Nvidia’s Titan RTX: A Deep Dive
Last summer, Nvidia made a significant impact with the introduction of its new line of RTX Turing architecture-based cards, including the powerful Titan RTX. This new technology showcased real-time reflections, hinting at the future of visual effects.
This is more than just hype. Recall the stunning short, Star Wars: Secrets of the Empire, created by ILMxLAB? It was rendered in real-time with the Unreal Engine, utilizing a bank of Nvidia cards. Now, this level of performance can be achieved with a single card. The Titan RTX combines 72 Turing RT Cores to accelerate ray tracing, 576 Turing Tensor Cores optimized for deep learning and neural network operations, and optimizations for display, particularly for VR HMDs. It also has software support using AI for image manipulation and simulation calculations. All of this is powered by a remarkable 24GB of GDDR6 memory. Moreover, you can connect a couple of these cards using Nvidia NVLink, provided your workstation has enough power.
The RT Cores are particularly exciting. As a visual effects enthusiast, I’m always looking for ways to speed up the process. Game engines like Unreal are already providing real-time ray-traced shadows, reflections, and refractions. Nvidia is collaborating closely with leading render engine developers to leverage the RT Cores to boost ray tracing in production. Chaos Group (V-Ray), OTOY (Octane), Autodesk (Arnold), RedShift, and Pixar (RenderMan) are among the developers who already showcased demos. While the results aren’t strictly “real-time,” they’re fast enough to enable look development and lighting artists to work with supervisors or directors interactively, which speeds up creative decision-making.
Delving into the AI features, I was even more impressed. I’m just starting to explore the world of deep learning, trying to understand the workflows and the immense computing power needed for training. The Tensor architecture within the single Titan RTX greatly reduces that learning time, possibly eliminating the need for cloud-based training through services like Google or Amazon. As a newcomer, I don’t have any significant neural network training data, and I haven’t explored the card’s full potential yet.
The Titan RTX is expensive, and it’s geared toward users focused on extensive computing tasks, rather than those purely seeking higher frame rates in games. The less costly GeForce RTX 2080 Ti may be a better option for that market. But if you are pushing the boundaries of ray tracing, deep learning, and simulations, and need the best performance, the Titan RTX is an excellent choice.
Website: nvidia.com
Price: $2,500
Autodesk’s Maya 2019: Key Improvements for Artists
Earlier this year, Autodesk released Maya 2019, arriving later than 3ds Max. This time around, the focus is on performance and bug fixes, rather than headline-grabbing new features. This is expected for a mature software like Maya, which already has a comprehensive set of features. The fix list is extensive, while the list of “known issues” is small, indicating a solid release.
For artists, the primary benefit is improved scene interactivity. Viewport 2.0 is much faster, with intelligent approaches to screen updates. Only changing elements are updated, which makes for lighter throughput and quicker refresh rates. Playback is also faster, with caching taking place as you make changes. Instead of using play-blasts to review animations, the cache is available for you to play back inside the viewport. Real-time playback is crucial for determining animation timing, and wasting time on play-blasts is inefficient. Scenes also load faster, and selection and interaction are more responsive. Textures down-rez to optimize data traffic, and Wire Deformers have GPU acceleration, which speeds up deformation, especially with a powerful graphics card.
From a rendering and lighting perspective, you can render with Arnold inside the viewport, rather than having a separate render buffer. Like Viewport 2.0, the response is quick with near-immediate results (depending on the scene complexity). For lighting, this type of interaction is a game-changer, and convenience of having it inside the viewport just helps because you don’t have to allocate real estate to the other window.
Other performance improvements focus on tools to help users optimize their own workflow. The Evaluation Toolkit and Profiler have received enhancements improving diagnoses of scenes to determine processing bottlenecks. This filters down data to get to the core of the problems.
While it’s tempting to always want the very latest technology, sometimes the upgrades are subtle. For animators, Maya’s Viewport enhancements alone are worth the move to 2019. Along with its performance updates and bug fixes, it improves the Maya experience for all users.
Website: autodesk.com
Price: $195 (monthly); $1,545 (yearly); $4,170 (3 years)
Foundry’s Modo 13: A Powerful Update
Modo 13, recently released by The Foundry, offers a wealth of new features. I could probably write a whole book on it, so this is just an overview.
Let’s start with arrays used in the procedural modelling/schematic area of Modo. For those not familiar, Modo, among its features, is both a direct and procedural modeling tool. The Max/Maya approach uses straight modeling, while Modo/Houdini use a series of operators that build the geometry procedurally. Procedural modeling is non-destructive.
Modo’s node-based system has many tools for direct and procedural functions. Modo 13 has “arrays.” In Max/Maya, arrays are about creating multiple copies of an object in new patterns of multiple versions. Modo “arrays” are like a programming definition—collections of elements, usually particles and vertices. In Modo 13, you have nodes to manipulate the content, reorder them, and create geometry and curves from them. These tools are flexible and powerful.
The next impressive feature is the MeshFusion updates. MeshFusion has always been impressive, but it gets heavy as it adds two SubD levels to Boolean objects. Users can now control where MeshFusion adds those SubDs, and retain zero-level SubDs which provides creases along chosen edges. You can additionally control which edges and polys the MeshFusion will affect. If you are Boolean-ing a tube and a sphere together, the tube is hollow, so you can choose the outside polys of the tube, it will fuse the sphere with the outside of the tube only, leaving the space inside the tube. The potential here is amazing. And Boolean operations remain live and adjustable.
Animation now has procedural layers. Direct modeling is also improved, with dynamic edge bevel parameters and edge relax tools to affect edges after they’ve been created. There are alignment tools to quickly align elements and objects to others and UV updates to a UV toolset that is one of the best. I’ve been away from Modo for too long, and this release makes me want to jump back in.
Website: foundry.com
Price: $1,799
