Here are the first technical slides about Mantle, the low level API for Radeon cards on PC. According to AMD, it will be possible to have up to 100K render calls per frames with Mantle, while today, optimized render codes achieve only 10k draw calls/frame.
Mantle is a low and high level API: some parts of Mantle are lower than OpenGL and Direct3D while other parts are similar to what OpenGL/D3D offer. Mantle will bring a better graphics memory management and better management of the render command lists. Nice consequence for multi-GPU setups (CrossFire): the developer will be able to decide which GPU will process the render command lists: asymmetric rendering, beyond AFR. For example one GPU for scene rendering, a second GPU for post processing. AMD introduced the concept of monolithic pipeline: it’s a big rendering context that defines all render states, all GPU programs. Mantle simplifies the resource management: there will be only memory zones and rendering zones (images).
The first public documentation will be published during the next GDC 2014 and the first final version of Mantle will be available not before Q3 2014.
Another interesting fact is that Mantle is not limited to AMD Radeon GCN architecture. The core of Mantle could be ported to other GPUs (NVIDIA for example, some NVIDIA GPU designers were present at the AMD Mantle talk #APU13).
Update (2013.11.14): AMD Mantle Technical Deep Dive – APU13 Event
AMDs Mantle will blow everything that Intel and nVidia have… The wait is almost over.
Looks like AMD PR disease continuous from Bulldozer. Just a few slides and lunatics like Rares coming from caves. Prepare for R9 290X 100k TDP and 100k °C.
@anon: give me a brake!
if it may be use in AMD APU? If yes, I think that it will be a powerful performance. Unified memory(hUMA) for GPU and CPU with Mantle API and additionaly AMD Steamroller. If somebody will want yet a more performance, may add Radeon X8 or thing different 😉
AMD are desperate to make a proper OpenGL drivers so they made Mantle. I wonder if they will be able to write proper Mantle drivers…
I’m sure Mantle would remove some of the overhead and enable more draw calls per frame, but this is still an “old fashion” approach to rendering.
Some of you might have notices Nvidia has increased the programming abilities of their GPUs over the last three generations including some of the following achievements: direct pointer access to vbos and textures (removes the binding), allowing general pointer operations and data manipulation, allowing GPU thread spawning and allowing streaming of data from PCIe devices. This is achieved through OpenGL extensions in conjunction with CUDA. CUDA 6 was announced today featuring unified memory. Back on the GT200 GPUs Nvidia achieved an impressive 7.5x vbo throughput. For those who hasn’t realised it yet, this might soon allow rendering (including animations) with very few calls from the CPU at all.
It’s about time graphics rendering moves behond the CPU pushing every instruction to the GPU in a dumb fashion. While even the current low end GPU’s are perfectly capable of doing much more if only they wouldn’t be constrained by the number of draw calls. It really needs to be replaced by a more modern approach. Where the CPU first copies the whole GPU program tot the GPU and then tells what parts to run and sync and so on.
“C has too much of an overhead. so let’s start again writing code in assembly, because that’s the way to improving performance, not get better cpus”.
this is simply kick-ass! go Mantle go!!
it was about time for something like this to happen anyway