AMD Reportedly Plans To Bring Mantle To Linux, Calls Mantle An Open-Source API

[CrystalGamma]

The point is, bindless graphics eliminates the need for binding which reduces the overhead and cache inefficiency, and therefore eliminates the need for Mantle all together. Using bindless graphics, the shader can access millions of texture and buffers directly through pointers. This allows the GPU itself to do efficient cache usage and data fetching, instead of this being controlled by the driver. It should be obvious to any programmer that removing the need for CPU interaction will boost performance.
BTW, I researched this stuff in my thesis.

If you are familiar with how shader programs revolutionized computer graphics, you know it was by moving computational complexity from the CPU to programming the GPU directly, and reducing the need for API calls. Bindless graphics is taking this a step further. The final goal is having a pipeline completely implemented in a shader language, controlling the execution of the pipeline stages from the GPU. We are still not there yet, but Maxwell and probably AMD's new architecture will take small steps in that direction.

What makes you think Mantle will not allow for similar functionality as OGL bindless? Are you breaking some kind of NDA with AMD by telling us this? Or are you just pulling this out of your ass?

[Kraut]

I love how you guys are talking all this about mantle yet not even the specification has been released to the public. You're literally basing this off of graphics AMD has released which mean absolutely nothing. I don't know about you but I don't like to speculate on unreleased and unproven technologies.

There did published infos on the principle framework. For example: http://www.amd.com/Documents/Mantle_White_Paper.pdf (Interessting stuff starts at page 5-6)

Also they have working drivers and a few games running on it.

[efikkan]

Mantle also reduces API calls.

If a company were making a cross platform title with only OGL the work to optimize for both AMD, and Nvidia, and PS4, And XB1, and Linux would be about the same as making a cross platform title for Mantle/Nvidia OGL/PS4/XB1/Linux. That way they only need to worry about Nvidia performance on OGL.

Besides MS is bringing DX12 to XB1. And Mantle porting to XB1(DX12) will be painless. So for only a bit more work a company will get great AMD support, Linux support (eventually) and AMD Steambox support (eventually)

http://wccftech.com/mantle-api-ported-directx-12-amd-white-paper/

This might lead to NV supporting Mantle because it will be available on Linux/Steambox and porting Mantle to DX12 will be practically effortless. This would give devs 1 main API Mantle/DX12 to support everything at the highest performance levels.

Mantle>PS4 is also a pretty easy port. On top of that Mantle might come to both PS4 and XB1. Mantle wouldn't be dissapearing anytime soon.
http://www.dualshockers.com/2014/03/...-both-further/

With Direct3D 12 arriving and OpenGL already supporting more low overhead optimizations than Mantle, why would lots of game developers spend lots of resources creating a new rendering pipeline for Mantle when it's easier to adopt the new features of OpenGL or Direct3D? Implementing the driver support is not too hard, but redesigning game engines requires a lot of effort. (Remember poorly done ports perform poorly)

Mantle only offers marginal performance gains for high-end CPUs, so there are few reasons to support Mantle except the free advertisement from AMD.

Don't expect support from Nvidia anytime soon, unless Mantle seriously take of... (I doubt it)

[jimbohale]

There did published infos on the principle framework. For example: http://www.amd.com/Documents/Mantle_White_Paper.pdf (Interessting stuff starts at page 5-6)

Also they have working drivers and a few games running on it.

That link means absolutely nothing. I don't trust AMD to release a working piece of software, and I highly doubt that Mantle will wind up even working if it is released. They are clearly incompetent (the management responsible for dealing with Linux), and assuming otherwise is foolish.

[Filiprino]

Mantle only offers marginal performance gains for high-end CPUs, so there are few reasons to support Mantle except the free advertisement from AMD.

That's the biggest problem against Mantle. Users who play games usually have a higher performant CPU, and Mantle doesn't bring much to the table when you're using high performant CPUs. But probably it would bring benefits to the power consumption.

[glisse]

With Direct3D 12 arriving and OpenGL already supporting more low overhead optimizations than Mantle, why would lots of game developers spend lots of resources creating a new rendering pipeline for Mantle when it's easier to adopt the new features of OpenGL or Direct3D? Implementing the driver support is not too hard, but redesigning game engines requires a lot of effort. (Remember poorly done ports perform poorly)

What does support your claim that current OpenGL through clever use has lower over head than Mantle ? Is there any data showing that ?

What you fail to see is that Mantle not only allow low overhead, but more importantly it reduce the number of entry point drastically in comparison to OpenGL (see http://stackoverflow.com/questions/2...eady-available for list of currently exported functions). This is the most compelling point for Mantle. In OpenGL there can be more then 20 ways to do the same thing, some with low overhead, some faster than other, but it all depends on the driver, ie on some driver one combination will be the fastest and on another it can an entirely different combination. This is the most problematic aspect of OpenGL the size of the API and the lack of consistency accross different implementation.

If you want a single example to prove how bad the number of entry point is for OpenGL just look at OpenGLES, this is the reason why no embedded GPU vendor is eagger to make a full GL driver even if their hardware has the capabilities. The OpenGL API is just too big.

[profoundWHALE]

DMAs are great for transporting information around
CPUs are great for serial workloads/glue code
GPUs are great for massively parallel workloads.

FTFY.

Thanks.

I love how you guys are talking all this about mantle yet not even the specification has been released to the public. You're literally basing this off of graphics AMD has released which mean absolutely nothing. I don't know about you but I don't like to speculate on unreleased and unproven technologies.

Take a look at this: http://www.guru3d.com/articles_pages...review,19.html



And there are other tests with people who have beefy CPUs where there is very little effect if any.
I say that in the first case the CPU and the GPU are very close as far as bottlenecks go, but then when you add that 290X, it's the CPU that's bottlenecking.

[efikkan]

That's the biggest problem against Mantle. Users who play games usually have a higher performant CPU, and Mantle doesn't bring much to the table when you're using high performant CPUs. But probably it would bring benefits to the power consumption.

Yes, most people with a decent desktop computer have an i5 with 3+ GHz, and Mantle is not going to give a huge difference here. And when games utilize DirectX 12 or new OpenGL features the benefits of Mantle will disappear.

BTW, the power consumption will actually slightly increase, since the GPU spend less cycles idling.

[grndzro]

Yes, most people with a decent desktop computer have an i5 with 3+ GHz, and Mantle is not going to give a huge difference here. And when games utilize DirectX 12 or new OpenGL features the benefits of Mantle will disappear.

BTW, the power consumption will actually slightly increase, since the GPU spend less cycles idling.

Mantle porting to DX12 will be very easy. The benefits of supporting Linux/Steambox probably outweigh the minimal porting effort of going Mantle/DX12. On top of that Mantle will work with Win7 which has the highest market share of windows.

Do you honestly think supporting OS locked DX12/OGL is a better deal than supporting Mantle/OGL? You are crazy.

[Filiprino]

Yes, most people with a decent desktop computer have an i5 with 3+ GHz, and Mantle is not going to give a huge difference here. And when games utilize DirectX 12 or new OpenGL features the benefits of Mantle will disappear.

BTW, the power consumption will actually slightly increase, since the GPU spend less cycles idling.

Nah, the power consumption would be lower because you'd spend less CPU cycles and the GPU would be used when needed for the same performance.

And now that I think about it, if Mantle is low overhead then it may free up resource from the CPU to run other type of tasks: physics, input/output, random generation algorithms, sound processing and others.