DX12 / Vulcan and new GPU architecture (?)

[TheELF]

True with DX11. Mantle, DX12, Vulkan, will address that. According to Johan, DX11 is "painfully serial" and that's it's biggest shortcoming.

That's exactly what people where saying about games before ps4 ,game code is soooo serial and ps4 will fix that.
It will be the same for the vga, parallel stuff is great but it will still come down to the speed of the main effects.



People seem to think that dx12 will affect the cpu in some way and make the game code more parallel or something.

 

[3DVagabond]

That's exactly what people where saying about games before ps4 ,game code is soooo serial and ps4 will fix that.
It will be the same for the vga, parallel stuff is great but it will still come down to the speed of the main effects.



People seem to think that dx12 will affect the cpu in some way and make the game code more parallel or something.

I understand that it's more out of order. You don't have to necessarily complete one task before running another which allows for less idle time.

 

[Azix]

That's exactly what people where saying about games before ps4 ,game code is soooo serial and ps4 will fix that.
It will be the same for the vga, parallel stuff is great but it will still come down to the speed of the main effects.



People seem to think that dx12 will affect the cpu in some way and make the game code more parallel or something.

It should affect the CPU. Almost all the demos people have shown mention using multiple cores better rather than just one core for the api/driver. Soemthing like that anyway.

Ps4 is already like dx12 but ps4 can't really fix the PC side of things without a new API. dx11 was still dx11 regardless of ps4. Now things should change some.

 

[Alessio1989]

It is always funny see people ranting about API conspiracy theories just because a single draft introduction manual page was copied from another API...

I've programmed enough to know that documentation wouldn't be that similar if I wrote documentation for a piece of my own code two different times.

You could argue that DX12's error handling is similar to Mantle and MS plagiarized the documentation, but there is no chance in hell it is a coincidence.

You probably have not programmed enough.
Mantle error handling is the same error handling used in tons of Win32 API, from Direct3D 11 to WioAPI (windows.h).
HRESULT, aka returning an integer for error handling, is a very common low-level error-handling technique. Moreover it was obvious that MS were going to remove a lot of runtime checks. Who wrote that page just made a bad homework, causing an embarrassing result.

 

[Silverforce11]

On one hand, we have a lot of game devs who make game engines for a living, some of whom are very renown, say Metal, Vulkan and DX12 are similar/deja-vu/variants of Mantle... on the other, a bunch of forum warriors who disregard the evidence and continue to spread fud...

Not sure why you guys bother. Until you're a software engineer responsible for making major game engines, please don't talk against those guys, its disrespectful.

 

[AnandThenMan]

MS makes amazing new API (by secretly partnering with DICE/Johan's team for years!), shares with AMD (or AMD steals it), and allow AMD to share it with Apple & Kronos..

About as possible as "mankind never landed on the moon" tinfoil theory.

Nailed it. There is simply no way AMD copied work done by MS and released it as Mantle and that morphed into Vulkan. MS would sue the pants off of AMD.

 

[railven]

That's exactly what people where saying about games before ps4 ,game code is soooo serial and ps4 will fix that.
It will be the same for the vga, parallel stuff is great but it will still come down to the speed of the main effects.



People seem to think that dx12 will affect the cpu in some way and make the game code more parallel or something.

Pretty much where I'm at right now. I'm eager to see how the new DX12 era affects AMD's hardware. A portion of it is because it would be nice to see Radeons in the limelight again. And also because some people here are so invested if it fails I worry for their safety.

End of the day, I see Nvidia pulling a rabbit out of it's hat. Call it Gameworks if you will. /shrug

 

[omek]

IIRC UE4 is capable of doing voxel cone tracing which I believe is a simplified version of ray tracing rendering by method of ray scattering. Because ray tracing is compute orientated it could give AMD another leg up as long as engines start to take advantage of it.

 

[Alessio1989]

On one hand, we have a lot of game devs who make game engines for a living, some of whom are very renown, say Metal, Vulkan and DX12 are similar/deja-vu/variants of Mantle... on the other, a bunch of forum warriors who disregard the evidence and continue to spread fud...

Not sure why you guys bother. Until you're a software engineer responsible for making major game engines, please don't talk against those guys, its disrespectful.

I am not a software engineer for making a major game engine, but I am a member of the DX12 EAP, I have accessed to a lot of NDA and unpublished documentation, I know piece of D3D12 API evolution that most people don't, so I am not spread fud.
That page of that stupid introduction programming manual (not the IHV/ISV specifications) was the ONLY that was shamelessly copied. The D3D12 API has evolved in the years, in the months, is still evolving (despite of what states the MSDN library) and will continue to evolve after Windows 10 RTM for PCs. If we consider WDDM 2.0, the first information that we can find come from WinHec 2006, but this will drive the discussion far away....
Of course they have a lot of similarities, they need to do quite the same jobs on basically the same class of hardware. Moreover, more low-level you go , less possibilities you have to do the right thing in the correct way on the same hardware. A lot of devs could also argue that Mantle has a lot of similarities with LibGCM (despite the different hardware), that's obvious and couldn't be the opposite.
Finally, don't forget that DICE (better, EA) invested al lot of money into Mantle implementation for their proprietary engines, it perfectly makes sense they will defend the API they made for/with AMD.

Nailed it. There is simply no way AMD copied work done by MS and released it as Mantle and that morphed into Vulkan. MS would sue the pants off of AMD.

No-one stole nothing from anyone. Do you guys never ever thought simply that AMD played an important role as one of the 3 major GPU HIVs in the D3D12 development process?

 

[Azix]

No-one stole nothing from anyone. Do you guys never ever thought simply that AMD played an important role as one of the 3 major GPU HIVs in the D3D12 development process?

Not sure who is talking about theft, but mantle contributing to dx12 is what I see people say and you seem to be confirming that.

 

[werepossum]

I don't think anyone is seriously saying that AMD or Microsoft stole anything. People ARE saying that DX12 and Mantle are extremely similar, and given that AMD is also partnered with Microsoft on DX12 development, we are all wondering how similar. Conceptually similar is I think a given. Command level? I think that is possible, given again that Mantle is intended to make AMD architecture more competitive and a new Microsoft DX that accomplishes that without the massive resources needed to support Mantle is obviously in AMD's best interests. I think here that the counter points are Microsoft's interests, the XBox One first but narrowly behind that, having Windows adopted and enjoyed by as many people as possible. Therefore while Microsoft certainly wants to move GPU architecture toward the XBone architecture, the Windows DX12 team is obviously going to do everything possible to maximize performance on current NVidia hardware and even Intel hardware. Otherwise they sabotage Windows 10 adoption, which means more years supporting Vista, 7 & 8 for no gain.

It is difficult to determine just who led and who followed, but I suggest that AMD in giving Mantle to the Vulcan team argues for AMD leading. AMD has an interest in AMD cards working well for Vulcan; I do not think Microsoft has that interest. Therefore I am guessing that AMD owned more of DX12 than Microsoft owned of Mantle, or that the similarities are largely conceptual rather that code-level.

 

[Silverforce11]

You don't have to worry about NV's DX12 performance. GW updated libraries for FL12_1 which only Maxwell 2 support. Buy a few more AAA devs to implement it, they win.

 

[Alessio1989]

Not sure who is talking about theft, but mantle contributing to dx12 is what I see people say and you seem to be confirming that.

I said AMD, not mantle. Of course AMD had a more recent study case (Mantle) compared to NVIDIA (libCGM) or Intel, but that does not means or state that D3D12 start with Mantle as base API.
I know that without public documentation no-one cannot create a valid bullet-proof history reconstruction of the events, but that does not means that some "journalists" are allowed to let the wings of their imagination fly and wrote what they dreamed. If someone here has the rights to sign-up into the DX12 EAP (which is still active!) and wants to know more about that, ie will able to find some interesting legacy documentations (of course NDA) about.

...Terefore while Microsoft certainly wants to move GPU architecture toward the XBone architecture, the Windows DX12 team is obviously going to do everything possible to maximize performance on current NVidia hardware and even Intel hardware. Otherwise they sabotage Windows 10 adoption, which means more years supporting Vista, 7 & 8 for no gain.
...

There is no such thing like "maximize performance". Never had been, that's not the use-case of a graphics API. The DirectX Development team made a pretty nice work creating an API that fits with such a huge class of different hardwares, not only of AMD, NVIDIA and INTEL but also other IHVs. I am not aware about any aspect of the API that can cause notable issues on a particular hardware implementation.
I also expect that Vulkan will made the same, but I still have the fear KG will screw up the things again in the name of the personal interests of every single member of the ARB.. Hopefully I will be wrong about that.

 

[Silverforce11]

Not journalists Alessio. Software engineers at major companies responsible for the big game engines. Not just Johan @ DICE either. Look on twitter for #Mantle & #DX12 chatter from way back.

What you're saying is DX12 just became very similar to Mantle by pure coincidence.

I recall when AMD made the Mantle SDK beta available for gamedevs who sign up, it was discussed as an "early access to accustom oneself to DX12 programming".

 

[Alessio1989]

Mantle SDK beta access started after Mantel was out, games using Mantle were out before. DX12 EAP started when D3D12 and related technologies was still in development, an advanced state of development thought, but it was not done. DX12 EAP is not just a private and preview access to the API, we receive and give feedback on the entire development process.
Also, looking on twitter to retrieving information is a joke, isn't that?

 

[Silverforce11]

It's no more a joke than looking to these tech forums. At least on twitter there's a real face with a real job (software engineers) to their names. A heck of a lot more reliable than randoms on here.

 

[Alessio1989]

On twitter you will post what best fits in 140 chars + /null.

 

[Azix]

On twitter you will post what best fits in 140 chars + /null.

only takes so many words to say something important.

I'm your father.

See? Just like that I turned your whole life upside down. Not even close to 140 characters.

 

[railven]

It's no more a joke than looking to these tech forums. At least on twitter there's a real face with a real job (software engineers) to their names. A heck of a lot more reliable than randoms on here.

I dunno, I've come to not trust Twitter. Did you see the recent fiasco with Gawker? Tim Hunt? Should I continue?

Shoot, JFAMD had a name and a real job and how did that turn out?

What I'm seeing now with the Gawker fiasco is people who are employees/fans of Gawker defending what Gawker did, I mean that is expected right?

 

[Silverforce11]

Let's discredit all the good people because you don't trust twitter (twitter is just a form of communications btw, what you are saying is you don't trust people). -_-

Only difference between these tech forums where anonymity is a cover, some folks on twitter are well known IRL.

I mean if the creator of Mantle, Johan (leaving all other devs aside) says DX12 is very similar, then I trust his words before any forum warriors.

 

[werepossum]

I said AMD, not mantle. Of course AMD had a more recent study case (Mantle) compared to NVIDIA (libCGM) or Intel, but that does not means or state that D3D12 start with Mantle as base API.
I know that without public documentation no-one cannot create a valid bullet-proof history reconstruction of the events, but that does not means that some "journalists" are allowed to let the wings of their imagination fly and wrote what they dreamed. If someone here has the rights to sign-up into the DX12 EAP (which is still active!) and wants to know more about that, ie will able to find some interesting legacy documentations (of course NDA) about.

There is no such thing like "maximize performance". Never had been, that's not the use-case of a graphics API. The DirectX Development team made a pretty nice work creating an API that fits with such a huge class of different hardwares, not only of AMD, NVIDIA and INTEL but also other IHVs. I am not aware about any aspect of the API that can cause notable issues on a particular hardware implementation.
I also expect that Vulkan will made the same, but I still have the fear KG will screw up the things again in the name of the personal interests of every single member of the ARB.. Hopefully I will be wrong about that.

C'mon, "no such thing like "maximize performance""? If NVidia and Intel hardware run like crap on Windows 10, then Windows 10 fails, period. No way in hell am I buying that Microsoft isn't concerned with maximizing how Windows 10 and DX12 perform on all platforms. Maximizing performance is the entire reason to create yet another Direct X system, much less one that radically departs from the existing software.

Again, no one (or almost no one) is claiming that DX12 (or the D3D part of it) used the actual Mantle code as its starting point, just its concepts. No way in hell are the two so conceptually similar as to use parts of the documentation word for word by coincidence. I'd also bet that some (maybe a lot) of the data structures are at the least function-for-function identical. AMD has a vested interest in its hardware running well within DX12 (especially with Microsoft taking over development and maintenance, which saves AMD a huge amount of money that would be spent on Mantle) and Microsoft has a vested interest in developing the best product in the least amount of programming time.

 

[werepossum]

You don't have to worry about NV's DX12 performance. GW updated libraries for FL12_1 which only Maxwell 2 support. Buy a few more AAA devs to implement it, they win.

I suspect you are correct. I think Microsoft will make sure there is a good NVidia path within DX12, and NVidia will pour money into supporting developers to make sure they are including those calls with the implementations that work best for NVidia hardware. It's a win-win-win scenario as Microsoft, NVidia and developers all want new games to run well on NVidia hardware. It's also trivial compared to the challenge of programming for the PS3 and that didn't stop developers.

I dunno, I've come to not trust Twitter. Did you see the recent fiasco with Gawker? Tim Hunt? Should I continue?

Shoot, JFAMD had a name and a real job and how did that turn out?

What I'm seeing now with the Gawker fiasco is people who are employees/fans of Gawker defending what Gawker did, I mean that is expected right?

But Silverforce's point is that these Tweets occurred long ago. If they were lying, we* should have seen other people calling them out and saying that DX12 and Mantle are not at all similar. At the least, we should have seen Microsoft people saying "WTF? We're starting from scratch here!"

* By "we" I mean everyone but me since I'm not a Twit. Um, a Twitterer. Whatever, I don't use Twitter.

 

[Alessio1989]

Let's discredit all the good people because you don't trust twitter (twitter is just a form of communications btw, what you are saying is you don't trust people). -_-

Only difference between these tech forums where anonymity is a cover, some folks on twitter are well known IRL.

I mean if the creator of Mantle, Johan (leaving all other devs aside) says DX12 is very similar, then I trust his words before any forum warriors.

I do not discredit those people, I sample state that you cannot explain well the things with a so limited number of characters. Twitter is a BAD form of communication, it limits your communication. "Very similar" doesn't mean a plagiarism or whatever have been said in these and other forums.
Try, as example, to explain the differences between Mantle and D3D12 in 140 characters. Try to do the same between LibGCM and Mantle. You simply cannot.

 

[Alessio1989]

C'mon, "no such thing like "maximize performance""? If NVidia and Intel hardware run like crap on Windows 10, then Windows 10 fails, period. No way in hell am I buying that Microsoft isn't concerned with maximizing how Windows 10 and DX12 perform on all platforms. Maximizing performance is the entire reason to create yet another Direct X system, much less one that radically departs from the existing software.

Again, no one (or almost no one) is claiming that DX12 (or the D3D part of it) used the actual Mantle code as its starting point, just its concepts. No way in hell are the two so conceptually similar as to use parts of the documentation word for word by coincidence. I'd also bet that some (maybe a lot) of the data structures are at the least function-for-function identical. AMD has a vested interest in its hardware running well within DX12 (especially with Microsoft taking over development and maintenance, which saves AMD a huge amount of money that would be spent on Mantle) and Microsoft has a vested interest in developing the best product in the least amount of programming time.

An API is not a magic tool that maximize hardware performance. You - developer - have to maximize your application performance through that tool. APIs like LibGCM, Mantle, Metal, Direct3D 12 and Vulkan are meant to reduce the impact of the API runtime on your application, to reduce the driver background workloads and give you less abstraction as possible to better control the hardware.
Just because you make an #include <d3d12.h> doesn't mean you are going to receive a free performance boost into my application. You are still able to make crap like like this, probably if you are a bad developer you will produce more "things" like that with a low level API.

AMD and DICE did not invented the wheel with Mantle. Mantle simply came out first, it was designed for a limited range of hardware and it was designed to run against WDDM 1.x. It was simple to put it out first. Mantle has a huge credit: it showed first how much limited were Direct3D and OpenGL with current modern hardware and it showed a part of the solution. No-one can deny that.
However, Direct3D 12 is meant to run with a larger range of hardware (which is not an easy challenge), it's meant to run with a new driver model (which required a lot of work by Microsoft), moreover it's meant to work with older and legacy graphics Windows API too (even if that's was not a primary goal though) and add set of new capabilities too (like SM 5.1), and there are the graphics tools. Build Direct3D 12 required a lot of work compared to Mantle. It is obvious that AMD and Dice where able to put it out first.

Intel and NVIDIA hardware do not have any issue caused by WDDM 2.0 and the D3D12 structure. I do not understand where such concerns come from. If we speak about rasterization, Maxwell 2.0 is the most advanced GPU currently on sale, no-one can negate that.

Finally you cannot directly compare the amount of money AMD invested into Mantle to the money Microsoft and it's partners invested into D3D12. There is at least one order of magnitude in difference, maybe two orders of magnitude. AMD itself stated they didn't invest a huge amount of money into mantle. If you wonna find someone that invested a lot of money into mantle it's called EA.

 

[railven]

But Silverforce's point is that these Tweets occurred long ago. If they were lying, we* should have seen other people calling them out and saying that DX12 and Mantle are not at all similar. At the least, we should have seen Microsoft people saying "WTF? We're starting from scratch here!"

* By "we" I mean everyone but me since I'm not a Twit. Um, a Twitterer. Whatever, I don't use Twitter.

Oh, guess I forgot to post last night.

"I didn't imply discredit anyone, I meant be skeptical."

Twitter has proven time and time again to be a terrible avenue to share information. Someone above said it best and the example "I'm your father" is perfect.

People post something quick - spur of the moment then it gets regurgitated and we start playing the telephone game until some site writes an article and it becomes almost dogma.