AMD GPU14 Tech Event Sept 25 - AMD Hawiian Islands

[EightySix Four]

Different topic, footnote:
*On whether Mantle truly is a close-to-the-metal API: I am getting mixed signals from AMD, or the press is misreporting. If Mantle is truly a close-to-the-metal API, then it cannot be "open" or cross-ISA/arch, because being locked to the specific arch is inherent to bare metal coding. Even now (decades since they arrived), we don't have assemblers that produce one type of machine code that can be run on any ISA like x86, ARM flavor, PowerPC, etc. It is simply impossible, because that machine code has to have a 1-1 target on the hardware side, and that is defined by the specific arch. But now there is talk and "confirmation" of Mantle being open or some such silliness.

You can't have it both ways. Either you are truly a close-to-the-metal API (therefore locked down to your target hardware arch), or you are still an abstraction layer (therefore can be "open" and used by other competing hardware archs). It can only be one or the other. If the press reports of being "open" are true, then this definitely isn't close-to-the-metal as we understand it, and simply another abstraction layer that supposedly has lower overhead than DX (i.e., far less useful and slower than if it were truly close-to-the-metal as initially thought). If it's just another abstraction layer that is abstracted high enough to accommodate multiple archs instead of just GCN, then it is a waste of time as the performance benefit is probably miniscule and not worth it.

I guess we'll only find out if they release more data or when the BF4 Mantle-powered version arrives, whichever happens first.

I mentioned that earlier and I find myself pretty confused by it. It's either a close to the hardware API or its just another abstraction layer with less overhead. If it is the former than we really won't see the gains we want, and if it is the latter the whole idea of open doesn't make sense.

I'm personally hoping for the latter because it would shake things up a lot more. I'll be pretty disappointed if AMD just introduced a competitor to OpenGL and DirectX instead of something special.

I agree completely, Windows has outlived it's usefulness. It's time to move on to *truly* open platforms and standards, enough with this proprietary BS already. It's fine in the early stages of a technology, but to make an attempt to close down what was previously open...it's just ridiculous.

Next thing you know, Cisco is going to announce a proprietary form of WiFi that's faster than anything else out there, but only works with Cisco gear. Can you imagine what a nightmare that would be?

More like Cisco offering a Cisco-Cisco wireless link that is faster than WiFi but doesn't interfere with WiFi operation with other devices.

 

[Revolution 11]

Well, I didn't expect so many replies to my question but it is interesting to see how different the descriptions of Mantle are on page 28.

 

[EightySix Four]

Well, I didn't expect so many replies to my question but it is interesting to see how different the descriptions of Mantle are on page 28.

They seem to all mention the same features of Mantle and provide different interpretations based on their understanding of how low level APIs work and what they wish would happen.

 

[Creig]

Well we went from rumours of a new card that could be 2X as fast as Titan

Who ever said that? And if somebody did, you actually believed them?

 

[itsmydamnation]

(Ok, since you brought this up, maybe you'll realize your mistake if I try one more time by leading from your own example).

That is EXACTLY what CUDA and Mantle (if Mantle really is a close-to-the-metal API*) are!

In your example, you can't support DX11 Tess up to 64 because maybe your hardware doesn't have Tess units. Your hardware physically has zero Tess units, therefore it can never support DX11 Tess. Correct.

In CUDA and Mantle (again, if it is truly a close-to-the-metal API), the same logic applies. Because CUDA and Mantle will keep making calls to X, Y, or Z hardware units (whatever hardware units they are expecting to be present) and only NV arch (for CUDA) or GCN (for Mantle) will have those hardware units. It is exactly like how your hardware can't support DX11 because of the lack of Tessellation fixed-function hardware.

Do you realize your earlier mistake now, in calling CUDA and Mantle artificially locked down? Or will you prove that you are just trolling here?


Different topic, footnote:
*On whether Mantle truly is a close-to-the-metal API: I am getting mixed signals from AMD, or the press is misreporting. If Mantle is truly a close-to-the-metal API, then it cannot be "open" or cross-ISA/arch, because being locked to the specific arch is inherent to bare metal coding. Even now (decades since they arrived), we don't have assemblers that produce one type of machine code that can be run on any ISA like x86, ARM flavor, PowerPC, etc. It is simply impossible, because that machine code has to have a 1-1 target on the hardware side, and that is defined by the specific arch. But now there is talk and "confirmation" of Mantle being open or some such silliness.

You can't have it both ways. Either you are truly a close-to-the-metal API (therefore locked down to your target hardware arch), or you are still an abstraction layer (therefore can be "open" and used by other competing hardware archs). It can only be one or the other. If the press reports of being "open" are true, then this definitely isn't close-to-the-metal as we understand it, and simply another abstraction layer that supposedly has lower overhead than DX (i.e., far less useful and slower than if it were truly close-to-the-metal as initially thought). If it's just another abstraction layer that is abstracted high enough to accommodate multiple archs instead of just GCN, then it is a waste of time as the performance benefit is probably miniscule and not worth it.

I guess we'll only find out if they release more data or when the BF4 Mantle-powered version arrives, whichever happens first.

i think you are grouping two things together and then calling it one thing.

Mantel can be open, it just doesn't have to be flexible, ie the hardware and OS has to meet Mantels specific requirements. From that perspective you have have an Open yet specific and highly controlled API. Now weather Nvidia could use mantel in that form is a completely different question, let alone would they actually support it.

 

[Skurge]

Competition haven't given us better prices with Nvidia just raising prices across the board and AMD dropping prices, but People still buying nVidia cards anyway. AMD really had no choice but to do this or die. The current situation where the sell faster cards for less and still get worse sales wasn't working. Competition for you huh.

 

[sontin]

In CUDA and Mantle (again, if it is truly a close-to-the-metal API), the same logic applies. Because CUDA and Mantle will keep making calls to X, Y, or Z hardware units (whatever hardware units they are expecting to be present) and only NV arch (for CUDA) or GCN (for Mantle) will have those hardware units. It is exactly like how your hardware can't support DX11 because of the lack of Tessellation fixed-function hardware.

Do you realize your earlier mistake now, in calling CUDA and Mantle artificially locked down? Or will you prove that you are just trolling here?

DX never said anything about the hardware implementation. It's only force the IHV to support certain features.
A low level API is only working in combination with the hardware level and how the API has access to the hardware. The whole translation stack between the driver and DX is much thicker. And nVidia will have no access to the documentation of AMD's hardware. Mantle needs certain ways to design the hardware. And this is impossible for every other vendor out there.

 

[BrightCandle]

I am getting the impression that mantle is an API like that of the consoles. Its not really low level but rather less abstracted than something like DirectX. Until I look at the API I can't really say for sure the what and how but principally right now I am taking this as meaning its a competitor to openGL and DirectX, but where AMD is in charge of its development. DirectX and OpenGL were both born in a world without shaders and I guess now we are moving into a world where AMD wants those to matter a great deal more, not just for graphics. Making an API that better fits the model everyone is developing in could very well make sense, but its success is going to depend on NVidia choosing to write a driver for it and being able to support it as well as it actually proving itself to be worth having. If BF4 doesn't show significant gains from it on PC it will be DOA, they really need to show that DirectX is really bad and its DX not the games special optimisation that is at fault. If the gains are marginal it wont be worth it, we already have a cross platform standard in openGL.

 

[EightySix Four]

DX never said anything about the hardware implementation. It's only force the IHV to support certain features.

Right, because the drivers do the work going from the DirectX call to the hardware implementation. You can't do that when you're talking directly to the hardware otherwise you're just building a different abstraction layer.

A low level API is only working in combination with the hardware level and how the API has access to the hardware. The whole translation stack between the driver and DX is much thicker. And nVidia will have no access to the documentation of AMD's hardware. Mantle needs certain ways to design the hardware. And this is impossible for every other vendor out there.

Now you seem to be making our point? This isn't an artificial limitation because there are physical hardware differences that can't be overcome to make this work. Mantle targets a specific hardware architecture directly... Do you want AMD to license GCN to nVidia or something?

I am getting the impression that mantle is an API like that of the consoles. Its not really low level but rather less abstracted than something like DirectX. Until I look at the API I can't really say for sure the what and how but principally right now I am taking this as meaning its a competitor to openGL and DirectX, but where AMD is in charge of its development. DirectX and OpenGL were both born in a world without shaders and I guess now we are moving into a world where AMD wants those to matter a great deal more, not just for graphics. Making an API that better fits the model everyone is developing in could very well make sense, but its success is going to depend on NVidia choosing to write a driver for it and being able to support it as well as it actually proving itself to be worth having. If BF4 doesn't show significant gains from it on PC it will be DOA, they really need to show that DirectX is really bad and its DX not the games special optimisation that is at fault. If the gains are marginal it wont be worth it, we already have a cross platform standard in openGL.

Here's the thing, any API like that that is not closer to the hardware is going to have the overhead they specifically said they were getting rid of. When that is combined with the "open" part of it, it's kind of hard to reconcile exactly what it is. If it's an abstraction layer, could it really be that much faster than DirectX and OpenGL?

 

[Xarick]

Do you all really believe Nvidia doesn't already have their own API that speaks directly to hardware? I find it hard to believe they don't have an in house one already.
Also.. how many studios are going spend the extra money it will cost to implement any other api in addition to DirectX.. When budgets are tight you develop the the most universal standard you can.

 

[brandonb]

Do you all really believe Nvidia doesn't already have their own API that speaks directly to hardware? I find it hard to believe they don't have an in house one already.
Also.. how many studios are going spend the extra money it will cost to implement any other api in addition to DirectX.. When budgets are tight you develop the the most universal standard you can.

You hit the nail on the head but the nail is different than what you are thinking.

It will be cheaper to use Mantle to develop for all consoles + Windows + Linux, than to use DirectX on windows, OpenGL on Linux, and all the console proprietary APIs. The cost will be loss of support to Intel + Nvidia GPUS (if they don't get access to the API, and they probably will).

This isn't an attack on Nvidia, it's an attack on Microsoft.

 

[AnandThenMan]

Do you all really believe Nvidia doesn't already have their own API that speaks directly to hardware? I find it hard to believe they don't have an in house one already.

If Nvidia has such a thing, they're not talking.

Also.. how many studios are going spend the extra money it will cost to implement any other api in addition to DirectX.. When budgets are tight you develop the the most universal standard you can.

Several so far and I'm sure many more to come, as shown at the tech event. Although I think the costs will actually be lower. If Mantle was a Windows only proposition, yea it would probably die a quick death. But given that AMD has hardware in all the consoles, changes everything.

Also people are ignoring the fact that Mantle is also going to take advantage of AMD 8-core processors.

 

[Abwx]

This slide say that BF4 is optimised for DX11.1.

Microsoft added optimisations on our request to DX11.1

http://www.hardware.fr/medias/photos_news/00/42/IMG0042728_1.jpg

http://www.hardware.fr/news/13368/amd-mantle-retour-glide.html

 

[Skurge]

Do you all really believe Nvidia doesn't already have their own API that speaks directly to hardware? I find it hard to believe they don't have an in house one already.
Also.. how many studios are going spend the extra money it will cost to implement any other api in addition to DirectX.. When budgets are tight you develop the the most universal standard you can.

Not unless someone is paying you to develop for their own API. AMD have frostbite which will power almost all of EA's games, Square Enix and Eidos are firmly in AMD's pocket. It's just a matter of weather AMD can convince them to develop for Mantle or not.

This could be a blessing as well. If AMD could make it really easy to port games to PC if Mantle is used we could see games like Final Fantasy XV and MGS come to PC.

 

[Revolution 11]

If all Mantle does is scare Microsoft into advancing DirectX in a meaningful way, job well done, I say. All this talk about Linux gaming, SteamOS, and now Mantle could be a reaction from the rest of the industry towards Microsoft's lackluster attitude towards DirectX.

Stop fragmenting the market, Microsoft.

 

[EightySix Four]

This slide say that BF4 is optimised for DX11.1.



http://www.hardware.fr/medias/photos_news/00/42/IMG0042728_1.jpg

http://www.hardware.fr/news/13368/amd-mantle-retour-glide.html

Being optimized for DirectX and supporting/optimizing for Mantle are not exclusive from one another.

 

[Freaksterz]

Do you all really believe Nvidia doesn't already have their own API that speaks directly to hardware? I find it hard to believe they don't have an in house one already.
Also.. how many studios are going spend the extra money it will cost to implement any other api in addition to DirectX.. When budgets are tight you develop the the most universal standard you can.

For starters, ALL Games running on frostbite 3.

 

[jpeyton]

Also.. how many studios are going spend the extra money it will cost to implement any other api in addition to DirectX.. When budgets are tight you develop the the most universal standard you can.

I'm assuming that because many major studios are already developing for consoles, and both next gen consoles use GCN with low-level API access, that developing for GCN on PC using low-level API access isn't a stretch of resources.

 

[Xarick]

Anyone thinking that games being easily ported to the PC is a good thing is fooling themselves. Games being redone for the pc to take advantage of the pc and it's control scheme is far more desirable than getting a bunch of ports that are easy.

 

[GaiaHunter]

Anyone thinking that games being easily ported to the PC is a good thing is fooling themselves. Games being redone for the pc to take advantage of the pc and it's control scheme is far more desirable than getting a bunch of ports that are easy.

But with mantle that will be easier as well.

 

[Edgy]

I see what could have been with Mantle if done differently than it is...

IF Nvidia somehow get this to work on their hardware as well as AMD and IF game devs start using it because it makes it so much easier to port from PC (Windows/Linux/Apple) to Console and vice versa...

Well - wouldn't that remove one of the biggest advantage Windows have over other OS like Linux and Apple - GAMES.

I wonder what the backstory is with Steam OS and Mantle - I would guess that Mantle was removed from consideration due to Nvidia/Intel not being involved in it from the beginning but if something like Mantle was agreed upon from the start between AMD/Nvidia/Intel, we might have had a future where avalanche of games being released to Linux/Apple ecosystem... the beginning of the end for Windows monopoly maybe?

 

[AnandThenMan]

Agreed, but Microsoft moves at the pace of a sloth. So we could be in for a long wait...

If all Mantle does is scare Microsoft into advancing DirectX in a meaningful way, job well done, I say. All this talk about Linux gaming, SteamOS, and now Mantle could be a reaction from the rest of the industry towards Microsoft's lackluster attitude towards DirectX.

Stop fragmenting the market, Microsoft.

 

[Xarick]

I am pretty interested in seeing the performance advantage mantle can bring. Hopefully it will play well with Intel CPUs as I don't see myself jumping to AMD cpu's any time soon.

 

[ozzy702]

Anyone thinking that games being easily ported to the PC is a good thing is fooling themselves. Games being redone for the pc to take advantage of the pc and it's control scheme is far more desirable than getting a bunch of ports that are easy.

??? Not sure you know what you're talking about. An easier port means MORE time to focus on pc control scheme and getting the pc version correct. If anything devs can now build for pc and scale down for the respective consoles.

 

[AnandThenMan]

Would Nvidia really consider writing a Mantle aware driver? I dunno, that's putting control too much in the hands of your direct competitor. What's stopping AMD from updating Mantle so that Nvidia hardware is always 2 steps behind. Same as if AMD were to obtain a PhysX license, risky.