OTOY Forums

I'm in the middle of a decision making process for proper GPU unit for octane-use..
I don't want to go up to a TITAN point.

Do you think going for a 4GB 770 is better, or 3GB 780ti? is 3GB too limiting for octane? If so, how much (and in what sense)? I am unable to judge which one is better since I don't know how easily/hardly 3GB is filled..

I just can't forecast since I don't have much experience with octane yet, so I want to know what you think.. Any ideas/suggstions are appreciated.

Hi zettam, the complete scene has to fit into the VRAM.
Geometry, textures for example.
A 4K texture is 160MB in the VRAM, 1Million polys is 1 GB iirc, your system need 3-400 MB too.
If you hit the memory limit Octane stop rendering.
The GTX 780Ti is much faster than a GTX 770, it is faster than Titan.
Take a look to your old scenes, how many RAM are they use.
Think about a second card for display, if possible on your mainboard, than you have full 3 GB for Octane.
There are many tricks to reduce memory, reduce texture size, use instancing for plants, low poly objects in background for example.

Cheers, mib.

hi, Zettam. By now Mib2berlin gave You some nice points to think of.

however there's one more thing. as You render, depending on Your output there is some space reserved in vram for what is called Film Buffer. It might not be so demanding at resolutions like 1024x1024 or 2048x2048 where it's only 20 & 80 MB acordingly, but if You step further..4096x4096 will use 320MB & 8192x8192MB will take up to ..1.3GB. So..If you are going to make some work for print resolutions, keep this in mind =)

cheers

I suggest to go with 1 780 ti. and in future get 6GB/12GB versions if the news is right.

Thank you very much for the replies,
In this case, which one would be a better choice? 770 4GB or 780ti 3GB?
I was planning on getting a 770 4GB but there is a really good discount here for 780ti right new, which makes its price quite close to 770 4gb.

As the price is close or money is no object here, what I would advise to look at a purpose of usage. What are You going to use if for?

780ti will be between 2 & 3 times faster compared to 770, but it has 1GB less of vRam..

If You are going to do simple scenes, sort of a product rendering with few bigger textures, web resolutions etc. - think 780ti in this case fits better. You just work 'faster'..

However if You're lookin' to more complex scenes, where You're going to have bit more of geometry or maybe more textures..or higher output resolution - extra 1GB could be a lot. Even if You end up rendering slower..



Recently I've found other interesting thing ( http://render.otoy.com/forum/viewtopic.php?f=6&t=37141 ) If I render let's say 2k square output, it might take me half an hour to clean an image (GTX670). However, that's a nice card with 4GB & I only use way under a 1gig for simple test scene, so I've updated textures switched to to higher res, changed output to 8k..on the same time tweaked some settings in Kernel & Imager to have a bit of blur - fireflies goes out after few minutes.. - so I can make cleaner image in less time at higher res (in expence of using more of vram - in my case the diference is 2-3times).

It's not a rule though, just a personal test, as an example to ilustrate where bigger card might come useful. In my case I would have biggest card possible, I've bought mine just before Titan came & if it were there, I would have grabbed it for sure (I was considering to send mine back). I've upgraded from two 460's (1.28GB), to 670 (4GB) - & this is more or less the same speed, but with additional space I could tweak my workflows to fit my needs better.


Some are making a lot of nice works with 3gb cards (or even 1.5GB) - Just look over the gallery, read a bit the technical info provided, take a look to these signatures below - & You'll get idea 'bout what is possible with what budget.


It's hard to recomend honestly. Without knowing what You're lookin' for, both of these are great cards.

What about particles?
It should be very easy to fill up the ram with particles -

- you have to make sure you use instancing if you use particles (depends on if you're using some external plugins to generate them) - octane handles instances better then your viewport probably does, that way you can have a lot more particles showing on octane than in you viewport

- I had a lot more problems trying to optimize the viewport on 3ds max cause it slows down incredibly with heavy scenes - plugins like multiscatter can show you the octane render but nothing in your scene, so you can rotate and move easily (so much for 3dmax new nitrous viewport display which is supposed to be incredibly faster, octane works faster rendering the geometry than 3dmax displaying wireframes, go figure..)

- and I am using 1.5 Gb VRAM with 580 and 590gtx cards, and used them for some very heavy landscape scenes, with a lot of instancing - you can optimize a lot and get away with lower memory usage

- you would need a second card though, because the system will eat up a few hundred MBs of VRAM if you're rendering everything on one GPU (any other application you open will use VRAM), and you will rage constantly because it will slow down refresh rates as well, not beeing able to do much of anything else on you computer while rendering

- actually I like the setup where the other GPUs burn out rendering on other 2 or 3 cards, and I actually play battlefield or something on the main GPU while it's working , kills the waiting time, nothing is affected, no crashes, I did that a lot actually

- in the end, you get much faster updates if you optimize anyway - doing it lazy and letting renders reach the limits is going to take you that much more to work on anything, if you're going to use 8gb scenes instead of 1gb - you'll have to wait 8 times more to load the octane render or update the scene with new geometry - it's like, who wants to work on 20gb scenes, unless you really have to (I remember the crappy 3gb RAM limits on 32 bit systems, where everything is loaded in RAM, this way ram takes on generating the scene, and VRAM only uses the geometry it generates)

- I'm wondering, if you take every single polygon from geometry and make it an instance do we get infinite scene sizes?

- anyway I would rather take a 3gb faster card than a 4gb slower one - I'm working on 1.5gb VRAM so I am sure this shouldn't be such a problem

Thank you very much for the detailed explanation. I made my pock and went for a gtx 780ti, crossing finger to see whats going to happen at this point.