CoreWeave Guide Helps Conductor Users Pick the Right GPU for Renders

CoreWeave joins the Conductor platform

According to CoreWeave, it is now integrated with Conductor, which the company describes as "the fastest and most economical way to render at scale in the cloud." Artists and studios can access NVIDIA GPUs from CoreWeave through the platform to speed up rendering for visual effects (VFX), animation, motion graphics, design projects and large-scale workloads.

The company says access to more GPU SKUs, including the latest architecture from NVIDIA, lets users run jobs more efficiently and cost-effectively by matching resources to their needs. CoreWeave also points to benchmarks that compare its GPU and CPU instances with a general cloud provider on Conductor, covering both time and cost.

Three factors for choosing an instance

CoreWeave lists three considerations: GPU type, number of GPUs, and price versus performance.

GPU type. Bigger GPUs, judged mainly by CUDA cores and RT Cores, are not always better. RT Cores matter for denoising, post-processing and accelerating ray tracing. According to CoreWeave, GPUs scale well on jobs where scene load times and memory footprint are small compared with render time. For those, it says nothing beats the NVIDIA RTX A40 and NVIDIA RTX A6000 for now, thanks to their large VRAM and maximum numbers of RT Cores and CUDA cores. Where render times are short but many frames are needed, bigger GPUs may not scale as expected. For that case CoreWeave recommends the NVIDIA Quadro RTX 4000 and NVIDIA RTX A4000 to push parallelism as far as possible.

Number of GPUs. Conductor offers options from a single GPU up to 6 or 8 per type. When weighing something like 4x NVIDIA RTX A4000 against 2x NVIDIA RTX A6000, CoreWeave advises looking at the scene's GPU memory footprint. Many 4k or 8k textures, or low GPU utilization during renders, may indicate that fewer GPUs with more memory are the better choice. Conductor lets users test jobs on many compute types and compare results.

Price and performance. The fastest option is not automatically the best. CoreWeave notes that a job may render 50% faster but cost slightly more. It suggests benchmarking ahead of time to learn which instance types work in a crunch and which save money on routine rendering, and proposes measuring a normalized render time (Render Time / Total Cost of Job).

Available GPUs and what they offer

CoreWeave says it currently makes five NVIDIA RTX professional GPU types available for rendering on Conductor:

• NVIDIA Quadro RTX 4000
• NVIDIA Quadro RTX 5000
• NVIDIA RTX A4000
• NVIDIA RTX A5000
• NVIDIA RTX A6000 (limited availability)

The company claims these GPUs deliver more compute power, GPU memory capacity and scalability than older hardware and other visualization hardware. On CoreWeave Cloud, it says, they accelerate final frame rendering through bare metal instances, rendering scenes in a fraction of the time a CPU needs. Moving up the range brings more GPU memory, up to 48GB, plus features such as ray tracing and stronger AI/ML capabilities. CoreWeave cautions that core counts are not directly comparable between architectures and points readers to its benchmarks.

What it means for Conductor users

For studios already rendering on Conductor, the guidance is mainly a decision aid: match the GPU to the workload instead of defaulting to the largest card, and measure cost against runtime. CoreWeave says Conductor supports most pipelines in as little as 30 minutes from sign-up to reviewing rendered shots, and invites interested teams to schedule a call to discuss their workloads. The performance and cost claims come from CoreWeave itself, so its own benchmarks are the place to check them.