AMD have always been first to market with features:
1. 2Gb Vram. My 5870 had it, which is why I bought it over the GTX 470.
2. Eyefinity, first iteration of proper multi monitor support.
3. DX 10.1 again AMD first.
4. Vulkan in the guise of Mantle developed by AMD.
5. DX12 feature sets more support by AMD than Nvidia.
6. Now we have HBCC and FP16 hardware support.
While there is no denying AMD often do add features that will be useful in the future, or have a forward looking architecture, as explained this often a waste of time because by the time those features are useful everything has moved on 2 generations and then AMd never seems to catch up w once the tech is utilized. For example, ATI introduced tessellation long before it became a DX requirement, it was basically never used. Then when tessellation was addes to the DX specs Nvidia produced a very good implementation while AMD struggled to get a performant version out.
If we look at your list:
1) AMD cards have mostly required additional bandwidth over Nvidia cards, because NVidia GPUs have more advanced BW saving techniques while AMD relies on brute force. One way to get additional bandwidth is to have a wider pipe with more chips, which leads naturally to higher VRAM amounts. this adds costs. This is only ever important if you have games that max out VRAm, which just doesn't really happen that often, and when it does it often all taxes the GPU heavily. And here we are today with Nvidia offering 11/12GB VRAm for their top GPUS that might come close to using that amount at 4K/SLI/ultra settings. AMD's best consumer offering is at 8GB. As some balance, I don;t think nvidia 1060 3GB models are a great buy if you wanted yo hold on to the card for a long time, the 6GB models or Polaris are btter. But for a year or 2 then the 3GB is liekly fine.
4). Nvidia had Vulkan drivers certified before AM. Nvidia had a load of developer support, example open-soruce code, documentation, etc. all online a long time before AMD did. And the last time I checked Nvidia were still well ahead with Vulkan developer support.
5). Nvidia hass offered high DX12 feature support in Pascal and Maxwell than the AMD competition. It is only with Vega has AMD caught up and taken over. In 6 months Volta will liekly be at the same level
6). HBCC is designed for HPC and not gaming. It can onyl really help in gaming when the GPU runs out of VRAM, which doesn't typically happen. AND then see your point 1.
As for Fp16, actually Nvidia had FP16 support way back on the FX 5 series cards, back when the lower precision was really important to get the msot performance out of the new pixel shader technology. Fp16 support was dropped because GPUs were powerful enough not to need to drop form FP32, and in fact the DX spec require F32 as a minimum. FP16 has only resurfaced now because it is useful for deep learning, which is why Nvidia added CFP16 support for Pascal Gp100 released 16 months ago. Since AMD hae yet to differentiate their compute and gaming cards, vega happens to also have FP16 support for gaming. There is some limited use cases where FP16 is sufficient and will lead to minor performance increase in specific scenarios. this isn;t some game changing feature, and it certainly isnlt some highly novel ground breaking technology that AMD are the first to develop. Nvidia sold GPUs with FP16 support 15 years ago.