Developers that aren't being paid by AMD aren't saying much at all!
I am a developer (not games) and I can tell the last thing anyone is bespoke platform specific APIs and are much more willing to trade a small CPU overhead for a simpler API.
yes, developers do want a graphics API with lower CPU overhead, but they don't want the industry jumping back 20 years. As graphics cards get more and more complex games can have more and more complex graphics but this takes more and more time to develop so there is less and less desire for multiple APIS. There was much hatred 20 years ago, that is only worse now.
Besides which mantle doesn't resolve the fundamental issue in any unique way that would make a paradigm shift. yes, draw calls are slow but that is mostly due to hardware limitations. Any time there is data or instructions going from the CPU and main memory to the graphics card and GPU memory then there is a time cost. Mantle doesn't resolve that because the problem is a hardware issue and even a new CPU interface BUS wouldn't change the fact that system memory is slower than GPU memory. Therefore things like mega textures, texture atlases, instancing etc. are still required even with Mantle
Furthermore, with the move to 4K screens GPUS become very pixel shader limited so CPU constraints matter less.
What is actually needed and coming due to the increased capabilities of GPGPU and increased memory sizes is to transfer more load from the CPU to the GPU so the drawcalls don't emanate from the CPU but the GPU issues them itself. Things like scene graphs, spatial partitioning trees, frustum culling, occlusion all can be shifted to the GPU so the GPU decided what to render, rather than the CPU deciding what to render and then sending the necessary draw calls. things like Physx go a step towards this direction, e.g. it already allows line of site and occlusion checking in the GPU. Nvidia has already provided OGL extension for efficient occlusion
http://http.developer.nvidia.com/GPUGems/gpugems_ch29.html
This is developer with strong AMD ties (can't remember if they work for AMD or gets spnsership) but the website is stuffed with the future of graphics programming:
http://rastergrid.com/blog/2010/10/opengl-4-0-mountains-demo-released/
Hierarchical-Z map based occlusion culling
http://rastergrid.com/blog/2010/10/hierarchical-z-map-based-occlusion-culling/
Instance culling using geometry shaders
http://rastergrid.com/blog/2010/02/instance-culling-using-geometry-shaders/
All of that stuff is where the developers are going and makes a far bigger difference than the Mantle API
This is the current state of most games
Code:
// Mantle/OGl/DX running on CPU
if ( expensiveCullingAndVisibilityTest(ScaryMonster) == VISIBLE)
{
// Send drawcall to GPU, if texture/geometry not loaded then do very expensive uploads.
sendDrawCall2GPU(ScaryMonster)...
}
// Mantle makes the draw call a little faster, but there is still a draw call which is inherently slow and we had to use a slow culling algorithm
This is where most games are heading.
Code:
// OGL/DX running on the GPU
// The culling is done on the GPU, and these algorithms are inherently parallel so suit GPGPU perfectly, takes load of CPU
if ( fastCullingAndVisibilityTest(ScaryMonster) == VISIBLE)
{
// No need to send anything to the GPU, it is already there, so no draw calls required. We have sufficient memory nt to need expensive data uploads.
draw(ScaryMonster)...
}
// CPU did nothing to render the scene, the GPU did everything, CPU concentrated on AI (with help form the GPU using physX)[/SPOILER]
