Graphics processing units (GPUs) are specialized computer processors designed especially for rendering images. They use their own video random-access memory (VRAM) chips and perform multi-core parallel processing to deliver high-speed rendering for 3D animation previews, video games, virtual reality (VR), and other interactive media.
GPUs are computer processors built for rendering images. They process in parallel using thousands of small, low-power cores, relieving serial-processing CPUs (central processing units) from graphics-heavy work. When combined with GPU-optimized rendering software like Autodesk Arnold, the massive parallel processing power of GPUs can offer a speed boost over CPU rendering. That productivity, along with GPUs’ lower power consumption compared to CPUs, makes them cost-effective.
However, there’s more to the CPU vs GPU rendering debate. GPUs use VRAM and usually have much less memory than comparable CPUs using RAM, so they may not perform as well as CPUs when rendering highly complex scenes and simulations requiring a lot of memory. GPU rendering excels for less complex scenes with lower memory requirements for polygon counts, textures, and mesh topology, as well as real-time rendering for video games, VR, and other interactive apps. For the best-quality results on highly detailed scenes and graphics based on complex physics, CPU rendering may be the better choice despite its longer processing times and higher energy use.
GPUs do have the advantage of scalability: It’s much easier to combine multiple GPUs into a rendering setup for added performance than it is to add CPUs. Another boon comes from GPU-accelerated rendering software such as Arnold. That software is optimized for GPU hardware and boosts performance for tasks like real-time viewport playback, interactive rendering, and denoising. GPU-accelerated rendering software has become popular for 3D-modeled animation and visual effects, gaming, virtual reality, and photorealistic visualization for architecture, product design, and entertainment media. With Arnold, artists can seamlessly switch between CPU and GPU rendering, depending on the task at hand.
Though GPU rendering can be a high-performance solution when artists need rapid iteration and fast rendering, such as with game development and 3D animation previews, it also comes with added caveats. Because they typically have less memory than CPUs, GPUs may become unstable if running both a computer display and rendering software. They also come with drivers that have to be updated to maintain stable operation with other computer hardware.
With that in mind, a number of other technologies can increase the visual quality and speed of GPU rendering. For example, NVIDIA NVLink technology can create a high-bandwidth interconnect between multiple GPUs or between a GPU and CPU. This enables more memory to be used for large rendering jobs and speeds up GPU rendering times for data-intensive scenes.
GPUs perform parallel processing and the more processor cores they have, such as the CUDA cores in NVIDIA GPUs, the more processing tasks they can handle simultaneously, such as the vertex processing, pixel shading, geometry processing, and texture mapping needed for rendering.
Certain modern GPUs are also optimized for ray tracing, a computationally intensive rendering technique for creating photorealistic lighting effects. With ray tracing, rendered graphics can realistically simulate light refracting through transparent objects or water, reflections, shadows, and other light path behavior.
GPU rendering can help 3D media projects of any size with benefits in performance and cost, including:
GPUs are built for multi-core parallel processing with potentially thousands of small computing cores on a single GPU. As a result, GPU rendering can render 3D scenes much faster than comparable CPUs.
For artists working in 3D animation, video games, and other media where it’s an advantage to see their work in real time, GPU rendering produces high-quality, real-time video and visualizations, where artists can make adjustments to their scenes, lighting effects, and other graphical elements smoothly.
High-end rendering GPUs compare favorably in cost to high-end CPUs and use less energy to render. And because they render faster than CPUs, they can result in long-term savings from increased productivity.
GPUs are still evolving, with more frequent hardware and software updates and optimizations than. Some GPUs are even optimized for certain tasks, with specialized rendering GPUs offering advanced features like ray tracing or deep learning super sampling (DLSS) for image upscaling.
When a single rendering GPU won’t suffice, it’s easy to scale up in power by adding more GPUs to a rendering system. It may also be possible to set up each GPU dedicated to specific tasks.
Tools for 3D visual effects, finishing, and 3D compositing—available as Flame, Flame Assist, Flare, Lustre
MEGALIS VFX
FX supervisors of Megalis VFX explain their partnership with Autodesk, which helped ensure that after switching to GPU rendering, Arnold software would successfully render the characters, large environments, and high-quality lighting of the Netflix show ONI: Thunder God’s Tale.
Image courtesy of Megalis VFX
IGNITE ANIMATION STUDIOS
Hear from the studio founders who built a custom remote-work pipeline to create the animated short Andy: A Dog’s Tale using Autodesk Maya, Flow Production Tracking (formerly ShotGrid), Unreal Engine, and Autodesk Arnold for rendering.
Image courtesy of Ignite Animation
AXIS STUDIOS
When Axis Studios created two hugely successful cinematic trailers for the Innistrad: Midnight Hunt, Magic: The Gathering cards packs, it built an evocative horror world of werewolves, vampires, and gothic architecture, relying on Autodesk Maya for character animation and Arnold for shading, rendering, and adding versatile effects ranging from photorealistic to stylistic.
Image courtesy of Axis Studios
The latest versions of Maya and 3ds Max include artist-friendly features like improvements to working with the OpenUSD standard, productivity enhancements, automation of mundane tasks, a GPU rendering overhaul, and more.
Find out everything you need to know to take advantage of the GPU rendering overhaul in Arnold’s user guide.
The next-generation technology of Autodesk Flame 3D VFX software unlocks advanced computing and graphics capabilities for artists, including support for rendering with the latest NVIDIA GPUs.
Get all the resources for Autodesk’s rendering specialist software, Arnold, including tutorials, documentation, research papers, a free trial, and more.
Learn all about Autodesk Maya’s GPU-based renderer in this one-hour Autodesk University video class. It covers efficient workflows, when to use different rendering options, and tools like shaders, lens effects, and more.
Arnold is the advanced Monte Carlo ray tracing renderer the industry relies on for intuitive use and full artistic control, as well as abilities like seamlessly switching between CPU and GPU rendering.
GPU rendering will work for your project if it requires real-time rendering or if the scenes for rendering are not based on complex simulations or physics. If rendering speed matters, the parallel processing of a rendering GPU is generally faster than CPU rendering for simpler scenes not based on complex simulations and physics.
Make sure your software fully supports GPU rendering; some software even supports the seamless switching between CPU and GPU rendering.
The choice between using CPU or GPU rendering depends on the specific project’s needs. Current CPUs can have much more RAM than GPUs, making CPU rendering good for large scenes and those based on complex simulations and physics. CPU rendering is slower but suited for long periods of heavy workloads.
GPUs on the other hand have parallel processing, often with many low-powered processing cores, making them faster and more cost-effective than CPUs for rendering less complex scenes. GPUs are also the preferred choice for real-time rendering of video games and interactive applications.
The disadvantages of GPU rendering include limited VRAM capacity compared to the RAM of CPUs. High-end GPUs may have less than half the memory of high-end CPUs, which may limit the complexity or number of elements in a scene a GPU can render. This may also affect the quality of GPU rendering and produce less accurate lighting effects than CPU rendering.
The RAM problem can compound if powering both a display and rendering processes from the same GPU, which will divide the available RAM and may make the GPU unstable. GPUs also need driver updates to maintain compatibility with certain hardware; not updating drivers can also affect GPU stability.
