The last two days a sort of interesting discussion on a forum unspun about the uses and performance gains of the just newly released NVidia GeForce GTX 1080. The starting point was someone looking to replace his four year old card and he was quite generically asking about the future-proof-ness of the 1080 without even knowing what VR solution he was going to get. In my usual charming way I told him not to jerk off on hardware. Then a few posts later the subject of GPU renderers came up and someone linked this article bemoaning the lack of double-precision features and other stuff in NVidia‘s latest consumer card. Now that’s all fine and well, but I think it is time to look at this for what it is. Let me summarize and elaborate some of my thoughts.
First, most obviously current GPU-accelerated renderers like Octane, Red Shift, VRay RT and so on work just fine even without these fancies and on older hardware as well. They will (have to) continue to do so for quite a while for the simple reason that not everyone will immediately throw out his just recently bought GTX 980, Titan Black or whatever. Of course this will complicate matters for developers, but when has it not? All this stuff is in a constant state of flux, anyways, and a lot of time is spent on sorting out those compatibility issues already. In the end, many of the reasons I’m not delirious with joy whenever stuff gets more dependent on GPU features are the bugs and quirks associated with this. I appreciate having fast rendering as much as the next guy, but as long as everyone just greedily usurps your GPU resources this will not change. You can have as much hardware power as you want, but the sad truth is that even today it’s a struggle to even run two 3D programs side by side because they struggle for OpenGL resources, not to speak of things like CUDA or OpenCL.
Second on the list would be the question, whether there are actually any benefits to you as a user. A programmer might think it fantastic not having to employ tricks like remapping value ranges and memory areas (and spend extra clock cycles on it), but as a user, why should you even care? It’s under-the-hood stuff you won’t even notice. Not saying that there couldn’t be benefits, but with regards to GPU renderers this is a somewhat ambiguous thing. Things like specific complex shaders might gain speed, but other stuff like geometry processing might not. And then of course the usual “if”s apply like scene size and complexity. For games this would be a bit different since things like double-precision could improve clipping calculations, collision precision for dynamics and so on. however, even so, the engines would have to support it first, which is why at this point I find any complaints about the absence of such features are misplaced. It wil ltake time before developers actualyl begin to implement it.
Which brings us near a point: People that have a genuine need for this stuff clearly have the wits and money to go with a “proper” Tesla card to use these features. Ultimately in my view this has a lot to do with people’s expectations. Everyone is wanting a free lunch because conveniently they have gotten used to GPU rendering and now the world is coming to an end when NVidia is not giving them their toys, they get upset. Does anyone even remember that once CUDA only worked on expensive professional Quadro cards? If it wasn’t for some hackers back then that patched BIOSes to coerce consumer cards into pretending they were Quadros and enabling these features, none of this might actually have happened. NVidia only reluctantly opened up this stuff. People these days just take it for granted.
Finally, to come back to that VR thing: I’m sure a 1080 would be fantastic to support whatever is currently on the market, but a year down the road? I’m not too sure. Currently nothing is standardized and it’s not even particularly affordable to buy such a headset. On a technical side, resolutions are “just” HD or slightly above, but 4K displays are already on the way and nobody even knows what requirements they will bring. These are certainly exiting times, but not necessarily in a positive way. This market is in its infancy and what’s hot today will be expensive scrap tomorrow. Your fancy Oculus could quickly become a memento of “good old times”…
Mylenium's Blog 