AMD revealed some slides that show details regarding the Mantle API itself and the way it reduces API overhead and allow developers to use the full potential of the GCN-based GPUs.

According to AMD, Mantle API will allow developers to optimize performance by distributing threads on the CPU and GPU, thus using both much more efficiently. The optimization can be further improved with HSA implementation which means that Mantle API might work quite well on the upcoming AMD APUs.

According to AMD, Mantle API can use multiple CPU cores much better and more efficiently than both the DirectX or OpenGL and things even get better when it comes to multi-GPU support. It appears that AMD Mantle API allows multiple GCN-based GPUs to be "treated" as a single GPU, thus improving load balancing and also eliminating micro-stuttering, which was quite a thorn in AMD's side.

According to AMD, Mantle API should bring anywhere between 20 to 50 percent boost in performance, which does sound quite impressive although we are quite sure that the gain will be much closer to 20 than 50 percent.

AMD's Mantle API will also have a big problem with adoption, as currently, DICE is the only big developer behind it and although AMD has more names to add to the list, these are coming later in 2014. While it is currently only supported on AMD GCN-based GPUs on Windows, DICE's Johan Andersson noted during the AMD APU13 conference that while it is a shortcoming, Johan noted that he would like to see the Mantle API on Linux and OS X, while pairing up the Mantle API with Valve's SteamOS would make a "powerful combination".

While currently limited to AMD GCN-based GPUs, Johan noted that Mantle API can work on most modern GPUs out today. We presume that he is talking about Nvidia GPUs as well, unless he does not count Nvidia ones as modern GPUs.



Source: via Hothardware.com.