PCI Express 3.0

[GeorgoSK]

I just wondered...with upcoming 3.0 version, with double bandwich of previous one, would it still be too big of a bottleneck to use mainboard memory ? Just asking

[ROUBAL]

Even if it was enough fast (and I don't think it is), only people with brand new motherboards could use Octane, so it would not be a good thing imho.

[GeorgoSK]

Even if it was enough fast (and I don't think it is), only people with brand new motherboards could use Octane, so it would not be a good thing imho.

Why did you assume that whole code would had to be rewritten so that it would exclude older mainboards... ? It could just be an option for those with newer ones.

And even if it would be slower, it could still be a good option for occasional use on bigger scenes. It's still in the stars whether middle-class gpus with higher vrams will appear on the market (it might never happen, this is super small market ! ), and we shouldn't fool ourselves that everything can be optimized under 1,5 gb of memory.

[matej]

Gaming videocards with 2GB appeared this year - even without real need for them (I doubt any games right now take advantage of it). So, I wouldn't worry - the market will always produce "unnecessary" stuff for those who want to make their neighbors jealous. In this case we can take advantage of it.

As I read around, copying stuff from vram to system ram in CUDA apps. is the major bottleneck in performance. Frequent copying doesn't just slow down things, it kills every bit of possible performance gain. Refractive might aswell implement Octane on CPU...

[GeorgoSK]

Well, let's then hope 580 will appear sooner than later

[GeorgoSK]

Reason for my neverending brag about short-comming of small GPU's memory is I want to be able to use Octane all the time, right now I still have to use Mental Ray for bigger projects, and I already hate to do that since I love everything the way Octane does it.
Right now there is no solution for that, optimisation of scene can only do so much. I am well aware proxies will help tremendeously, so will if these higher vram GPUs will appear, but there is still time in meantime to speculate about alternative/other solutions Don't dismiss that, such thinking is unhelpful

[matej]

I know what you mean, it's just that Octane is young and beta. Memory management features and optimizations are yet to come. As will new material possibilities that could help save texture memory (new procedurals, mapping options, etc..)

I don't doubt that Refractive is aware that most Octane users don't have teslas & quadros (and probably never will).

[DayVids]

We probably won't ever have that, but there were some ideas thrown out there, that Octane may look into offering Tesla high VRAM rendering for rental fee's to accommodate these sorts of needs.

We'll have to see what comes of that as well, as it could be very worth it.