Mantle not restricted to GCN

[ajoy39]

NVAPI? Nvidia API? It's been around for 2 years (ish)? 

not the same

[ACatWithThumbs]

Sooo..... Performance numbers?

They didn't release any

So we have to wait now till december for the Bf4 Mantle Patch.

[GoodBytes]

Invalid point!

AMD SAYS that Mantle is Open.. but it needs to comply to a set of requirements and specification in order to work.

Does that involve using AMD patents? Does that mean that essentially Nvidia needs to build an AMD GPU architecture in order to use it?

We don't know any of this, and AMD is not saying a thing.

So, I call B.S on that claim.

[TechFan@ic]

Invalid point!

AMD SAYS that Mantle is Open.. but it needs to comply to a set of requirements and specification in order to work.

Does that involve using AMD patents? Does that mean that essentially Nvidia needs to build an AMD GPU architecture in order to use it?

We don't know any of this, and AMD is not saying a thing.

We actually do.

The first question is irrelevant since TressFX & MLAA are patented AMD technologies and Nvidia products can utilize both.

The answer to the second question is an unequivocal no.

Most Mantle functionality can be supported on today's modern GPUs

What are "modern GPUs" ? they are GCN based ones from AMD (7000 series & up) & Kepler based ones from Nvidia (600 series & up).

Johann also mentions "Extensions for architecture specific functionality"

This means Nvidia can develop its own extensions and driver layer for Mantle if it chooses to do so. Just like AMD can and will develop new extensions for future GPU architectures.

This was already confirmed by AMD's software head Raja Koduri.

What AMD may request in exchange is unknown at this point in time, my guess would be nothing, because establishing Mantle as an industry standard would already be a massive win for AMD.

 

[GoodBytes]

We actually do.

The first question is irrelevant since TressFX & MLAA are patented AMD technologies and Nvidia products can utilize both.

The answer to the second question is an unequivocal no.

What are "modern GPUs" ? they are GCN based ones from AMD (7000 series & up) & Kepler based ones from Nvidia (600 series & up).

There is a difference between... "We have a set of API that does this and that.. you can use it but pay us", and In order to implement Mantle, it will work with the latest architecture that you have, you just need to do this and that, plus you need to modify 100000000 things. Where these things can essentially turn Nvidia GPU to an AMD one at the end of the day. Also, it creates dependency on AMD graphic card, which Nvidia must implement in order to continue to get mantle to work, block Nvidia from making something drastically different (innovating), and make new more powerful products.

It's a trap for Nvidia.

[ajoy39]

There is a difference between... "We have a set of API that does this and that.. you can use it but pay us", and In order to implement Mantle, it will work with the latest architecture that you have, you just need to do this and that, plus you need to modify 100000000 things. Where these things can essentially turn Nvidia GPU to an AMD one at the end of the day. Also, it creates dependency on AMD graphic card, which Nvidia must implement in order to continue to get mantle to work, block Nvidia from making something drastically different (innovating), and make new more powerful products.

It's a trap for Nvidia.

Err, no thats not what it is at all

[AlexGoesHigh]

There is a difference between... "We have a set of API that does this and that.. you can use it but pay us", and In order to implement Mantle, it will work with the latest architecture that you have, you just need to do this and that, plus you need to modify 100000000 things. Where these things can essentially turn Nvidia GPU to an AMD one at the end of the day. Also, it creates dependency on AMD graphic card, which Nvidia must implement in order to continue to get mantle to work, block Nvidia from making something drastically different (innovating), and make new more powerful products.

It's a trap for Nvidia.

 

if that's what AMD wants, is a very dark clever idea...

[TechFan@ic]

There is a difference between... "We have a set of API that does this and that.. you can use it but pay us", and In order to implement Mantle, it will work with the latest architecture that you have, you just need to do this and that, plus you need to modify 100000000 things. Where these things can essentially turn Nvidia GPU to an AMD one at the end of the day. Also, it creates dependency on AMD graphic card, which Nvidia must implement in order to continue to get mantle to work, block Nvidia from making something drastically different (innovating), and make new more powerful products.

It's a trap for Nvidia.

How can additional performance be a trap ?

Mantle at its current state without any specific architectural extensions is already going to be faster than DX or OpenGL.

Nvidia can essentially add Mantle support for additional performance at no developmental cost & that's what AMD is offering.

If Nvidia wanted even more performance it can get into the nitty gritty architectural optimizations via the aforementioned extensions.

Otherwise they get a performance improvement regardless.

 

The only reasons for Nvidia not to support Mantle are corporate and the only reasons for Nvidia to support Mantle is the benefit to all end-users and the greater good of the industry.

Which will end up making Nvidia more money ? that's what it comes down to.

[LAwLz]

I don't get how this is suppose to be low level access if it isn't tied to a specific architecture. My guess is that 1) it's not actually that low level, 2) it is not low level if you implement it on other architecture, making it kind of useless, 3) Johan is wrong 4) they are using black magic.

 

Still not going to buy into the Mantle hype. Still barely any solid facts and I have been disappointed by overhyped products way to many times in the past. When we start seeing real numbers, then I'll get excited.

[mgsstar]

So mantle won't be as unoptimized as Physx if it was on Nvidia cards? AMD would probably would want to gain some profit if they were to give to Nvidia.

[LOST TALE]

So mantle won't be as unoptimized as Physx if it was on Nvidia cards? AMD would probably would want to gain some profit if they were to give to Nvidia.

they should with nvidia being reserving with G-sync

[ACatWithThumbs]

PCper recorded the hole event where there talking about Mantle here:

http://www.youtube.com/watch?v=tDPgJB2x7dQ

[martinrox1568]

It means AMD's GCN I think, If Nvidia made a GCN card it would work then.

you realize what gcn is, right?? It's not just a name on the box. It's an architecture. Nvidia can't make GCN because hell, they wouldn't no how it works, and vice versa.

[ACatWithThumbs]

They even said at the conferrence that they hope other GPU vendors to use it.
And the only other vendor is Nvidia.
The best thing that could happen to PC gaming would be that Nvidia and AMD use Mantle and boost up the PC gaming peformance for everybody.
Nvidia and AMD should fight with Hardware not with software.

[Apocalypse]

wait its not tied to GCN ?

the why is it called "low level"

 

anyways if works with anything then just throw away Mantle and pour more money on opencl

 

It is called low level because if gives you access nearly driver level to the hardware. There are still some abstraction layers. 

It is low level like c vs java, and not like assembler vs c 

 

Actually one of their driver developers even stated that you can do nearly all of the usual driver optimisation now on your own, because you have that level of control over your code. Same goes for async and asymmetric use of GPU cores. With mantle it is for example possible to use the GPU of your APU for post-processing or compute or whatever, while your dedicated card(s) is doing the major stuff. Even more important: The game developer can decide what parts of his code should run on which gpu, if he needs to duplicate parts of the memory or not, etc

 

Yes, this means as well, that 2x290 for example can use with mantle 8gb of ram, not just 4gb like in crossfire. Mind in this case, you as developer may still decide to use memory duplicated for performance reasons, you may decide against for again performance reasons. Its a lot more power, but as well a lot more responsibility for developers. 

 

 

They even said at the conferrence that they hope other GPU vendors to use it.

And the only other vendor is Nvidia.

The best thing that could happen to PC gaming would be that Nvidia and AMD use Mantle and boost up the PC gaming peformance for everybody.

Nvidia and AMD should fight with Hardware not with software.

 

There is Intel, PowerVR (iphone, galaxy s4), Qualcomm's Adreno (Nexus 5, LG G2), ARM Mali (Nexus 10, and I think some Galaxy Note III and Chromebooks). Mantle is not aimed on PC alone. It is aimed to simply be faster and better than openGL. 

[brownninja97]

In a few months depending on how well it goes nvidia will probably pay a few million to buy it.

[Humbug]

Johan wants to see Mantle on Linux, also he said that they are building an opencl/gl extension in FB3

Ya AMD is also making significant contributions now to the Linux kernel to improve performance

http://www.tomshardware.com/reviews/amd-ama-toms-hardware,3672-5.html

 

It's pretty impressive Gabe has actually managed to get important parties to work on Linux stuff.

[TheGhostVII]

Mantle and G-Sync would work wonders!

Maybe this is a sign G-Sync will be available for AMD as well.

[digitalnav]

Mantle and G-Sync would work wonders!

Maybe this is a sign G-Sync will be available for AMD as well.

I wouldn't count on it haha

[TheGhostVII]

I wouldn't count on it haha

A man can dream.

 

But seriously, I don't think monitor manufacturers will put G-Sync in their product if only half of their customers can use it.

[Humbug]

I hope AMD makes mantle an open standard rather licensing it.

 

I doubt Nvidia would let AMD touch G-Sync, but i'm sure AMD can develop their own version to allow their VGAs to sync with monitors. Although this may end up working in a similar fashion to gsync which may be a problem if nvidia has patented it.

[TheGhostVII]

I hope AMD makes mantle an open standard rather licensing it.

 

I doubt Nvidia would let AMD touch G-Sync, but i'm sure AMD can develop their own version to allow their VGAs to sync with monitors. Although this may end up working in a similar fashion to gsync which may be a problem if nvidia has patented it.

Call it A-Sync and say it works by magic and Christmas cheer.

Nobody can patent magic and Christmas cheer!

[Boo Berry]

And the only other vendor is Nvidia.

 

You forgot Intel and its HD graphics. There's two graphics vendors.

[GIJew]

I guess Nvidia will have to buy the blueprints/rights to GCN much like non-reference GPU makers do.

 

At the moment I'm a little pensive on Mantle. it hasn't exactly blown me away.

 

If however we could get AMD and Nvidia to "work together" on something that would be awesome

[ACatWithThumbs]

You forgot Intel and its HD graphics. There's two graphics vendors.

Try to run Battlefield 4 on HD graphics.