AMD Mantle API [updating]

[ams]

Targeting Mantle is not that much of an overhead because it:

1. Can reuse console code 'in whole or part (AMD)"

XboxOne supports DirectX at a high level and PS4 supports OpenGL at a high level, which facilitates porting this code to PC's, so I don't see the purpose of trying to port "part" of lower level code. And at a lower level, there are likely to be some differences in the console API's due to console-specific features added.

2. Already has a large GPU user base that can benefit from Mantle - the 7xxx cards.

That is incorrect. The percentage of PC gaming systems with GCN-equipped graphics is very very small compared to everything else out there. The majority of PC gaming systems with discrete graphics cards use NVIDIA graphics (while the rest use primarily non-GCN graphics).

3. Addresses both the GPU and CPU

Wrong again. The vast majority of PC systems with integrated graphics use Intel CPU's + Intel iGPU's. The vast majority of PC systems with discrete graphics use Intel CPU's + NVIDIA GPU's. The vast majority of Android/iOS/WinRT systems use Qualcomm/ImgTech/ARM/etc. CPU's + GPU's. Even many popular consoles today (and possibly in the future too) use non-AMD CPU's and non-GCN graphics such as Xbox360, PS3, 3dS, etc

4. GCN is a long term architecture. Mantle optimizations done today will work on AMD processors far into the future.

GCN is a generic name for AMD's current GPU architecture. GPU architectures evolve and change over time. Even according to AMD, there is no guarantee of backward compatibility with future lower level API's and current GCN hardware.

5. Mantle is designed from the start to be multi platform - Linux and Android can be easily accommodated.

That is not very relevant because Android has zero GCN-equipped hardware, iOS has zero GCN-equipped hardware, WinRT has zero GCN-equipped hardware, and even Linux support from AMD has been dubious at best.

With most of the major developers (and engines) on board and cross platform future proofing Mantle is looking very strong.

Actually the long-term prospects for a very low level IHV-specific graphics API such as Mantle are very dim when you consider the need for game developers to increasingly target more and more [ultra mobile] platforms and more and more [ultra mobile] graphics vendors.

 

[3dilettante]

I would have said the second definition was the common one...

Gimmick has a more negative connotation than the plain text of the second definition, which doesn't comment on whether a thing has an impact beyond the generation of interest.
It implies heavily that there's no substance, which I don't think is accurate. It may not work out, but there is meat to it.

The amount of investment in these efforts and the pervasiveness of the changes are already signs that there something more serious than that going on.
In addition, restricting things to a marketing ploy misses the soft-sell of making porting on similar hardware more compelling, and the implication that if AMD can get enough buy-in amongst competing engines that holdouts will find themselves hurt.
This is a market play that does attract eyeballs, but it also hits at a technical and potentially financial level as well.


All this aside, I'm curious if Mantle includes the user-level queuing capabilities mentioned for HSA and its compute.
That would seem to be useful here.

Another question I have is whether the idea that if the idea is that an application can use Mantle to talk to the GPU, what if said application were the D3D runtime or the user driver?
The low-level access wouldn't be there, but if coupled with a divorcing of functionality from the driver model, it might allow for the newest features to be applied even if the system's DX version doesn't support them.
Possibly, that could allow Mantle functions to be used in parts of a standard DX game (requires more work on the part of the system software), or a game's DX11.n path running unchanged on a virtual DX path even when the user system has a lower DX version (requires faking out either the application or Windows).

Could AMD try making a user-mode driver that takes input from the D3D runtime, spits out work on a hidden work queue, and passes back bookkeeping command entries to keep Windows happy?

 

[Malo]

Correct. It were Batman fans which have been "screwed over" by Rocksteady, who have been "screwed over" by their middle-ware provider, for not providing native & generic MSAA support.

If you think there was some Nvidia conspiracy going on, you might ask yourself who is really the sinner in this case, Nvidia or Rocksteady. Who defines and implements the engine's features?

What if MSAA would be a Nvidia only feature?

I don't believe the Batman MSAA situation is comparable to Mantle. The code used for MSAA was generic code straight from SDK's that would have worked fine with AMD's (and indeed did so for the GoTY edition which wasn't covered under the same SKU legal obligations to Nvidia).

With Mantle obviously it's not going to work on Nvidia.

 

[jimbo75]

That is incorrect. The percentage of PC gaming systems with GCN-equipped graphics is very very small compared to everything else out there. The majority of PC gaming systems with discrete graphics cards use NVIDIA graphics (while the rest use primarily non-GCN graphics).

No it's quite correct. AMD will have sold (at a rough guess) ~50 million GCN cards, and while that is a small part of the entire PC user base, it's still a massive number that will probably multiply ten-fold over the coming years.

Actually the long-term prospects for a very low level IHV-specific graphics API such as Mantle are very dim when you consider the need for game developers to increasingly target more and more [ultra mobile] platforms and more and more [ultra mobile] graphics vendors.

Great, there's no more need for you to go on an on about it being bad for consumers, developers and everybody else.

 

[ams]

LOL @ jimbo75: While we are pulling numbers out of thin air, imagine how many NVIDIA/Intel/Qualcomm/ARM/ImgTech/etc. GPU's have been sold over the last two years, and how many will be sold over the coming years.

 

[willardjuice]

There's a lot to respond to (and only so much compile time!!!), but...

XboxOne supports DirectX at a high level and PS4 supports OpenGL at a high level, which facilitates porting this code to PC's

Is 100% wrong. No one uses ogl on ps3/ps4 and I highly doubt developers are using a "high level" api on X1.

 

[Albuquerque]

LOL @ jimbo75: While we are pulling numbers out of thin air, imagine how many NVIDIA/Intel/Qualcomm/ARM/ImgTech/etc. GPU's have been sold over the last two years, and how many will be sold over the coming years.

And why does this matter? An API doesn't have to apply to everyone to be useful.

Mantle is meant for high-performance video games, and there are ONLY two companies catering to high performance video gaming needs. You already know the two companies, and you already know one of them is who rolled out Mantle.

Thus, the target audience of Mantle has zero to do with all the other shit you continue to spew about an ecosystem that you arbitrarily assigned. As such, your claims about pain and suffering are null and void.

You've stated it yourself in no uncertain terms: Performance increases for one IHV is fine, even if the other IHV(s) don't get it. I verbatim quoted you saying it. But your original context was NVIDIA getting the bump, which explains why you hate this.

 

[itsmydamnation]

LOL @ jimbo75: While we are pulling numbers out of thin air, imagine how many NVIDIA/Intel/Qualcomm/ARM/ImgTech/etc. GPU's have been sold over the last two years, and how many will be sold over the coming years.

and imagine how many of those apply to mantles target market.

 

[Brad Grenz]

If BF4 doesn't run well on my old, hacked HP TouchPad I know I will be livid and blame Mantle.

 

[Blazkowicz]

And imagine how many not-GCN GPUs AMD has been selling.
E-350, 5450, 6450, 6670, Trinity, Richland and the Radeon 6870 was not so long ago.

Yes people do buy Radeon 6670 to play games, it's what it's made for .

 

[pjbliverpool]

LOL @ jimbo75: While we are pulling numbers out of thin air, imagine how many Qualcomm/ARM/ImgTech/etc. GPU's have been sold over the last two years, and how many will be sold over the coming years.

Please stop banging on about those IHV's. They have precisely nothing to do with the high end PC gaming space that Mantle is targeted at. The games that PC gamers are playing on GCN and equivalent NV architectures are never going to go anywhere near GPU's from those IHV's so they don't contribute to a fracturing of the market for those games.

Sure developers these days may be developing mobile games with the same title as the desktop equivalent but its not the same game at all. Mantle or not developers would have to produce a completely separate code base for the mobile versions of games so if anything that makes the impact of developing to one extra API for your high end version of the game only a marginal increase in workload.

 

[Blazkowicz]

It's funny how we're back to "low end graphics card = old gen", i.e. Radeon 7000 was the low end when there was Radeon 8500, Radeon 9200 was the low end when there was Radeon X800 pro, same thing with nvidia.

Then both vendors switched to offering current gen GPU on the low end, first with Radeon X300 and geforce 6200, Radeon X1300 and geforce 7300, HD2400 and 8400GS etc.
Now the 5000/6000 series is starting to look old. In the old times AMD would have been selling a 128 SP 64bit GCN for low end cards by now. Sure, they don't want to and it's understandable. What messes up the situation is Kaveri is still missing.

 

[jimbo75]

LOL @ jimbo75: While we are pulling numbers out of thin air, imagine how many NVIDIA/Intel/Qualcomm/ARM/ImgTech/etc. GPU's have been sold over the last two years, and how many will be sold over the coming years.

You're (deliberately) missing the point. The console devs realistically have a choice of AMD or Nvidia to cater for on PC. Intel is still mostly irrelevant.

The overall graphics market share is -

Right now a lot of that AMD share (probably around 1/3rd) isn't GCN because Trinity isn't GCN. By this time next year however - everything else being equal - all of that 21.9% for AMD will be GCN based PC.

The console devs are looking at numbers of GCN-equipped devices and seeing AMD's entire lineup of PC chips and both consoles. By this time next year that market will be some 25-30 million AMD (Kaveri) APU's, 100 million discrete (7-series and R-series) graphics and say 30 million consoles on GCN. These numbers are far, far higher than anything Nvidia will be able to muster.

Now here's the kicker - a lot of Nvidia's market is crap level anyway (GT 610, GT 620 etc) that can't game worth a shit. That same crap level for AMD could well be catapulted into "gameable" level due to Mantle. More devices - cheaper devices - become good enough to play certain titles. Then Mantle starts really becoming true "cross platform". AMD will be in ARM based tablets and phones in less than 2 years, and guess what? Yup, It'll be GCN graphics, boosted with Mantle. This could revolutionise mobile gaming.

AMD will be targeting ~1 billion GCN based devices within 5 years. Even if they fail to get near than number it's still going to be an absolutely huge amount of devices simply through consoles and what they have left in PC's. There will be at least a half billion GCN devices within 5 years.

TLDR - forget any ideas of the devs not bothering with Mantle because of insignificant user base. This is AMD's big play for graphics dominance and they hold at least 2 Aces right now.

 

[ams]

No one uses ogl on ps3/ps4 and I highly doubt developers are using a "high level" api on X1.

That's not what I meant. Obviously game developers will use a very specific low level API for PS4 and XboxOne, respectively. But to port games to a PC platform (which, by definition, is an open platform where a variety of different hardware needs to be supported), developers will use OpenGL and DirectX, respectively.

 

[ams]

You're (deliberately) missing the point. The console devs realistically have a choice of AMD or Nvidia to cater for on PC. Intel is still mostly irrelevant.

No I am not missing the point. The argument that Mantle should be adopted because of user base "strength in numbers" makes no sense, because GCN doesn't have "strength in numbers" at all. Both today and in the future, GCN graphics will represent a very small fraction of all graphics systems used and sold. Due to the exponential growth in graphics performance in the ultra mobile space, and due to the growth in ultra mobile systems sold, game developers will need to target a variety of different platforms, including ultra mobile platforms where there is great diversity in graphics hardware. AMD has literally zero presence in this space right now (ie. zero % marketshare in Android/iOS/WinRT). And in the near future, console game developers will still need to target Xbox360 and PS3 as the lowest common denominator, where GCN graphics have zero presence (same goes for all the handheld consoles). Finally, in the Windows PC space, with respect to discrete graphics, the overwhelming majority of PC gamers (> 65%) prefer and use Intel CPU's + NVIDIA GPU's, and with respect to integrated graphics, the overwhelming majority of PC users use Intel CPU's + Intel iGPU's.

 

[Blazkowicz]

Sure developers these days may be developing mobile games with the same title as the desktop equivalent but its not the same game at all. Mantle or not developers would have to produce a completely separate code base for the mobile versions of games so if anything that makes the impact of developing to one extra API for your high end version of the game only a marginal increase in workload.

Nvidia seems to be moving in that direction of convergence, they'll sell mobile GPUs (Tegra 5 and up) with the full desktop feature set and desktop APIs (OpenGL, CUDA).
So you could target Android, Linux (i.e. Steam and SteamOS) and Windows at once with mostly the same code and with the same graphics API.

That has nothing to do with Mantle, and it's maybe not natural or sensible to do AAA games that way. You would have indie and casual games that way, at first.
It's an anti-Mantle I'm talking about actually, if there's an OpenGL rebirth, the OpenGL games will run on AMD stuff too - including Temash, Kabini which are just too weak for the big Direct3D / Mantle games.

 

[jimbo75]

No I am not missing the point. The argument that Mantle should be adopted because of "strength in numbers" is completely false. Both today and in the future, GCN graphics will represent a very small fraction of all graphics systems used and sold. Due to the exponential growth in graphics performance in the ultra mobile space, and due to the growth in ultra mobile systems sold, game developers will need to target a variety of different graphics hardware vendors and operating systems in the near future.

What exponential growth in ultramobile? We've been there, done that - http://www.taipeitimes.com/News/biz/archives/2013/10/12/2003574285

“Some of TSMC’s major customers are seeing demand for high-end smartphones slowing down and carrying high inventory to boot,”
Now it's possible they are just talking about Nvidia but I doubt that would be worthy of a paragraph so it's more likely to be a proper slowdown affecting the industry. Exponential growth is long gone.

And in the PC space, with respect to discrete graphics, the overwhelming majority of PC gamers (> 65%) prefer and use Intel CPU's + NVIDIA GPU's.

How many of them are using GT 610's and 620's? AMD's worst Trinity is at least comparable to those.

 

[jimbo75]

delete

 

[jimbo75]

Finally, in the Windows PC space, with respect to discrete graphics, the overwhelming majority of PC gamers (> 65%) prefer and use Intel CPU's + NVIDIA GPU's, and with respect to integrated graphics, the overwhelming majority of PC users use Intel CPU's + Intel iGPU's.

Intel CPU's with Intel iGPU's are a total irrelevance. They aren't good enough to be considered part of the gamer market. Intel doesn't have 60%+ of the graphics market because of graphics, it's because their graphics are tied to their CPU's.

The majority of Nvidia's discrete GPU's sales aren't good enough for true gaming either - the type we're discussing here. You're making the classic mistake of assuming "discrete" is the same as "enthusiast".

 

[Blazkowicz]

Fucking wrong, people are busy playing League of Legends, Valve games and such on their laptop with Intel graphics.
Millions upon millions gamers.
That's maybe afterall one of the reason for AMD's APU strategy, Intel sells tons of its own APUs and people use them because it's what's available on the cheap laptops.