AMD Mantle API [updating]

Ah, I understand now. I would assume so, but that may be a terrible assumption on my part

 

AMD fans cant really hear you from the sound of partying.

And lets not forget sentence written just after your quote:
"He also tried to brush aside comparisons with Glide but then stated: If a competitor were to approach AMD to make their own backend and drivers for Mantle, AMD would not dismiss them right away."

 

Let me know when it's on github...

 

I wouldnt say thats true, I dont really want to be faced with a choice of " I can buy an amd card and enjoy games a, b and c enhanced with mantle or I can buy a nvidia card and enjoy games x, y and z enhanced by nv's to the metal api".

 

Surely the more immediate worry is having to make a choice between buying AMD to enjoy game A (and possibly game B) enhanced with Mantle, whilst at the same time having to put up patchy driver issues and cheesy-grin-excuses performance in games C though Z that happen not to support Mantle. Is A+B > sum(C:Z)?

 

Yeah at a minimum Amd has forced them to do so with Mantle, as there is no way NVidia can now sit back and be badly beat on gaming benchmarks on AAA games. It's just Frostbite games today but Unreal tends to support everything so I'd presume them supporting Mantle is all but guaranteed. Cryengine prides itself as being the best looking and performing engine out there so likewise there is no way they can sit out of this either. Between those three engines, Amd would be beating NVidia on most game benchmarks out there.

That's why I don't really mind this now. Back in the glide days it was a different story, but most games today use a handful of engines so the time is right to go this route. What I wonder though is how clean and lean Mantle will be in 5 gpu hardware revisions from now.

 

You just touched on what's probably the biggest difference between now and the Glide days: middleware. How many games roll their own tech now? You get Unreal Engine, Cryengine, Unity and Frostbite and you've got almost the entirety of the game market covered.

 

Things get weird if they don't get all the engines, or a competing API divides things up. Hopefully that doesn't happen.


Some of the recent changes for GCN, Sea Islands, and the consoles make me wonder which of those are precursors or hints as to what AMD was going for.

There are a number of things that may have contributed to making Mantle practical from a technical standpoint.

GCN aligned its virtual memory system with that of x86, and it's been pushing hard on updating the IOMMU.
These would add an extra layer of security for the rest of the system, by keeping malformed low-level GPU code from going where it shouldn't. The API and DX path for Windows would have been barriers, although there have been some graphics layer exploits or priviledge escalations.

Sea Islands and one or more consoles adds user-level compute queues, so AMD has been (hopefully) working on GPU hardware, firmware, or software that is more robust in taking inputs that haven't gone through the traditional submission chain.
Mantle may mean that further work has been done for other command types.

The consoles both appear to have dual graphics front ends. Hawaii apprently has doubled triangle setup capability, although CP count and setup pipe numbers don't necessarily affect each other.
Possibly this is one way to keep the OS on any platform happy that the GPU is responsive?
AMD does want to bring out GPU QoS and preemption, which someday would keep funky Mantle code from DoSing a system and might eventually give standard graphics drivers a way to not freak out when GPU shaders start having longer and longer run times.

One possibly faint silver lining for DirectX: whatever issues AMD and DICE run into getting this to work is something Microsoft might want to see someone else take the hits for, anyway. If something scary comes up that wrecks things, Microsoft has clean hands or a starting point for a working version.
As far as standards go, AMD might want a CUDA/OpenCL situation, where bothering to do things first and getting it out there influences what future standards internalize.

 

The only reason to want a monopoly from any graphics vendor is if you dislike performance increases and feature innovation. If AMD had had a graphics monopoly since the 360 launched we'd be lucky to GPUs as fast as the 5870 by now.

As for mantle, I think its awesome as long as NV matches it quickly and big devs are happy to support 2 extra API's. I'm locked into NV now thanks to my 3D vision monitor but if I weren't my sites would be firmly set on an AMD GPU.

 

This kind of bias (DSC as well) seems pretty silly to me. We should be PC gamers, not AMD or Nvidia gamers. And while I do worry about the advantage this could give AMD (on top of everything else) if it pushes NV to do something similar and devs support both, it could be pretty incredible. Bye bye PC overhead!

 

Maybe I'm just imagining this, but last night I heard the word "open" mentioned several times in conjunction with Mantle. It was my understanding from that, that if nVidia wanted to get in on Mantle AMD was not going to refuse them. The only problem may lie in what Mantle is...nVidia may want to look at it & set up some drivers to run the API (just as AMD and nVidia both run the same D3d high-level API but with different drivers and hardware), but being low-level based on specifics in AMD's hardware, nVidia may find the API less that amenable to its driver and hardware environment. I don't know enough about Mantle to guess in that direction, though...

Key, though, I think to Mantle is that it transcends Microsoft's D3d and brings to light D3d's obvious weakness for game developers--its lack of cross-platform cohesion. I mean, these new consoles, which are more like PC's than they are like traditional consoles, finally!, present an unprecedented opportunity for game developers--the ability to use the same basic sets of tools to develop for PC, xBone, and PS4 simultaneously! All that's needed is an API exactly like Mantle--and AMD is obviously the only company capable of doing that, since it manufactures both consoles and has been selling AMD cpus and (ATi) gpus in the PC space...! Nobody but AMD could pull this off.

This surely isn't a repeat of Glide, despite what the spoilsports prefer to think... I cannot see how this remotely resembles GLIDE even superficially. Glide was one company, one hardware IHV, and Windows. Prior to Microsoft announcing D3d, I recall the scuttlebutt being widespread that Microsoft had selected GLIDE as the official Windows 3d API--GLIDE development being way ahead of D3d development. It never happened, frankly, because 3dfx insisted on keeping GLIDE proprietary and was not amenable to opening it up to anyone else who wanted in. As a result, it wasn't until DX7 or so that D3d gained a rough parity with Glide on the feature-support front. (I was a great 3dfx cheerleader in those days--hey, many of us were! Nobody had much of a crystal ball back then and everyone was flying by the seat of his pants.)

Today things just got organized to a much higher degree, with AMD being at the helm of both consoles and selling the former ATi brand hardware to very good effect. As I said, AMD is in an unprecedented situation here and Mantle seem like just what the doctor ordered.

What nVidia will do with it is anybody's guess, but I think nVidia would be rather silly not to get in on the Mantle fun if that proves possible for nVidia to do. Most importantly for Mantle, however, will be how it plays out with developers in actual game development. First signs are very encouraging, though.

 

AMD said it wouldn't refuse them off the bat.
That's not the same as saying it's open, and Nvidia would need to ask.
Even then, it might not work out--assuming both parties have an interest in things working out.
If Mantle catches on and dominates, AMD might rather Nvidia not be along for the ride.
If Nvidia can make things uncertain enough that Mantle founders, then it might want to step aside while AMD walks off a cliff.

 

Raja's response left me thinking, did ATI/AMD ever approach Nvidia to license CUDA and they were refused?

 

Guys... there is a huge difference between being willing to disclose a spec for an architecture-specific API and designing a portable API. All indications are that Mantle is designed quite specifically for GCN, and I doubt AMD made any compromises for portability's sake (that was not the goal of it). Thus saying "well it's open and we'd let other people support it" is likely just PR nonsense, since they know very well that likely no one else can support it as it is defined.

Of course this is speculation, but seriously... if it wasn't specific to GCN then there wouldn't really be a need for it. They could do the same thing with GL extensions otherwise.

... and that's totally fine; the goal here was for a very low level API to drive GCN, not something portable or forwards/backwards compatible. But I think it's safe to assume that it won't be supportable on other hardware until we get more details that show otherwise.

 

My interpretation of the "open" statement was more along the lines of not charging a fee to use their language / compiler / tools / whatever is necessary to "write in Mantle" for your production, profitable code. I also similarly extend that interpretation to mean they aren't going to license it at an operating system implementation level, either. Of course, since you depend on AMD's driver to support your OS, that's probably more of a red herring.

 

There are already extensions for AMD that only GCN supports, so is that the reason?
They could just not want to share, so even if there were an AMD not-GCN architecture for Mantle, they would still not create extensions.

Are there semantics or state requirements that the low-level API can violate if it were implemented that way?

GL_AMD_Fake_OpenGL_Bro?

 

Isn't this an academic discussion anyway? HSA is a very open initiative with multiple tech founders and most of the tech industry participating in some fashion, yet who is completely absent from participation? NVIDIA and Intel; both of these companies are making or developing products that have direct applicability to HSA yet they choose not to participate. If Mantle is considered open or not, would if make a blind bit of difference in relation to NVIDIA?

 

IMG is a founding member of the HSA foundation, as is ARM and Qualcomm.
There are other contributing members with graphics IP or with more direct ties to AMD's APU products like Sony.

If there's a question of being a graphics provider with membership in HSA, Mantle currently doesn't consider it relevant.

 

Yes, if Nvidia created their own low-level API and developers started to support both Mantle and "Glide", age of DirectX PC dominance would truly start to fade.

Now that's a good thought.

AMD could left some openings for expansion of API when some future different GPU architecture arrives. There is no point in creating Mantle if it will become obsolete in 5-6 years.

 

Yup, "Mantle" is not open at all really. Mantle can be used across different platforms (aka "open" platforms), but it is still an IHV-specific API for GCN-equipped products. Since Mantle requires some extra development work vs. industry-standard API's, and since Mantle is essentially unuseable for NVIDIA/Intel/Qualcomm/ImgTech/ARM GPU-equipped products, Mantle will ultimately fail to gain traction in the marketplace in my humble opinion. AMD will need to pay developers to use Mantle, and that strategy will only go so far considering their dwindling cash balance and heavy debt load.

AMD is really opening up a can of worms here. If all the top companies decided to pursue IHV-specific graphics API's, the result would be disastrous and chaotic to PC game development. And the ramifications go beyond just a chaotic environment, because as John Carmack said, Sony and Microsoft are surely going to be very upset about this.