AMD at SIGGRAPH 2017 revealed their new efficient feature called the Rapid Packed Math, which basically halves the floating point calculations in processing a data request, resulting in a quicker distribution of information.
AMD and Ubisoft revealed that Wolfenstein 2: The New Colossus & Far Cry 5 will support Rapid Packed Math, this feature as of now is exclusive to the RX Vega series of GPUs. According to the developers at AMD, RPM will process two math instructions for the time consumed for one.
This means that both the games will perform much faster floating point 16 FP16 computing than with the standard Floating point 32bit computing. Rapid Packed Math is currently an exclusive AMD feature and NVIDIA has made no announcements what so ever to support it. Since it’s a hardware feature, owners of current NVIDIA GPUs won’t benefit from it.
Certain enthusiasts also criticized the idea portrayed by AMD for using 16 bit FP compute against the FP32. Based on the discussions from Neogaf forum, FP16 is a lower quality calculation being used by the GPU, directly affecting the image quality being produced, except when it’s used in those areas of the game where object detail would not matter and will be less noticeable to the eyes, may result in certain level of performance gain without a visual quality loss. The extent of which is yet to be seen.
Apart from the fact that this is a hardware feature and without most of the game developers supporting this feature, it will become pointless. You may Google for 16 bit precision VS 32 bit precision to check image quality differences.
AMD also claims that it’s the complete support for MS DX12 API. The fact still remains that if the developers are willing to take advantage of all the DX12 features, and will they prove good enough to complete with NVIDIA GPUs. Considering the fact that DX12 and VULCAN are still struggling to get mainstream attention from game developers.
