Art Meets Innovation: Inside Pixar’s RenderMan XPU Revolution

🎨 The philosophy behind the work

I carry a simple expression with me in everything I do at Pixar: "The art challenges the technology, and the technology inspires the art." That line is more than a slogan. It is a compact guide to how we approach filmmaking as both a technical and artistic endeavor. I tell teams that when an artistic need appears, it becomes a concrete problem that our engineers solve. When engineers solve new problems, the artists discover new possibilities. That interaction sits at the heart of what RenderMan XPU is trying to achieve.

"The art challenges the technology, and the technology inspires the art."

RenderMan has long been the engine that performs the heavy lifting — the math and physics that determine the color of every pixel across entire films and shorts. Historically, that work has been dominated by CPU rendering. But as I see it, we are entering a new phase where GPUs are not just an adjunct but a core enabler. That shift changes not only how we compute pixels but how we tell stories.

🚀 What RenderMan XPU is and why it matters

RenderMan XPU is Pixar’s push to harness GPU acceleration within the RenderMan ecosystem. In practical terms, this means moving significant portions of rendering workloads from CPUs to GPUs so artists can iterate faster, handle more complex scenes, and free up engineering and farm resources.

There are three reasons this matters right now:

Memory capacity: Recent GPU architectures have increased onboard memory dramatically. That lets entire shots that previously required CPU memory to be processed on a single GPU.
Raw parallel throughput: GPUs excel at the highly parallel math that path tracing and shading require. For many workloads, that translates directly into faster renders.
Pipeline efficiency: GPU acceleration changes the economics of the render farm. Faster frames and greater local capacity reduce queuing, cut costs, and enable different workflows.

I describe RenderMan as the piece of software that does all of the math and physics to produce the final color for each pixel. XPU extends that capability by using the parallel horsepower of GPUs while preserving the quality, robustness, and feature set that artists expect from RenderMan.

⚙️ How GPU acceleration changes production workflows

In my experience, the biggest practical effect of moving to XPU and GPU acceleration is a qualitative change in how artists work. Rendering becomes less of a waiting game and more of an immediate creative tool.

Lighting is the clearest example. Lighting has traditionally been a bottleneck in production because every lighting adjustment required substantial render time. That forced lighters to make conservative choices, rely on approximations, or batch work in ways that constrained their creative instinct. With GPU acceleration, the turn-around on lighting passes shortens. Lighters can try more ideas, refine subtleties, and apply principles of cinematography rather than focusing on technical constraints.

Here are some workflow changes I’ve observed or expect to see:

More interactive iteration: Faster renders mean artists explore a wider variety of lighting, camera, and compositional options.
Higher-quality look development: Artists can afford to run full-quality renders earlier and more often, catching issues before they propagate downstream.
Distributed farm efficiency: When a larger fraction of shots fit on a single GPU, farm scheduling becomes simpler and utilization improves.
New role dynamics: Engineers and artists collaborate more closely on renderer features rather than merely performance tweaks.

That last point is important. For creative teams, the time saved by faster iteration translates directly into better storytelling. That is the real return on investment for adopting GPU-accelerated rendering.

🔬 Technical shifts: from CPU to XPU and modern GPUs

Understanding why XPU is possible now requires a short detour into hardware trends. GPUs have evolved rapidly over the last several generations. Where they were once confined by smaller amounts of memory, recent architectures have vastly expanded onboard RAM and memory bandwidth.

In presentations and internal discussions, I frequently point to the progression of recent cards. The memory limitations that used to force us to split scenes across machines are largely gone. With modern GPUs, a surprisingly large percentage of our shots can fit entirely on a single GPU. In fact, I now estimate that roughly 99 percent of our shots will fit on the latest Blackwell-class GPUs.

That is not just a statement about capacity. It also affects how we architect the renderer. When scenes fit in GPU memory, we can:

Keep geometry, textures, and shading data local to the GPU, avoiding expensive CPU-GPU transfers.
Run shading, ray traversal, and sampling entirely on the GPU for most of the frame, maximizing throughput.
Use GPU-native denoising and filtering techniques to reduce the number of samples needed for acceptable noise levels.

Moving workloads to GPUs also forces us to re-examine how rendering algorithms are implemented. GPUs reward massively parallel work and penalize serial bottlenecks. So the software must be redesigned in places to avoid thread divergence, to improve memory access patterns, and to provide a consistent developer experience across CPU and GPU paths.

From a software standpoint, RenderMan XPU is a hybrid approach: it keeps the proven, high-fidelity RenderMan features intact while executing them efficiently where GPUs excel.

💡 Lighting: the biggest near-term win

Lighting has always been a discipline where art and technology collide. Cinematography is essentially the art of lighting motion pictures. For our supervisors and lighters, the ability to make dozens of adjustments and see results quickly is transformative.

When lighting takes hours or days to converge, artists make conservative choices and test fewer options. Faster renders change that calculus. Lighters can take risks, try more stylistic or risky approaches, and iterate into more nuanced photographic choices. We see a direct correlation between iteration speed and creative quality.

Concretely, GPU acceleration helps lighting in these ways:

Faster feedback loops: Shorter render times mean you can test multiple exposures, fill intensities, and shadow directions in the time it used to take to run one test.
Real-time inspiration: Sometimes a subtle change to a rim light or the color temperature of a key can reveal new character beats. Those discoveries happen more often when you can try things instantly.
Better integration with color grading and compositing: Having high-fidelity renders earlier in the pipeline allows colorists and compositors to work with higher-quality inputs sooner.
More consistent cinematography: With iteration, lighting decisions can be made based on intent and mood rather than technical constraints.

My observation is that lighting benefits disproportionately compared to other stages. This is because lighting is fundamentally iterative and aesthetically driven. Anything that shortens the feedback loop directly empowers the artist.

🎬 Real-world application: rendering a portion of Toy Story 5

One of the most exciting outcomes of this work is that we are rendering a portion of Toy Story 5 using XPU. That is not a token experiment. It is a real creative piece of a major production, chosen because the technology enables a distinct artistic advantage.

Rendering a major feature film or a segment of one with a new rendering path is a careful, deliberate process. It requires ensuring parity with existing shots, validating artistic intent, and proving that the renderer handles all the production cases we throw at it — complex characters, layered shaders, volumetrics, hair and fur, and large environments. We only commit to such a change when the risks are understood and the benefits clearly outweigh the costs.

For Toy Story 5, XPU gives the artists:

Greater flexibility: Artists can try new lighting setups, camera moves, and shading tweaks without long waits.
More efficiency: We can render more frames faster, enabling more creative exploration within production timelines.
Improved farm utilization: Using GPUs efficiently reduces the need to overprovision CPU resources.

This is not just about speed. It is about expanding the set of creative choices available to directors, cinematographers, and lighting artists. When those choices expand, the stories we can tell become richer.

🤝 The partnership with NVIDIA

I often talk about technical partnerships as strategic relationships. Our collaboration with NVIDIA reaches into the core of what Pixar is about. They provide the hardware platforms and architectural roadmaps that make XPU practical, and we contribute production expertise about how that hardware will be used in real-world feature animation.

The partnership is not purely transactional. It is a conversation about the future of storytelling. NVIDIA's hardware advances — both in compute and memory — have been instrumental. At the same time, our production requirements drive new features and optimizations in drivers, SDKs, and hardware priorities.

That synergy is essential. It means that when a new GPU architecture arrives, we are not faced with a porting exercise alone. Instead, there is a path to incorporate those architectural gains into tools that artists use every day. For instance, when GPUs reduce memory constraints, we can stop designing artificial scene-splitting workarounds and start focusing on shaders, volumetrics, and global illumination at higher fidelity.

The practical implications of this partnership are felt during production and in the research labs. Together we are exploring how to make rendering both faster and artist-friendly, which is ultimately the point: enabling storytellers to do more.

🧰 Practical takeaways for studios and artists

Switching to GPU-accelerated rendering is a strategic decision. From my perspective, the path to adoption should be pragmatic and staged. Here are actionable takeaways I recommend for studios considering this transition.

Start with hybrid workflows: Don’t expect an overnight migration. Run GPU-accelerated paths where they make sense and keep CPU paths for corner cases until those are solved.
Profile and measure: Identify where your current pipeline spends the most time. Focus GPU efforts on those hotspots, such as heavy shading networks, complex lighting, or volumetrics.
Invest in tools and training: Artists need interfaces and presets that make GPU features accessible. Training reduces friction and accelerates adoption.
Validate parity: Establish tests that verify visual parity between CPU and GPU renders, especially for subtle effects and compositing workflows.
Upgrade infrastructure thoughtfully: Buying GPUs without considering cooling, power, and scheduling software is a mistake. Plan data center changes and cloud strategies in advance.
Leverage denoising: Using modern GPU denoisers can reduce sample counts and overall render time without compromising quality.

These are practical steps I have seen work. They allow studios to get the benefits of XPU while managing the risks of introducing significant platform changes into production pipelines.

⚖️ Challenges and considerations

No technology transition is without trade-offs. From my vantage point, here are the most important considerations to keep in mind before committing fully to GPU-accelerated rendering.

Compatibility of shaders and plugins: Some legacy shader code and third-party plugins may not be GPU-ready. Porting or re-authoring may be necessary.
Determinism and debugging: GPU execution models can make reproducing certain subtle bugs more difficult. You need robust debugging tools and validation testing.
Memory management: Even with larger GPU memory footprints, extremely large scenes still require careful asset management and streaming strategies.
Operational costs: High-end GPUs are power-hungry. Plan for electrical and cooling costs in on-prem environments or weigh cloud options and pricing carefully.
Vendor coupling: While GPU ecosystems are maturing, be mindful of potential lock-in and design software layers that can be maintained across hardware generations.
Quality control: Ensure that denoising and accelerated algorithms do not degrade subtle visual cues that matter for storytelling.

Addressing these challenges early and pragmatically is what separates a smooth rollout from a costly experiment. When done right, the benefits outweigh the complexities by a wide margin.

🔭 The future: where art and technology go next

Looking forward, GPU acceleration is one step on a longer trajectory. Rendering is becoming more immediate and interactive, and that has implications for everything from previsualization to the final frame.

Here are a few future directions I expect to see accelerate:

More immediate creative feedback: The line between lookdev and final rendering will blur. Artists will work in environments where the quality of final renders is available much earlier.
Integration with AI: AI-driven tools for denoising, material generation, and even lighting suggestions will augment artists. Machine learning will help compress complex data and speed workflows.
Virtual production and real-time techniques: Real-time engines will borrow more from offline renderers to deliver higher visual fidelity in live contexts.
Democratization of tools: Advances in hardware and cloud availability will make high-fidelity rendering accessible to smaller teams and independent creators.
New forms of storytelling: With fewer technical constraints, creators will experiment with visuals and narrative structures that were previously impossible.

All of this points toward a future where technology acts as an amplifier of artistic intent rather than a limiter. For me, that is the most exciting aspect of the current moment. It is not simply about rendering more pixels faster. It is about enabling artists to focus on shape, color, light, and storytelling.

🔁 Why this is more than just a performance upgrade

Performance matters, but the true value of RenderMan XPU is the way it reframes work. When technical limitations recede, creative problems take center stage. With that shift, teams spend more time asking "What does the scene need emotionally?" and less time asking "How can I make the technical budget fit?"

That qualitative shift has ripple effects across production. Directors can test cinematic choices earlier. Lighting leads can mentor and iterate faster. Compositors receive stronger inputs. And storytellers can conceive sequences that would have been impractical before.

Put another way, XPU is not just a tool for faster renders. It is an investment in a different creative economy — one in which time is less often the limiting resource, and artistic exploration gets a disproportionate share of the budget.

📌 My closing thoughts

I believe we are at a pivotal moment for visual storytelling. The combination of RenderMan's proven capabilities and modern GPU architectures creates opportunities that were hard to imagine even a few years ago. When artists spend less time waiting and more time experimenting, the final film benefits in ways that are hard to quantify but immediately visible.

Rendering a portion of Toy Story 5 with XPU is a milestone because it signals confidence in the new path. It is not just about one film. It is about how the next generation of films will be conceived and made — with fewer technical compromises and greater artistic freedom.

The art challenges the technology, and the technology inspires the art. That interplay has always defined what we do. With RenderMan XPU and modern GPUs, I feel like we have a new set of tools to answer that call.