• OCL Hybrid Render Engine
Yes only this is future - are too many limits with only GPU rendering.
Actually I read a bit about this, if somebody is Interested:
- problem with sharing GPU memory , so if You have two GFX first has 2 Giga second 4 giga then visible will be only 2 Giga from the first (of course speed will be double because two cards)....so must be hybrid system CPU/GPU to avoid it.
- in new GFX cards You have over 2000 CUDA cores (!). You calculate in the same time by 2000 threats ...but is one problem: the amount of data for calculating must be small (exactly perfect for that is Path Tracing code , which is easy to write and very light). Bigger procedures needs more amount of data to calculate - and then starts problems with speed because in some cases increasing amount of cores will bring slower calculations. The same as with intel CPU - the maximum and profitable in the same time amount of virtual cores is 1 . So we have one physical and one virtual. I read: increasing virtual cores (In CPU) will bring slow calculation
- why we have maximum one virtual core....because block of date to execute and calculate are too big and too complicate to calculate. Why must be Hybrid, because in some cases CPU is better then GPU for example calculate of Motion Blur , AA or reflection blur)
- in Octane for example render with full Path Tracing 5 bounce light (this is only MINIMUM to get good renders) take "years" ...even with two High class GFX cards. Why when You see interiors renders from Octane (or similar engines ) they looks quite artificially because to avoid problem with time render they use Direct Light system + OCC (so is not Pure Path Tracing)
- there is also problem with speed - is not that fast as everybody imagine . the first seconds/minutes of render are really fast (Is special trick in code for that to see something, kind of interpolated pixels) but real render comes after many minutes (depends from GFX card) and to see clean render You must wait 2-5 hours....
- unbiased engine renders are fast in basic scenes because is almost nothing to calculate! (for example cars or studio) There is nothing to render: basic models shadows (is not that important amount of polygones), one source of light, OCC . The real render comes in complicate scenes when photons must be calculated in dark corners (for example interiors) with many objects inside. So testing Octane with basic scenes and one source of light is completely nonsense and more nonsense with Direct Light and OCC.
- Unbiased renders engines are very easy to write (everybody can start
) But problems stars when You will try add complicated materials: glass, SSS, reflection blur, Motion blur, DoF , displacement , nodes ,etc
- there are problems also with "hot pixels" randomly especially with reflection/blured surfaces You can notice very bright pixels. This is normal in that kind of engines renders. There are some special modules to reduce it. This coming from not perfect accurate calculations (sic! in Path Tracing even...) ...and this is really big problem
- lags in system during render - in demo Octane for example I could not do anything in system! (firefox, watching video, skype ...even move windows are difficult) - maybe I did something wrong...but lags was very big.
- to render smoothly with Octane for example You need minimum two Titanium GFX card with 6 Giga GPU RAM (each ~ 1000 Euro) and a lot of electricity, one card consume up to 300W! and still You are limited to 6 Giga GPU RAM
If I mistook something or write wrongly please correct me. I can be wrong in some details.
