it would have been better to say the cpu has been the master and the gpu has been the slave.. never before more untrue with next gen api's. your cpu and gpu are more like equal partners than ever before. even if nvidia is able to fully optimize overhead at levels of amd its just not enough..
what are you going to buy if you have heavy gpu accelerated workflow? well at the top is probably going to be a complete intel xeon setup followed by quadro and firepro.
amd gaming gpu's hold a spot over nv gaming gpu's for being able to do both except pushing cuda has made sales.
I think the amount of overhead you think exists is vastly overstated. Also, we've seen what Mantle can do and while it's good, it's not amazing, and we're bound to see the same thing in DX12 with nVidia, something AMD already has with Mantle. Also GPU and CPU can only be truly equal partners if you're using an APU with something like HSA where the two cores are practically physically next to each other. You can't expect external hardware like a GPU on a PCI-E bus to respond quickly enough to general purpose compute instructions on the fly which is why most GPGPU applications are compiled then sent to the GPU. They're not good at doing things on the fly, that's what a CPU is for. Also I've made this point time and time again, more threads doesn't always mean more performance. Your workload has to already be sufficiently functional (lacking side effects,) to do such a thing and have it be helpful. Otherwise you run into the race condition and deadlock problems which are solved using methods that may ultimately reduce performance because the bit of your application that you're trying to make multi-threaded may be inherently serial in nature, in which case threads don't do anything for you.
All in all, I think some more waiting is in order because there are a number of things going on here.
- Speculation about unreleased and unreviewed hardware.
- Speculation on an API that isn't used by any hardware yet.
- False assumptions about how computers work.
With that said, I think closing this thread and waiting for some reviews to come out would be a wise idea because I see more bias and false assumptions than what should be acceptable in a thread. On top of that, several knowledgable members (in particular,
@BiggieShady ) have given their input to explain exactly what's going on here but yet you maintain your claims seemingly based on faith and not information considering you have no rebuttal other than "OMG AMD, DX12, and Win10, rah, rah rah," which isn't constructive or helpful considering no piece of technology will make your code run in parallel for you and that you're still constrained by reality when doing such things.
so your just onto trolling now?? must have actually read and thought about what dx12 means since thats all you have?
Calling you out on the bias and BS doesn't make him a troll.