AMD Mantle API [updating]

[WaltC]

According to Peddie, nVidia is taking a beating in total gpu market share.

http://jonpeddie.com/press-releases/details/amd-winner-in-q2-intel-up-nvidia-down/

The data as of 8/13/2013.

 

[Andrew Lauritzen]

... big post on mobile stuff ...

I think there was some mis-communication on my point there. I'll avoid replying here due to us straying increasingly off-topic, but if people are interested in continuing the discussion I'd be happy to reply in another split thread.

 

[grokzilla]

Look, this isn't rocket science. For the nth time, what I said is that for pre-built systems that use Intel CPU's AND discrete graphics, NVIDIA GPU's are used the majority of the time (>90%)..

Misses the point. What counts is what happens into the FUTURE and that's clear enough:

1. Low and midrange desktops and laptops/netbooks/etc. will be purely Intel/AMD APUs with AMD APUs growing market share with gaming oriented buyers.

2. AMD will increasingly own high end/gaming desktops and laptops as more and more Mantle optimized games come on line. It simply won't make sense to buy Nvidia when AMD will provide cost /performance parity on all PC games and a far better gaming experience on Mantle optimized games.

 

[Andrew Lauritzen]

2. AMD will increasingly own high end/gaming desktops and laptops as more and more Mantle optimized games come on line. It simply won't make sense to buy Nvidia when AMD will provide cost /performance parity on all PC games and a far better gaming experience on Mantle optimized games.

Nothing in that statement is "clear"... certainly that's what they want you to believe, but there's a multitude of other likely scenarios.

Honestly the message that no one wants to see a future where we go back to vendor specific APIs has been pretty clear from the game developers. Thus ultimately the point of Mantle will likely be to effect positive change to other APIs and platforms (which is great), not so much to be an end unto itself.

 

[grokzilla]

Nothing in that statement is "clear"... certainly that's what they want you to believe, but there's a multitude of other likely scenarios.

Name three of those 'multitude of other likely scenarios'?

The fact is that ultimately if Mantle really does make much difference that the majority of game developers want, it'll just force change (i.e. to DX/GL, etc). You really think NVIDIA or anyone else is just going to lie down and accept AMD's amazing victory, even if the rest of the speculation in your statement does magically hold up? Silly notion.

What can Nvidia do about it?

They for sure have zero influence at EA/Dice where Mantle is being gleefully embraced and 16 games SO FAR have been verified as Mantle games, Ea is so enthusiastic about Mantle it's not a stretch that Mantle will next be integrated into Ignite and all EA sports games will be Mantle optimized.

AMD has stated they have 'several' more Mantle partners to be revealed at APU13. That would be several more game engines and major developers. That's a whole lot of Mantle games in development.

Honestly the message that no one wants to see a future where we go back to vendor specific APIs has been pretty clear from the game developers. Thus ultimately the point of Mantle will be to effect positive change (which is great), not so much to be an end unto itself.

It's well known AAA game developers have been frustrated with DX for years and have been wanting, and asking for, a way to program closer to the metal on PC games. Johan Andersson said exactly that during a Q&A session at the Nvidia event. With Mantle AMD, leveraging their console wins, is now able to effective supply what several developers have been asking for.

 

[Andrew Lauritzen]

Name three of those 'multitude of other likely scenarios'?

Why three (beyond the typical goal-post moving)? I already gave you the one that I think is the most likely: DX/GL improve their overhead/multithreading to a similar level. As a free-bee, you can have game developers move more of the submission logic to the application itself via bind-less and indirect drawing extensions and thereby avoid the need for tons of API calls.

What can Nvidia do about it?

Make their own API? Johan did say he has been asking all three of the main folks for this for years and consequently would likely support others as well. Or drive improvements to the two that we already have? As I've said earlier in this thread, either Mantle is very specific to GCN or - at best - it'll have a short period of lead over the standard APIs in performance. But the swiftness of the response from Microsoft and Khronos and others will obviously be relative to the actual gains there. You really think everyone is just going to capitulate if there's important gains to be had? I'm not sure how you think the industry works...

If it's at-all portable then by definition there nothing fundamental to the software that AMD is writing that can't be reproduced. If it's not portable, then there are other problems. Which camp are you in here?

AMD has stated they have 'several' more Mantle partners to be revealed at APU13. That would be several more game engines and major developers. That's a whole lot of Mantle games in development.

No argument on the number of games, but that doesn't support the rest of your argument. Furthermore since Johan helped push and design the API and he's the lead for the engine that these Mantle games will be using, the logic that a large number of other developers want this doesn't necessarily follow. In fact John Carmack and Tim Sweeney both made it pretty clear that they didn't like it as an industry direction recently so I think it's far from clear that everyone is going to hop on-board. You're making a lot of assumptions about what Mantle brings to the table that are far from known at this point.

Anyways, upon re-reading, your post and post history are sounding pretty fanboyish, so given that you're a new poster I might just exit from this particular sub-thread before getting too deep in that rat-hole.

 

[homerdog]

Anyways, upon re-reading your post and post history are sounding pretty fanboyish, so given that you're a new poster I might just exit from this particular sub-thread before getting too deep in that rat-hole.

I'm surprised you banged your head against that wall as long as you did

 

[gkar1]

Heh, developers have egos and can also be fanboys too.

Out of the 3 from that panel Johan sounded the most pragmatic. He got past their ego barrier pretty quickly and got them to talk about what they wished to do with hardware (Carmack specifically talking about wanting to get his hands on GCNs async compute engines, etc) and later Sweeney just to not be left behind trying to get a word in edgewise.

 

[pjbliverpool]

Is it just me or does the PC forum feel a lot like the console forum recently?

 

[grokzilla]

Why three (beyond the typical goal-post moving)? I already gave you the one that I think is the most likely: DX/GL improve their overhead/multithreading to a similar level. As a free-bee, you can have game developers move more of the submission logic to the application itself via bind-less and indirect drawing extensions and thereby avoid the need for tons of API calls.

You said 'multitude'. Asking for three examples is more than reasonable and hardly 'moving the goalposts'.

DX/GL are of necessity sufficiently abstracted to cover all GPU architectures. They can be considerably improved and expanded, but they cannot improve their overhead/multithreading to a similar level as Mantle which targets a single architecture.

Make their own API? Johan did say he has been asking all three of the main folks for this for years and consequently would likely support others as well. Or drive improvements to the two that we already have? As I've said earlier in this thread, either Mantle is very specific to GCN or - at best - it'll have a short period of lead over the standard APIs in performance. But the swiftness of the response from Microsoft and Khronos and others will obviously be relative to the actual gains there. You really think everyone is just going to capitulate if there's important gains to be had? I'm not sure how you think the industry works...

Mantle is workable solely because it can leverage large chunks, or in some cases, apparently, 'all', of the coding done for the consoles and once the initial work is done to incorporate a Mantle path into an engine, BF4 in the case of Frostbite, the engine can then do the heavy lifting of creating a mantle path for all subsequent games - as per Johan's 'out of the box' phrase. And common sense.

If it's at-all portable then by definition there nothing fundamental to the software that AMD is writing that can't be reproduced. If it's not portable, then there are other problems. Which camp are you in here?

AMD said it was designed from the start to be cross platform and that they intend to open source it when it is 'sufficiently developed' - a rather nebulous timetable - which would apply not only to NVidia, but also the Android graphics players. Will have to see how that all falls out.

No argument on the number of games, but that doesn't support the rest of your argument. Furthermore since Johan helped push and design the API and he's the lead for the engine that these Mantle games will be using, the logic that a large number of other developers want this doesn't necessarily follow. In fact John Carmack and Tim Sweeney both made it pretty clear that they didn't like it as an industry direction recently so I think it's far from clear that everyone is going to hop on-board. You're making a lot of assumptions about what Mantle brings to the table that are far from known at this point.

That Nvidia was only able to round up ONE major developer for their three man Montreal panel that was definitely against Mantle, one that was on the fence 'pending' developments, and one that will be a keynote speaker at APU13 is telling. What it's telling me is the rest of the major players are either already one of the 'several partners' to be revealed at APU13 or are on the fence waiting to see what happens.

Anyways, upon re-reading, your post and post history are sounding pretty fanboyish, so given that you're a new poster I might just exit from this particular sub-thread before getting too deep in that rat-hole.

As you will.

 

[grokzilla]

Heh, developers have egos and can also be fanboys too.

Out of the 3 from that panel Johan sounded the most pragmatic. He got past their ego barrier pretty quickly and got them to talk about what they wished to do with hardware (Carmack specifically talking about wanting to get his hands on GCNs async compute engines, etc) and later Sweeney just to not be left behind trying to get a word in edgewise.

I'm interested to see how long it takes Sweeney to change his tune if APU13 has the rest of the major engines and developers getting on the Mantle bandwagon, BF4 Mantle proves to be a stunner, AMD GPUs suddenly becomes the red hot 'must have' hardware and developers using the Unreal engine for their next gen titles look on in dismay and consternation.

 

[3dilettante]

Is it just me or does the PC forum feel a lot like the console forum recently?

Easier porting of console code = easier porting of console forum.

DX/GL are of necessity sufficiently abstracted to cover all GPU architectures. They can be considerably improved and expanded, but they cannot improve their overhead/multithreading to a similar level as Mantle which targets a single architecture.

I don't know if this is true. The low-hanging fruit for overhead and mulithreading has more to do with queueing, driver model, and state management. This is more of an overarching system architecture issue, where most things incur penalties before touching the GPU.

This is something orthogonal to feature exposure and manipulation of architectural elements within the space of the GPU's domain.
The things being changed or worked around with bindless extensions (that AMD is also pledging to implement) apply to multiple architectures.

It may be that AMD has implemented specific optimizations, protections, or processing paths with Mantle. However why would the particulars of one architecture be the only possible way, and how does that preclude a standard bypass?

I'm interested to see how long it takes Sweeney to change his tune if APU13 has the rest of the major engines and developers getting on the Mantle bandwagon, BF4 Mantle proves to be a stunner, AMD GPUs suddenly becomes the red hot 'must have' hardware and developers using the Unreal engine for their next gen titles look on in dismay and consternation.

This depends on many other things besides the technical merits or drawbacks to Mantle.
There are other things in addition to the already long list of ifs you put in there.

 

[homerdog]

grokzilla, since most in this forum are already fairly technically proficient, feel free to take your expertise of low-level graphics APIs somewhere it would be more appreciated. Running off Andy from the thread is not appreciated.

 

[willardjuice]

Agreed, too much cheerleading on both sides. I'm amazed how there's such hard feelings about an api that we know essentially nothing about.

 

[willardjuice]

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?

...

It's of course Rocksteady's fault. I don't blame nvidia for trying to make money. It's called capitalism.

But are you suggesting that there wasn't a downfall to this decision? I think it's fair to say amd customers felt cheated. I'm not even passing judgement on whether this was a worthy trade-off, but it was clear that this choice made many customers unhappy. I never said proprietary apis are inherently bad, but they do have cons. That's not to say the pros never outweigh those cons, but they do exist.

Mantle is just another trade-off. It too has pros and cons. In this case, it seems repi and company feel the pros dwarf the cons.

No, they can't. If a consumer has no contract with a game producer, then he can't. He can't dictate CryTek to make RPGs, he can't dictate Crystal Dynamics to make a side-scroller, he can't dictate DICE to make iPhone games. He can't dictate them to make the games accessible to the blind, which would actually pretty cool, I would love to hear high quality aural games.

Particular developers don't even go to the market and ask people what kind of game they would like and then do that. They just make the game they are either familiar with making, or they want to make for "personal" reasons.
In fact, lots of games come from individuals thinking about it for some time, and manage to find a sponsor to produce it, when the sponsor thinks it's marketable.
Monster companies like EA might look to the market and think it's time for another RPG, and then put one of their studio to do it. Which in effect might end up boring because of it.

So having a middleman changes everything? Developers produce games for publishers, publishers then sell their products to paying customers. If customers don't want X, publishers won't fund developers to make X. There's a reason why studios get shut down after they release bombs...

 

[Dominik D]

DX/GL are of necessity sufficiently abstracted to cover all GPU architectures. They can be considerably improved and expanded, but they cannot improve their overhead/multithreading to a similar level as Mantle which targets a single architecture.

Overhead can be improved in the same orders as Mantle does it. Can it reach the level of "to the metal" programming? No. But it could get darn close. This has nothing to do with with application APIs (DX/OGL) and everything to do with driver model. It so happens that Andrew knows that first hand, you know? So do several other people on this forum.

WDDM is designed around the concurrency and multitasking which introduces necessary levels of abstraction to alleviate drivers from having to deal with dependencies and what not (DXGK's scheduler does that). This comes at a cost but there's nothing stopping WDDM from fast-tracking execution for the single-consumer case. There are special paths for accelerated composition (DXGKDdiRenderKM and DxgkDdiPresent) it just so happens that extra performance on the regular rendering path wasn't priority recently (power consumption seems to be the hot thing right now).

The bottom line is: you have no idea about topics you're trying to discuss. My signature is dedicated to people like you.

 

[Silent_Buddha]

It's of course Rocksteady's fault. I don't blame nvidia for trying to make money. It's called capitalism.

But are you suggesting that there wasn't a downfall to this decision? I think it's fair to say amd customers felt cheated. I'm not even passing judgement on whether this was a worthy trade-off, but it was clear that this choice made many customers unhappy. I never said proprietary apis are inherently bad, but they do have cons. That's not to say the pros never outweigh those cons, but they do exist.

Yup, ever since Batman: AA, I will never buy a Rocksteady game. Ever. I have Batman: AA and Batman: AC right now only because a friend gifted them to me on Steam because he thought it was a shame that I never played them. Batman: AA was certainly a decent game, but doesn't change my mind. No matter how good a game they make, they'll never get a single penny from me due to agreeing to Nvidia's legal terms with regards to the AA used.

It would have been a different matter if it was technically impossible on AMD hardware, but the fact that it was a common AA method that works just fine on AMD hardware as long as you fake an Nvidia device ID just makes me want to puke.

Not that I hold Nvidia completely blameless, either. I extended my personal ban on considering their hardware for purchase a bit longer due to that. But they are right in the mix with AMD cards for my next video card upgrade now (and for the past year+).

With Mantle, I hope it's a similar situation to when they implemented the hair in Tomb Raider. Where all game/graphics features are still available on all hardware of a given DX level, just that it's faster on some hardware than other hardware. Rather than something like hardware PhysX or CUDA where the features are limited to certain cards. IE - Mantle obviously won't work for non-supported cards, but as long as all graphical features are still there, it's not a big deal if it's slower on other cards.

Regards,
SB

 

[Davros]

Davros salutes Silent_Buddha
May the Gaming Gods smile upon you....

ps:

Well, you gotta love AMD and its engineers. Mantle looks – in paper – amazing, however that’s not the only thing the red team is currently working on. According to Graham Sellers, OpenGL guy at AMD, the red team will be supporting this open API with some high performance extensions that will offer almost similar performance to AMD’s upcoming API, Mantle.

As Sellers claimed, AMD aims to expose all of the hardware of their GPUs with these upcoming high performance extensions of OpenGL, and gamers will be able to get close to theoretical peak and performance. Not only that, but Sellers claimed that games using the modern versions of OpenGL won’t be bottlenecked by the API anymore, meaning that gamers will hit HW limits first.

 

[homerdog]

Yes, AMD has plans to match NVIDIA's bindless texture extension in OpenGL. That alone should bring most of Mantle's efficiency advantages to OGL.

The question I would like to ask is, why didn't AMD just do that in the first place instead of going the Mantle route? I'm sure there is a good reason (HLSL vs GLSL?), but as a layman I lack the knowledge and insight to come up with a good answer.

The good news is, if Mantle ends up bringing noticeable benefits to BF4, I will swap out my GTX670 for the HD7950 I just bought and donated to my brother

 

[3dilettante]

It seems to ride the coattails of an API and OS that AMD has the strongest competency in; it hits a broad slice of the gamer market it has incumbency in, and its choice of making it a publisher's pet project allows it to strike more quickly and without giving any competing graphics provider immediate admittance.

I suppose that AMD could have introduced OpenGL extensions and then waited for them to be used, but by whom, how many, and what year I don't really know.
There's a gloom thread that should give a picture of how much time AMD has to leverage its GCN IP and what amount of compelling x86 value it still has.