AMD Mantle API [updating]

Would that be a Mantle piece ?

*available for weddings and bar mitzvahs

How does one go about booking your services?

I'm still not seeing anywhere that Mantle is completely open with no royalties etc. Everything seems to indicate it's only specifically coded to GCN at the moment.

I'm sure the legal aspects will be covered in the November conference.

It's the technical front I'm a bit more worried about. If Mantle is specifically design to be the best, most efficient, closest-to-the-metal API for GCN, then how easy/difficult would it be to port it for, say, Kepler or Maxwell?

And for that matter, how disruptive would it be if AMD decided to make significant changes to GCN?
 
The way I'm reading the tea leaves is that it'll be designed with no consideration for competitor's HW. So I would expect it to be quite closely tied to GCN features. Maybe even at the "here's our command buffer format, feel free to build them yourself and chuck pointers at them to our GPU" level.

Regarding significant revisions to GCN being disruptive: I could see that being the case, but... given that this seems mainly aimed at huge engine development houses and that the 2 next-gen consoles are probably going to be hanging around for at least half a decade, does that matter? The big middleware devs will want to use it for the console versions of their engines anyway, so the fact that some range of GCN equipped PCs can piggyback on those optimizations and (at least for a while) make the competition look bad just seems like gravy.

My prediction is that yes, at some point the mantle will get quite crusty, and AMD will seize the opportunity to use volcanic eruptions, huge fault lines, or asteroids as the basis for future product code names.
 
There is absolutely zero chance that MS doesn't know about Mantle because Drivers need to be certified for Windows right?!
Certification verifies whether stuff to be certified works as expected. That means it checks whether WDDM interfaces and DX APIs work as expected. It would never touch non-WDDM stuff. There may be other non-technical conditions I'm not aware of but from the code point of view I could probably expose coffee brewing interfaces in our driver and it should be OK (except that it wouldn't pass code review and I'd get fired for trying to do that).
 
To me it's pretty obvious that Nvidia will work on their own API. Their roadmap kind of imply it.Maxwell should integrate Denver cores, while Volta stacked DRAM. How can you expose this kind of innovation to devs? How can you stay competitive with AMD if you are wasting transistors on something that won't be used outside GPU computing?

I don't see it as a problem: 90% of games will run on Cryengine, Unreal Engine 4, Frostbite 3 and few more. They can target two APIs, as long as Nvidia mantains a good marketshare in the pc world.

I'm excited about this move because I think it will lead to a quicker innovation pace in the GPU industry: until now innovation was slowed down from what was exposed in DirectX (ie trueform/tessellation tooks years). Giving the bigger and bigger constrain over power-consuption, we may see a "come-back" of fixed function units. (i.e DSP integrated in GPU)

If AnandTech guess is right, then we may see interesting things on next-generation GNC architectures... something like a on-die scratchpad. :D They have the chance to expose it and if it's used in games for X1, it could be used also on desktop.
 
High performance API for future and existing hardware: Good.
Not being able available to competitors nVidia and Intel (if true): All kinds of bad.

Hey AMD, how about trading Mantle support for official CUDA/PhysX support in GCN GPUs?
Everyone would win.

And then Intel could just pay for it. They have money.


Honestly though, if this doesn't leave AMD's territory, I hope it fails. No customer wants a monopoly.
 
Honestly though, if this doesn't leave AMD's territory, I hope it fails. No customer wants a monopoly.

The way I see it, Nvidia would eventually find something proprietary to monopolise the market with. God knows they try hard enough. Better that AMD does it with this.
 
High performance API for future and existing hardware: Good.
Not being able available to competitors nVidia and Intel (if true): All kinds of bad.

Hey AMD, how about trading Mantle support for official CUDA/PhysX support in GCN GPUs?
Everyone would win.

And then Intel could just pay for it. They have money.


Honestly though, if this doesn't leave AMD's territory, I hope it fails. No customer wants a monopoly.

Where do you see monopoly when AMD is the smallest out of the three and needs time and space to grow?

I honestly would welcome struggling nvidia and intel first because I don't need them and second because they did enough, so I hate them...

Me as a customer- I want a monopoly of AMD ;)

Anyway, as someone else pointed out- if Mantle goes to uniform support by all vendors, it will introduce performance losses and its purpose to great extent would be lost...
 
Oh come on Dave, don't tease us like that !
Release the specification & SDK damnit !
:p
 
Where do you see monopoly when AMD is the smallest out of the three and needs time and space to grow?

I honestly would welcome struggling nvidia and intel first because I don't need them and second because they did enough, so I hate them...

Me as a customer- I want a monopoly of AMD ;)

Anyway, as someone else pointed out- if Mantle goes to uniform support by all vendors, it will introduce performance losses and its purpose to great extent would be lost...

I'm not sure what metric you're using, but I think most people would consider NVIDIA to be the smallest player, by an ever larger margin if you factor in the console GPUs.
 
Oh come on Dave, don't tease us like that !
Release the specification & SDK damnit !
:p

It is coming right up after the HSA stuff that was coming right up last year. No really.

UniversalTruth: Take a month, ponder upon what I explicitly and repeatedly told you not to do and you keep on doing, and then figure out if B3D is the place for you.
 
You had to put that in both threads? Didn't you realise that nobody really cares about whether or not it's proprietary?

What's more, it's absurd to believe that AMD would make this as open as possible asap? Time is of the essence here, they have to get a working version out before they can consider another lagging company's welfare.
 
Last edited by a moderator:
Also nice bit of selective quoting, leaving out that he said if a competitor came over to talk about implementing Mantle for their hardware, they wouldn't dismiss that out of hand.
 
http://forum.beyond3d.com/showpost.php?p=1789339&postcount=18



So much for the myth about AMD promoting only open standards and Nvidia being the big bad proprietary company. Lets see how the AMD defense force tries to spin and damage control this one. Enjoy your Glide API 2013.

Howabout the part where x86 + GCN underpins both new console platforms, which makes for a stronger likelihood of cross-platform games via Mantle? Open or not, this is the sort of thing that NVIDIA simply cannot claim with CUDA or PhysX.
 
Back
Top