While I know Digistor 3D/VFX customers are no stranger to V-Ray, it's used across many industries and the Chaos Group tours always see heavy turnouts, I still wanted to take a moment to describe it. V-Ray is an advanced CPU based global illumination raytracer that can efficiently create photo-real results like these featured in the V-ray Gallery and Showreel.

V-Ray 2016 Showreel

While some customers are constantly experimenting with it's numerous expert-mode options or using every last feature, a majority find their preferred workflow as quickly as possible and use it to produce consistently amazing work without exploring too deep. It's these customers this post is predominately intended for.

conor harll design architecture vray 3ds max 870 2

V-Ray RT is a modern next-generation 3D render engine included with V-Ray designed for real-time throughput. Essentially what that means is it's able to render live, updated, scene data and display regular result outputs as it progresses. Over the years I've only recommended this for look development and preview renders, and even though it's still not quite at feature parity with regular V-Ray it's definitely become mature enough that this is no longer the case. Provided your scenes only use it's supported features you can use it for final production renders.

V-Ray RT not only runs on the CPU, but can run on the GPU too

While V-Ray RT can be used on existing workstation and render-farms using the CPU it also includes a powerful GPU powered kernel designed for high performance on multiple graphics cards. While the cross-vendor OpenCL option is available Digistor only recommends using it in CUDA mode which is only available on nVidia GPUS, the more modern the better - and preferably on a dedicated card so your main graphics system can continue driving the desktop and main 3D viewports.

V-Ray RT Network Distributed GPU Rendering

Similar to V-Ray's "bucket" style Distributed Rendering, V-Ray RT is also supported on renderfarms and can be used over the network in both CPU and GPU modes.

Unlike some versions of regular V-Ray, V-Ray RT's GPU network spawner mode is not dependent the plugin host application being installed to render, as the scene data must be fully translated into a V-Ray RT GPU compatible format (which is a normally undesirable overhead). This makes it simpler to purchase dedicated GPU powered devices, such as the nVidia Visual Computing Appliance (VCA) and other solutions are available to give an instant speed boost to your V-Ray RT rendering.

Current Limitations

Unfortunately not all V-Ray features are yet supported in RT so Chaos Group publish the up-to-date feature comparison list here, showing what's supported in both CPU and GPU modes. Beyond that, CPU mode works as you'd expect, where as GPU mode has an additional resource you'll need to be aware of. GPU RAM.

Until relatively recently the only way to get a reasonable amount of RAM on a GPU was on the highest end Quadro and Tesla graphics cards, but the nVidia Titan series changed everything and each iteration has only gotten stronger. The latest model, dubbed Titan X (Pascal) has 12GB of high bandwidth graphics RAM. This resource limit is extremely strict. If your scene data fits within this hard limit it can be rendered on that GPU, if not, it can't. Adding multiple cards to a system unfortunately doesn't increased this limit as they all must load an independent copy of the scene data. Should you go over your limit you'll need to either optimise your scene complexity or drop back to CPU mode. This resource limit is easily the number 1 limitation to the technology.

robert hennings panda art vray modo 870 c

The near future of GPU rendering

Thankfully technology never stands still and nVidia have been busy working on a solution - NV-Link . NV-Link is a high speed GPU-to-GPU data bus designed for graphics cards to access each others RAM directly, bypassing the PCI-Express bus they use to communicate with the host CPUs. This allows multi-card GPU RAM to become cumulative, thus changing things dramatically. Once you can expand your available GPU RAM by simply linking them together it makes expandable, stackable, large scene GPU rendering appliances possible, so watch this space. In the mean time, the speed returns of working within the current limits is more than worth it.

Please get in touch with Digistor and check out Chaos Group's in depth guide to GPU rendering.

You can see some great example of how a GPU powered network renderfarm greatly increases performance for production renderers in this example tutorial using Forest Pack and this Chaosgroup VCA Demonstration.

GPUS are also in the cloud, it's a bit old now but here's an interesting MotionBuilder GPU Cloud Integration Experiment.