The blog states that if the Mantle specification is adopted by other manufacturers, particularly on mobile, then it will become the new cross platform standard and OpenGL will cease to be relevant. 67% of native speed is astronomically slow in the world of real-time games where improving per-frame performance by 3ms is considered a huge win. Getting all of your nice special effects and post-processing implemented while still running at a smooth 60fps is very difficult - especially on mobile.
I see, thanks. Yes, if that happens it could certainly upset OpenGL's position. What, though, is the likelihood of that happening? Have there been signs of cross-vendor Mantle adoption, and of support by the OS vendors Google and Apple?
> Getting all of your nice special effects and post-processing implemented while still running at a smooth 60fps is very difficult
And has very little to do with CPU speed. There are some games where CPU speed is important but there's a large subset (I'd be willing to bet 95%) of AAA games where only GPU speed is important and they run just fine on low end CPUs.
The whole point of D3D12 and Mantle is to decrease driver overhead, improve multi-threading support, and increase the number of per-frame draw calls that can be made. Those are all CPU performance improvements so I think it's safe to say that CPU speed is quite important or else AMD and Microsoft wouldn't be going through the trouble.
There is an reason why review sites use games for CPU benchmarks. As you can see some games are indeed GPU bound. But vast majority is actually bit of both, e.g some section of it is purely CPU bound while another is purely GPU bound. So improving either component will increase FPS.
My experience with low-latency programming suggests that often the biggest cost is memory management -- garbage collection. In some cases, the GC compacter may even be limited by CPU, although usually its the collection pauses themselves which hurt the most.
