Mousemonkey, you have mentioned a small sample size and an amateur developer more than once, so we might as well address those things and try to figure out how you think they matter here.
If you could clarify who you are aiming the amateur developer comment at it might help.
If you are talking about Oxide, their team is made of non-amateur developers. Read their bios, as they have worked on quite a few AAA titles and even have one of the Microsoft engineers from the HLSL team. If you're referring to an engineer who helped develop DX APIs as amateur, heaven help the industry.
If on the other hand, you're referring to the person who wrote the test program to try and confirm whether both GPU architectures were indeed running async compute functions, well, what's the point of calling him an amateur if he accomplished his goal? It really doesn't matter his skill set, as the results of his test seem to be able to stand on their own. Does it matter if an amateur baseball player or a pro ball player makes a home run? It could, but if we're just talking about the home run as an end in itself, not really. It's more of a red herring at that point.
Saying there is a sample set of one is inaccurate on your part, if referring to NVidia's apparent lack of asynchronous compute. There is a sample set of 2 pointing toward that conclusion. Not a whole order better, but neither is saying the sample set is too small to draw conclusions from.
The AotS "benchmark" is perfectly fine as a real world example of just how the two different GPU architectures will behave running the Nitrous engine, and also how different CPU architectures behave, whether it's a sample size of one, or not. It's not as though you can play the AotS RTS game on another game engine. People looking to see how that game behaves under certain conditions will never get accurate indications by benchmarking other titles. Each game engine will have unique performance characteristics under DX 12, especially when more of it's feature set is tapped into. It's clear that both GPU vendors have focused on differing feature level support so we will have to wait and see what software developers choose to implement, and the ramifications of those choices.
Not sure the relevance of your last comment, about running a GPU compute program on both AMD and NVidia cards. Nobody is saying either GPU can't run compute in parallel, they are simply saying, only one of those GPUs can run compute in parallel with graphics without a large performance hit. In the end, whether NVidia is far faster at compute may end up mattering very little, if you can't use the idle resources on your NVidia GPU because a graphics operation is taking place.
Facebook
Twitter
Instagram