Best Used GPU for Rendering 3D and Video

[jiw2033]

So i'm putting a work station together and my plan is to build on the Ryzen 1700.

The rig is designed for content creation and whilst video is part of my plan i'm far more interested in creating 3D content.

Ultimately the packages i would be looking at are 3DS Max, Blender, Daz3D and Reallusion iClone. But the software isn't the only point but also which renderer you use to render the scene. For example Blender has at least 6 render engines you can use. Including Cycles, LuxRender, Maxwell, Mitsuba, Octane, and V-Ray. All producing slightly different results and taking a different amount of time to render the scene.

https://www.blenderguru.com/articles...les-vs-giants/

My plan was to use two GTX 780 Ti in SLI. Each card has 2,880 Cuda cores so in SLI i'll have 5,760 but still have just 3GB vram. The combined price of both cards was £250.

I could opt for a single GTX 980 Ti which has just 2,816 cores but is over 30% faster in benchmarks than a single 780 Ti and has 6gb vram. A single card would cost £250 used.

Finally, now that the 1080 Ti has been released you can find the standard 1080 for just £320, used, on eBay. This has just 2,560 Cuda cores; the least of them all. However it does have 8gb vram and in gaming out performs the 980 Ti by 30%. However i understand in Adobe premier it's no better than the 980 Ti.

I tend to think that more slower cores is better than faster fewer cores. Like this example comparing rendering using a GPU over a CPU: https://www.youtube.com/watch?v=-P28LKWTzrI

So the question is which is the better setup for 3D rendering, watts aside? 2x 780 Ti, a single 980 Ti, or a single 1080?

Anybody with experience in this area i would value your opinions. Thanks

 

[Chuk_Chuk]

I think render engines prefer more cuda cores.
Here is the benchmark thread for blender https://blenderartists.org/forum/showthread.php?419670-Cycles-Nvidia-Ultimate-Benchmark-Spreadsheet

you will get faster rneders with 2x 780 TI but you'll need to look into your needs. If most of the work will fit within the ram constraints then you should go for it.

 

[Quixote]

For the adobe stuff, I wouldn't worry too much. Sure, it's accelerated but it isn't really too important & you'll get some benefit from any card.
For general viewport/ editor work in your 3D apps (3ds Max, Blender Cinema4D, Maya etc) you'll more often be bottle-necked by CPU performance when it comes to deformers, skinning, cloners etc The graphics card will come into play when you have large static polygon counts ie very detailed models with little animation or deformation - performance here is broadly similar to gaming performance.

It's really if you are intending to mainly do GPU rendering that the Graphics card becomes very important.
Firstly you can't compare cuda core counts directly between different generations of card - you can compare core counts between a 970 gtx & a 980ti gtx but not between a 780ti & a 1080ti for example.
Be aware that you don't add up the VRam of the cards for GPU rendering - each card has to be able to load the scene in memory.
Also be aware that Windows 10 makes a chunk of your VRam unavailable for GPU rendering - a chunk which increases the larger your VRam - on my new 1080ti it eats nearly 2.5 of the 11gb of VRam & there is nothing you can do about it.
Although Redshift & Octane (maybe others too) can use 'out of core' rendering to store some of the textures in system RAM, the meshes & frame buffer still have to fit on the main card.
For these reasons I would rule out the 780ti's -3GB just isn't enough.
Bang for buck the 6GB 980ti's are probably the best, you can always add another at a later date. Or if you have more money, the new 1080ti's are actually very good value for GPU rendering but still £650 a pop.