Gpu and cpu rendering

[mitviz]

How does both stack up against each other when it comes to rendering in vray, i havent really used RT to know enough but is it wise to invest in more powerful graphics cards for full gpu rendering or more cores for cpu rendering? How does one compare rendering speed of both in anyway? Is it possible to make a rendering time comparison in vray of both? I keep hearing gpu is the future because of the price n power etc n now i know of atleast two people using iray but i have limited knoweledge here, can anyone shed some light here. Can you say compare a machine as mine i7 5960, running at 4 ghz n 32 gigs of ram to a graphics card of similar rendering power? Just need some clarity

[super gnu]

as a very rough ballpark figure, a titan X gpu would be equivalent to 3-4 of your machine (its roughly 4x my 6 core at 4.6 ghz...)

it does depend on the scene though. however architectural scenes, esp. interiors, would be in this ballpark.


-however, you have to consider the following:

* only a subset of features are supported,, basically only bitmaps in the bump slot (with a couple of exceptions) and lots of plugins not supported. render elements can be a bit problematic.

* as well as being limited to the ram of your gpu ( 12 gb for the titan x) you have to consider, it also uses more *main ram* to render. i have a scene that renders in 17 gb using normal vray, but with rt gpu, it uses 31gb of system ram and 9 gb of my gpu ram.

this can bit you on the arse quite easily.

* even things that are supposedly supported ( forestpack, vraynormalmap, multimatte) are not fully supported.. and esxpecially in the case of forestpack, are unusable for larger scenes.

* displacement/fur/proxies are computed *in full* before rendering, and loaded onto the gpu. this means its extremely easy to max out your gpu ram with things that dont use much ram when using normal vray.


-did i mention its FAST?

[super gnu]

a bit of reading back over benchmark threads and the millions of rt gpu posts here would be a good idea, rather than asking people to repeat it all again

[mitviz]

a bit of reading back over benchmark threads and the millions of rt gpu posts here would be a good idea, rather than asking people to repeat it all again

Thanks for the info, i guess gpu isnt such a great idea based on what you said but i wonder if vlado has plans to change all the unsupported features. Also i kno this forum probably has alot of info on the topic but who knows what changed since those posts,better to ask for myself anew

[savage309]

Bump will be supported with other textures very soon. Displacement, subdivision, etc, are calculated based on how far is the object, optimizing the memory usage using camera dependent tessellation. We have fixed a couple of bugs with the VRayNormalMap as well, so things should get better.
Please report those to us, just to be sure that we have not missed anything.

[glorybound]

I have tried using it several times on my 8GB card, however, it took longer to load the scene then it took a LC pass to complete. Basically, I was trying to use it to set up lighting; it wasn't real time, which is what I was expecting. When it finally did load, I would have to hit render again to see the scene update. Having said that, once in a while it work in real time like it is supposed to, but it was not predictable.

[savage309]

Did you use it as Active Shade or as a Production Render ?

[glorybound]

Did you use it as Active Shade or as a Production Render ?

I tried both. It crashes in production and it doesn't live update in active shade.

[savage309]

This is definitely not how I think it works.
If you happen to have a scene that can show these problems I would very much like to look at it (mailto blagovest.taskov@chaosgroup.com).
Thanks !
Best,
Blago.

[glorybound]

I remember doing some work a year ago and my render times where 1-2 hours and the person I was doing work for was doing them in 20 minutes, with Octane. I'm sure a lot of things were not supported, but I was hoping that's what I would have experienced in RT. I was going to purchase a Titan X, however, I didn't because I feared I would not be able to use RT, anyway. I went from a 4GB card to an 8GB and it didn't make a difference with RT, it still failed and was not predictable.

[savage309]

I remember doing some work a year ago and my render times where 1-2 hours and the person I was doing work for was doing them in 20 minutes, with Octane. I'm sure a lot of things were not supported, but I was hoping that's what I would have experienced in RT. I was going to purchase a Titan X, however, I didn't because I feared I would not be able to use RT, anyway. I went from a 4GB card to an 8GB and it didn't make a difference with RT, it still failed and was not predictable.

I see.
If you happen to have any scene that shows a particular problem you can send it over and we will fix it
Doing a comparison between render engines running on different hardware (CPU vs GPU in your case) is like doing a benchmark on two different types of fuel, using two completely different cars (lets say 500HP vs 150HP) and measuring how fast each of them will go. It doesn't say much.

[glorybound]

I understand. If we turn off everything that makes V-Ray special, I might be seeing 20 minute RT times.

I see.
If you happen to have any scene that shows a particular problem you can send it over and we will fix it
Doing a comparison between render engines running on different hardware (CPU vs GPU in your case) is like doing a benchmark on two different types of fuel, using two completely different cars (lets say 500HP vs 150HP) and measuring how fast each of them will go. It doesn't say much.

[mitviz]

So i guess better to stay away from it from now in my case n i know its weired to compare the two but just wanted to know what say a certain type of card would compare to the 5960 interms of rendertime but if its so much trouble maybe best to forget about the gpu stuff for now, in my case that is until its at a point where its worth it

[savage309]

So i guess better to stay away from it from now in my case n i know its weired to compare the two but just wanted to know what say a certain type of card would compare to the 5960 interms of rendertime but if its so much trouble maybe best to forget about the gpu stuff for now, in my case that is until its at a point where its worth it

I would say that it would be best to test for yourself and see if it works for you.

[glorybound]

I wouldn't base your decision on me, at all. I haven't invested much time in seeing what the issue is, which could very well be me.