The AMD Mantle Thread

[psoomah]

Ah ok...you weren't clear, and I know the whole "Mantle isn't on consoles" thing has been beaten to death. Thanks.

You're welcome.

 

[Erenhardt]

/rant end

One reason - mantle is not ready yet.
The SDK is still under development. AMD is working with their partners - Mantle bandwagon.
What you have to understand is, mantle patch for BF4 will be here before the mantle SDK is even out.
Given the problems BF4 had. I wouldn't be surprised to see mantle patch delayed.
One thing to note aswell is, it is not BF4 patch, but Frostbite patch. The work they are doing it implement mantle hits the core - game engine itself. That can have snowball effect.

 

[blastingcap]

Sorry if this is a repost but I thought it was a good explanation of Mantle that should address some of the misinformation I hear sometimes:

http://techreport.com/review/25683/delving-deeper-into-amd-mantle-api/2

I suspect AMD learned a lot when cooperating with Microsoft and Sony on next-gen consoles and are using some of what they learned in Mantle.

 

[krumme]

Sorry if this is a repost but I thought it was a good explanation of Mantle that should address some of the misinformation I hear sometimes:

http://techreport.com/review/25683/delving-deeper-into-amd-mantle-api/2

I suspect AMD learned a lot when cooperating with Microsoft and Sony on next-gen consoles and are using some of what they learned in Mantle.

There is tons of compute power in a ps4 or 7850. I think our fps games lack dynamics. Bf4 is superior but a bf5 with use of all the compute grunt and true sound will take it to a new level for the gaming experience. I think we will have to wait 2 years for that effect to truely shine true.

Ofcourse its smart from amd to use the consoles as a leverage for compute. Asymetric use of mumtiple gpu. And pave the way for hsa with a new memory model.

I also think Avx2/3 is taking a hit back here. There is only so much focus you can have. Until mantle i thought compute was dead but it looks like that have changed dramatickly.

And yes in many ways its natural amd made THE API as they now have all the experience and competence. They can make it the best because they have broad inside knowledge from x86, gpu, apu, consoles, and also to a certain degree still production. So in that perspective its no surprise.

 

[psoomah]

This appeared on a slide near the end of AMD's Mantle presentation at APU13. (around the 35 minute mark)
...........................................................
CrossFire Unleashed

►Going beyond AFR rendering
►Flexible workload scaling and partitioning
►Asymmetric configurations (APU+dGPU)
►Unlocking novel usage scenarios
.............................................................

Basically it means Mantle can address, and flexibly distribute the workload between, an APU and a GCN dGPU at the same time.

Now your APU's iGPU is additive to your AMD AIB!!!

Makes pairing your AMD APU with an Nvidia AIB a losing proposition.

If that APU is a Kaveri with hUMA, heterogeneous Unified Memory Address capability, and the dGPU is an R9 290 card with it's system unified addressing capability (third Q. from bottom), you essentially have a PS4 console APU with 4 Kaveri cores and an R9 290xxx iGPU.

I'm salivating at the thought of seeing Kaveri + 290 combo BF4 Mantle reviews.

Until Mantle becomes available to, and is implemented by, Intel and Nvidia, any AMD APU + GPU combo will lead pipe any Intel CPU + Nvidia GPU combo on cost/performance in Mantle games.

I'm thinking AMD Kaveri + GCN GPU combo Steam Machines will sell like maple bacon doritos.

 

[krumme]

This appeared on a slide near the end of AMD's Mantle presentation at APU13. (around the 35 minute mark)
...........................................................
CrossFire Unleashed

 

[psoomah]

Uh huh, that's right.

 

[krumme]

The execution model seems to add complexity.

But listening to the devs at the q&a the paradox is that added complexity removed complexity from eg debugging/optimizing. They now know eg how much mem a texture uses.

So the end result is by adding complexity and control in one part it actually made it simpler to program for.

The thin driver layer and using mantle is making it far more easy to program for. I think its incredible and a bit of a paradox.

Dx with all the added layers during the years have ended with a bloated products that is both relatively slow and very difficult to use for performance games. Ms was sleeping as usual. And is now 5 years late.

 

[krumme]

Uh huh, that's right.

Its a bug with tapatalk. Happens sometimes at AT. My post was lost

 

[psoomah]

Its a bug with tapatalk. Happens sometimes at AT. My post was lost

Was wondering. Thanks.

 

[psoomah]

Dx with all the added layers during the years have ended with a bloated products that is both relatively slow and very difficult to use for performance games. Ms was sleeping as usual. And is now 5 years late.

Yep, DX is a dead API walking. Mantle is such a developer wet dream, such a Holy Grail + Swiss Army Knife, I don't see DX having any chance to become relevant again, no matter how busy Microsoft gets with it.

Well, if AMD makes Mantle available to Nvidia and Intel in a timely manner anyways.

 

[Tuna-Fish]

Basically it means Mantle can address, and flexibly distribute the workload between, an APU and a GCN dGPU at the same time.

It can't, you misunderstood. Mantle allows the game devs to do that if they choose to do it. The API does nothing but allows them the tools to do so. It all then boils down to whether or not they think it's worth the time.

This is actually one of the weaker points of Mantle. Traditional SLI/XFire simply won't work because providing the same interface to one or two gpus can only work with a high-level api that hides memory allocations. Because Mantle allows the devs fine control over memory allocation and work management, multi-GPU support turns from something done in the driver to something that needs to be done in the game.

The good part of this is that at it's best, it allows much better multigpu support -- no AFR, no stutter. At it's worst, it means no multi-gpu support, if the devs simply don't find it worth their time.

 

[SiliconWars]

It can't, you misunderstood. Mantle allows the game devs to do that if they choose to do it. The API does nothing but allows them the tools to do so. It all then boils down to whether or not they think it's worth the time.

This is actually one of the weaker points of Mantle. Traditional SLI/XFire simply won't work because providing the same interface to one or two gpus can only work with a high-level api that hides memory allocations. Because Mantle allows the devs fine control over memory allocation and work management, multi-GPU support turns from something done in the driver to something that needs to be done in the game.

The good part of this is that at it's best, it allows much better multigpu support -- no AFR, no stutter. At it's worst, it means no multi-gpu support, if the devs simply don't find it worth their time.

You're right that the devs need to do it themselves but they'll probably all come to the same kind of conclusion. Perhaps the core game runs on the first dGPU with an APU doing SSAA for "free" as it were. That alone is pretty huge. A second dGPU might be capable of SSAA and some other compute based stuff as well. I guess this is one area that will see some real advancement over the coming years. We could be set for some incredibly advanced visuals that just weren't possible before.

 

[psoomah]

It can't, you misunderstood. Mantle allows the game devs to do that if they choose to do it. The API does nothing but allows them the tools to do so. It all then boils down to whether or not they think it's worth the time.

Oh ... ... ... you mean like in Tron?

 

[mindbomb]

I think crossfire under early mantle games is actually not gonna turn out great. It has the potential to be really good, but idk if game developers are gonna put in the effort to find good ways to make it scale when only a small fraction of the total mantle users are gonna be using crossfire. They already have a lot on their plate as it is managing both directx and mantle versions of the game.

 

[selni]

This appeared on a slide near the end of AMD's Mantle presentation at APU13. (around the 35 minute mark)
...........................................................
CrossFire Unleashed
►Going beyond AFR rendering
►Flexible workload scaling and partitioning
►Asymmetric configurations (APU+dGPU)
►Unlocking novel usage scenarios
.............................................................
Basically it means Mantle can address, and flexibly distribute the workload between, an APU and a GCN dGPU at the same time.

Now your APU's iGPU is additive to your AMD AIB!!!
Makes pairing your AMD APU with an Nvidia AIB a losing proposition.
If that APU is a Kaveri with hUMA, heterogeneous Unified Memory Address capability, and the dGPU is an R9 290 card with it's system unified addressing capability (third Q. from bottom), you essentially have a PS4 console APU with 4 Kaveri cores and an R9 290xxx iGPU.
I'm salivating at the thought of seeing Kaveri + 290 combo BF4 Mantle reviews.

Until Mantle becomes available to, and is implemented by, Intel and Nvidia, any AMD APU + GPU combo will lead pipe any Intel CPU + Nvidia GPU combo on cost/performance in Mantle games.

I'm thinking AMD Kaveri + GCN GPU combo Steam Machines will sell like maple bacon doritos.

So a few things on this. I think some scepticism is warranted if for no other reason than AMD have been claiming this works for some time (hybrid crossfire) when anyone who's tested it knows it's a horrible mess.

There's also a huge difference between unified address space and unified memory. Even what NVidia is doing in the latest version of CUDA (essentially just hiding the memcpys) is a unified address space. The APU and GPU still have physically separate memory - yes you can do better than what NVidia have done, but the APU and GPU accessing each other's memory is still clearly going to have to be over PCIe and that's both already used for other data and even slower than dual channel DDR3 letalone a wide GDDR5 bus. You can't avoid this with a discrete GPU.

There's also the sheer difference in scale here - your example (kavari + R290) isn't exactly balanced - the R290 is an absolute monster in comparison. AFR is totally out of the picture, so what do you offload to the APU? You can't do anything too bandwidth heavy because you've got a unified address space, not unified memory and everything has to go over PCIe. There's certainly performance to be gained here (perhaps just using the APU for general compute, but why do you need mantle for that?), but it's tricky.

 

[3DVagabond]

It can't, you misunderstood. Mantle allows the game devs to do that if they choose to do it. The API does nothing but allows them the tools to do so. It all then boils down to whether or not they think it's worth the time.

This is actually one of the weaker points of Mantle. Traditional SLI/XFire simply won't work because providing the same interface to one or two gpus can only work with a high-level api that hides memory allocations. Because Mantle allows the devs fine control over memory allocation and work management, multi-GPU support turns from something done in the driver to something that needs to be done in the game.

The good part of this is that at it's best, it allows much better multigpu support -- no AFR, no stutter. At it's worst, it means no multi-gpu support, if the devs simply don't find it worth their time.

You say it like it's a bad thing. This is exactly the kind of control the devs have been asking for. Besides, are you sure it has to be done by the app (per game), or is it something that can be incorporated into the game engine?

 

[3DVagabond]

I think crossfire under early mantle games is actually not gonna turn out great. It has the potential to be really good, but idk if game developers are gonna put in the effort to find good ways to make it scale when only a small fraction of the total mantle users are gonna be using crossfire. They already have a lot on their plate as it is managing both directx and mantle versions of the game.

Face it, we have no idea. We can rationalize it any way that suits us, but we have no actual knowledge of how any of this is going to end up.

 

[monstercameron]

I cant substantiate this claim but I don't believe engine dev takes the lion share of resources. Asset production seems to be the big issue, especially when most studios typically license game engines like cryengine, unreal or unity. If amd can get these engine devs to write mantle backends [or better custom crossfire profiles] then mantle could literally be made into the majority of games...
all this talk about x money going being paid off by amd and so forth makes no sense, we are in an era where a studio might not even need a classical backend developer to develop game but a well rounded team of artists who can use the tools bundled with the licensed game engines. I am pretty disappointed in the vision many of the mantle detractors on this forum, sure mantle isn't fully baked nor has it proved its metal but the technology is promising.

 

[Obsoleet]

Yep, DX is a dead API walking. Mantle is such a developer wet dream, such a Holy Grail + Swiss Army Knife, I don't see DX having any chance to become relevant again, no matter how busy Microsoft gets with it.

Well, if AMD makes Mantle available to Nvidia and Intel in a timely manner anyways.

Wait.
So now DX is a dead API walking, yet less than a year ago I posted about my general enthusiasm about OpenGL's future and everyone walked out with anti-OGL tirades that sounded like they were from 10 years ago.
http://forums.anandtech.com/showthread.php?t=2255915&highlight=

With SteamOS on the way, and Macs being as popular as they've ever been I think we can say that targeting OGL+Mantle and leaving out DX is starting to make sense.

Oh, and I think they should open it up to Intel but not Nvidia. NV wouldn't open it up to them, and Intel actually has something to offer AMD- cooperation in the x86 space (opening their fabs to AMD, patent sharing ect), and making Mantle an unquestioned standard overnight if Intel HD graphics support it. NV is not necessary, and should be left in the dark.

 

[blastingcap]

DX isn't dead, it's not like Microsoft can't spend more effort making it better. But don't get me wrong, I really want Linux/SteamOS to succeed to break the current Microsoft PC gaming deathgrip. I can't trust them to really try very hard so long as they have XBOX.

P.S. The whole "single virtual GPU" thing would be great if it actually worked, or even half-worked so that iGPUs/APUs weren't wastes of space when paired with dGPUs.

 

[krumme]

The asymmetric use of dgpu and apu will be very low priority for devs as the market is small. I mean what is the market for kaveri + 290. And what sense does it make to program for assymetric here?

Ofcource it all depends how easy it is for devs to implement in the engines. I dont think it will fly and amd shouldnt promote it so hard. Its an imidiate solution before its 99% apus in 6 years. I wouldnt invest in it if it was ea.

Regarding hsa/huma i dont think the consoles or mantle is going to be the decisive factor for the future development. But amd betting on the mobiles and java development here seems to be spot on. Its a very strong move long term strategically imho.

 

[selni]

DX isn't dead, it's not like Microsoft can't spend more effort making it better. But don't get me wrong, I really want Linux/SteamOS to succeed to break the current Microsoft PC gaming deathgrip. I can't trust them to really try very hard so long as they have XBOX.

P.S. The whole "single virtual GPU" thing would be great if it actually worked, or even half-worked so that iGPUs/APUs weren't wastes of space when paired with dGPUs.

It's not the xbox specifically, Microsoft only try when they're losing. When they're winning they sit on their hands and do literally nothing for years until something unexpected bites them in the ass and they're behind again. It's happened so many times now...

 

[VulgarDisplay]

DX isn't dead, it's not like Microsoft can't spend more effort making it better. But don't get me wrong, I really want Linux/SteamOS to succeed to break the current Microsoft PC gaming deathgrip. I can't trust them to really try very hard so long as they have XBOX.

P.S. The whole "single virtual GPU" thing would be great if it actually worked, or even half-worked so that iGPUs/APUs weren't wastes of space when paired with dGPUs.

I have less than no confidence that microsoft is smart enough to dig themselves out of any of the holes they have put themselves in recently.

 

[Skurge]

Also, while Graphics Core Next is the "hardware foundation" for Mantle, AMD's Guennadi Riguer and some of the other Mantle luminaries at APU13 made it clear that the API is by no means tied down to GCN hardware. Some of Mantle's features are targeted at GCN, but others are generic. "We don't want to paint ourselves in a corner," Riguer explained. "What we would like to do with Mantle is to have [the] ability to innovate on future graphics architectures for years to come, and possibly even enable our competitors to run Mantle." Jurjen Katsman of Nixxes was even bolder in his assessment, stating, "There's nothing that I can see from my perspective that stops [Mantle] from running on pretty much any hardware out there that is somewhat recent."

For People who kept saying Mantle is locked to GCN.