice, built without any issue by simply cloning the repo and loading up the solution. That's a rare treat

(Could you quote all those full-size images at least? Makes the thread hard to read)

I'd like to hear your insight to all of this from a (software) developer standpoint so would you be fairly active in this thread? It would be fascinating to read your thoughts on it all.

 

[Durante]

I'd like to hear your insight to all of this from a (software) developer standpoint so would you be fairly active in this thread? It would be fascinating to read your thoughts on it all.

I don't really have a direct use-case for anything other than AO right now, so I probably won't look into that, but for that particular implementation/sample:

• Works immediately on checkout, so +1 point on that
• No external dependencies for the library itself as far as I can tell, which is always a huge plus for integration
• Easily accessible (on github) -- I wish the same could be said for something like HBAO+!

What remains to be seen is how active they are in maintaining the github release (bug reports, pull requests etc.). Especially if something interesting happens, e.g. if someone offers a pull request for a shader change which improves performance by 25% on non-GCN architectures (while not affecting GCN). If some long-term commitment is there it could be a valuable resource.

 

[LCGeek]

Please tell me this is a joke.

Not it's not AA support is still shit. Same for using AO to the same degree nvidia inspector allows me too. No forcing quality upscaling/downsampling. Forcing my sync which I actually need for proper lightboost and downsampling is something I like. Cannot do this on my 7950, 4850 and two 5XXX cards.

Mentioning 3 features when inspector lets me use more complexity and ensure they always work even vs crimson drivers which from experience still had the same faults the old ones did does nothing for my existing problems if I were to use my amd card. To be fair I haven't used crimson since they updated it.Don't care for fps control and flip queue was one of the few things radeonpro allowed so it's not much has changed from before for me. I wouldn't be surprised if a few games like GTA5 and the like still perform like crap when you force them and this was not more than 2 months ago for me.

 

[Danlord]

I don't really have a direct use-case for anything other than AO right now, so I probably won't look into that, but for that particular implementation/sample:

• Works immediately on checkout, so +1 point on that
• No external dependencies for the library itself as far as I can tell, which is always a huge plus for integration
• Easily accessible (on github) -- I wish the same could be said for something like HBAO+!

What remains to be seen is how active they are in maintaining the github release (bug reports, pull requests etc.). Especially if something interesting happens, e.g. if someone offers a pull request for a shader change which improves performance by 25% on non-GCN architectures (while not affecting GCN). If some long-term commitment is there it could be a valuable resource.

Thanks for this.

What would be the biggest thing to you that AMD could open-source through this GPUOpen initiative?

Similarly, what do you think would be the biggest thing AMD could open up that would affect/help most developers, or at least a significant portion of developers?

 

[Caayn]

So, the first thing I saw launching the demo application was the typical behaviour I associate with HDAO, that is very small-scale AO which overdarkens tiny crevices and most crucially introduces a lot of haloing that I find really ugly.

However, the good news is that playing with the settings in the sample (which is really convenient, I wish I had a setup like that for injecting AO!) I arrived at this:


I think that's quite a visually pleasing result. Not quite HBAO+ level, still too much overdarkening in some spots and not enough coverage in others for that, but ~2.6 ms (it actually averaged lower than what you see in the shot) on a 970 for 1080p isn't really such a high asking price for quality AO.

One issue is that I really could only get satisfactory results with Normals, which makes it much harder to use in an injection scenario.

I'm relatively unknown when it comes to programming graphics so forgive me for my ignorance.

Are those settings that devs can set for the default AO solution in a game as a preset?

 

[Durante]

I'm relatively unknown when it comes to programming graphics so forgive me for my ignorance.

Are those settings that devs can set for the default AO solution in a game as a preset?

Absolutely. I've never seen a game which uses HDAO with a preset like that though, mostly they are much closer to the default (which, to be fair, is just 1ms rather than 2.6ms on my 970). I really think tweaking settings for 2 hours to create three performance/quality tradeoff settings for highly scalable AO like this wouldn't be too much to ask.

 

[Kezen]

I really can't agree with that part of your post. It was used in a handful or two of shipping games, and in some of those you could get worse performance than with DX with the wrong CPU/card/driver combinations.

Mantle has been used in a very decent number of games, nearly all of which performed better on Mantle than DX11 although you made the remark in the past that the comparison mostly took place between AMD's DX11 drivers and Mantle. To my recollection only Dragon Age Inquisition ran particularly dreafully on Mantle, with huge loading times and various bugs. It was not a success but I would not hold it as representative of what Mantle was actually capable of, Civilization Beyond Earth also had a very interesting Mantle implementation.
https://community.amd.com/community/gaming/blog/2015/05/12/amd-reigns-supreme-in-sid-meiers-civilization-beyond-earth

My initial use of the word "uptake" referred to the raw number of games having supported AMD's API, it certainly was not perfect but growing pains I guess.

I suppose supporting Mantle also a wise move to be better prepared for DX12.

 

[icecold1983]

its an uphill battle for sure but it would be great if this initiative took off

 

[Kayant]

I don't really have a direct use-case for anything other than AO right now, so I probably won't look into that, but for that particular implementation/sample:

• Works immediately on checkout, so +1 point on that
• No external dependencies for the library itself as far as I can tell, which is always a huge plus for integration
• Easily accessible (on github) -- I wish the same could be said for something like HBAO+!

What remains to be seen is how active they are in maintaining the github release (bug reports, pull requests etc.). Especially if something interesting happens, e.g. if someone offers a pull request for a shader change which improves performance by 25% on non-GCN architectures (while not affecting GCN). If some long-term commitment is there it could be a valuable resource.

Looks like it may take a while before that happens -

Note that we (the AMD employees working on TressFX) are not currently working directly in GitHub. We may be eventually, but for now, for these types of small, isolated changes, it is easier for me to manually incorporate the suggested fixes in our internal code, and then they will be part of the next release we push out to the public GitHub repo.

Thanks for the feedback. It is very useful.

 

[RazorbackDB]

They open-sourced their media engine.

http://gpuopen.com/gaming-product/advanced-media-framework/

Hopefully this means better Steam IHS and OBS support and maybe someone can make a Shadowplay alternative which doesn't suck