Speed comparison vs CPU

[Alex]

Now before you all say GPU is way way way faster, hear me out .

I have done my fair share with mentalray in 3ds max and those renders take very long for a nice architectural scene (lets limit this discussion to stills). Now from what I have seen from octane I'm impressed but most of the images in the gallery say render times around 2-6 hours while a mental ray scene takes around an hour. I have to admit that the initial feeling and visual feedback you get almost instantly is very nice in octane but is it really faster for a fully settled image? I can imagine just purchasing octane for workflow reasons but I'm unsure about fully finished renders.

Can anyone with more experience than me please clarify on this?

Regards,

Alex

[Reggie]

Mental ray is trash (even nVidia agrees). Try the Octane demo and make your own choice. I could go on for days about how awful MR is, but everyone has their own needs.

[pixelrush]

How long it takes to get a clean image depends on a number of things.
The number of cores/gpu, the size of image, the type of materials and lighting/kernel etc.
Some people would be producing a decent image for their purposes in as little as 5 mins but of course others will take longer and some people are obsessive about wringing every last unresolved pixel out of one.
I usually dont go over about half an hour but my scenes and hardware are relatively simple. GPU is certainly faster than CPU for unbiased renders though. Definitely wouldn't want to go back.
Probably the well cooked renders you are referring to are for arch viz clients who expect and are willing to pay for best quality.
Those renders would also take a fair time to set up with a full compliment of textures, very neat unwraps and considerable attention to detail in the scene, lighting and so forth. Honestly I haven't yet heard anyone complain that Octane doesn't perform adequately in comparison.
Just having the ability to navigate around the scene and tweak things until you are happy is worth the price of admission for some and frankly Octane is very good value for what it is capable of.
How about trying the demo for yourself and satisfy your doubts.

[matej]

In my experience a CPU renderer with photon-mapping (or similar optimized algorithms) is currently a faster option for 'interiors' (in general: closed scenes with obstructed paths to light sources) than Octane pathtracing. Of course Octane will produce a more realistic result, but you will have to wait a while for it (and have some good hardware). When (and if) Octane will use optimized algorithms (like bidirectional pathtracing) this may change.

Another story are 'exteriors' or 'product-viz', ie. scenes where paths to light sources are not obstructed by walls, small openings etc. In this scenario Octane is very fast and worth every penny.

Stil, for example for a decent resolution (6 megapixels & above) with a complex 'exterior' scene (lots of complex materials & geometry) I have to spend about 3 - 5 hours of rendering for a scene, on my GTX 460. To make the render perfect some denoising in GIMP is still needed afterwards. (for comparision; a similar scene with non-fake DOF would take 3x - 5x more time with a CPU renderer on my i5 2500K, which is comparable in price to my GPU, so Octane is still faster & better)

EDIT: And yes, the interactive workflow with Octane is worth it's price alone.

[Alex]

Thank you all for your replies. What made me wonder was this topic http://www.refractivesoftware.com/forum ... f=21&t=996

I know this is unbiased and that is more realistic. However as long as the end result looks the same it's a fair comparison right? Also, is there a future plan from the octane guys to incorporate the denoising in the pipeline? An option to apply it to the image before saving it?

As for animation support (which is most of my work), I have seen the maya plugin video and I was wondering if it will all be supported in the engine itself or will it remain completely restarting octane for every frame?

I do think I'll buy octane because basically it looks like fun and sorry for going a bit OT with my last question.

[glimpse]

well, it's not wise to compare GPU renderrenr like Octane to Let's say Vray or any other that basicly fakes out the result using aproximations. It's unbiased render so have to compare it to let's say Maxwell Render, that works on the same unbiased mode, only for CPU.

To elevate the comparison CPU versus GPU let's make a small scene to compare. Because Octane doesn't have CPU vertion let's take let's say Vray and Compare speeds for RT both in CPU & GPU modes. Simple Cube with two lights, DOF..Let's hit 'Make It Nice' button - a.k.a. RENDER.

For the record my system QX6700 (quad core runing 2.66GHZ) & two 460GTXs.

need any explanation? =)

You can get the same result from GPU in 30sec, when You have to render for almost 10min on CPU mode. So if You Take those 10 min and spend them for GPU rendering, the result is much more cleaner than CPU.

But That's only one aspect. What is much more interesting for me is EXPANDABILITY.

Let's say I want to upgrade the system. How much it is going to cost me to cut render times to 1/4th of what I have now?

If I go for CPU route I Have to change everything (cpu, motherboard, ram ) that is costly - that Would probably cost me no less that 2,5k$. BUT if I take GPU route I just found a guy selling two GTX590 for 500$ each. So coupling these with new PSU I would probably spend like 1,2k$..or in other words just half =)

Having in mind that my curent GPUs do the work faster than CPUs like 20times..if I would ivest into upgrade it's clear where to go, at least for me.

So in this case, when You Do comparison, compare apples to apples, not apples to oranges =)
(but always keep in mind the trade-offs, like vRam limitations and other things, like power, heat, etc =)

Hope it helps =)