Smarter Memory, Smoother 3D Workflows

Modern media and entertainment industry workloads—including 3D simulation, rendering, and beyond—are more demanding than ever. Studios rely on applications like SideFX Houdini, DreamWorks’ MoonRay, Pixar’s RenderMan, or others to produce visually stunning content, but these applications often push systems to their compute and memory limits. High-resolution assets, complex shading models, and the need for physically accurate rendering usually translate to massive memory footprints—which leads to teams having to overprovision DRAM at significant cost.

At MEXT, we set out to change that equation. MEXT’s AI-Powered Predictive Memory™ software expands the memory capacity of the system by transparently using flash as a lower-cost memory tier, while predictive algorithms ensure workloads perform as if everything lived in DRAM. The result? Studios get the performance they need without the cost, complexity, or supply constraints of adding more physical DRAM.

Run MoonRay with HALF the DRAM

We evaluated MEXT performance running high-fidelity MoonRay scenes, comparing a 64GB all-DRAM system with a 32GB DRAM + MEXT system. The MEXT-enabled system delivered comparable performance with 40% lower memory-related costs and 29% lower total costs—translating to a 1.4X performance-per-dollar improvement. 

This means that when running on workstations with just half the DRAM, animators saw virtually identical rendering speeds and frame quality compared to the traditional all-DRAM configuration. MEXT’s predictive engine maintained performance with 98% accuracy in anticipating which memory pages MoonRay would need next.

Stop RenderMan Crashes Before They Happen

We evaluated MEXT performance for a RenderMan workload that required around 98GB of DRAM to function properly. Running it on the original 64GB all-DRAM system caused repeated out-of-memory errors. The mere addition of MEXT software to the original system, however, managed to eliminate these errors by expanding the effective memory capacity to 128GB. 

In this way, MEXT was able to circumvent the need for hardware upgrades in the range of $10K or more per user. For creative teams, this means more reliable production, fewer interruptions, and greater ability to push creative boundaries—without having to waste the time and money involved with procuring larger-memory systems. 

Boost Performance up to 4.4X for SideFX Houdini

We also tested a Houdini workload (that needed about 68GB of DRAM) across multiple configurations. On a 16GB system, it took around 22,000 seconds to complete. Upon adding MEXT, the completion time shrank to 5,000 seconds—a dramatic 4.4X improvement. In another test, we had a 32GB system running a similarly-sized workload. It took 3,400 seconds to complete. After adding MEXT, this reduced down to 2,700 seconds—a 1.3X improvement. 

Scaling Beyond

While our testing focused on these 3 flagship applications, the implications go far beyond. Many creative production workloads share common traits: large datasets, irregular memory access patterns, and sensitivity to latency. MEXT’s approach—offloading cold memory pages to flash, predicting future accesses, and transparently moving data back to DRAM before it’s needed—delivers consistent benefits across the spectrum, regardless of the particular application your team may be running. 

• Expanded effective memory capacity without new hardware (purely achieved through existing system flash) 
• Lower infrastructure costs while preserving performance
• Seamless deployment with installation happening in <5 minutes, with no changes needed to applications

Whether running on workstations or scaling across a server farm, studios can leverage MEXT to achieve a level of elasticity and efficiency that dramatically exceeds that of traditional pure-DRAM configurations. Animators can work faster, IT teams can stretch budgets further, and production pipelines can scale with confidence. For an industry where every frame matters, smarter memory is now the smarter choice.

Want to dive deeper? Explore the whitepapers on these results here.