How to Choose a Rendering GPU for Conductor

Strategies to find balanced speed, cost, and parallelism

Artists and studios can now use CoreWeave GPUs on the Conductor platform to accelerate rendering for visual effects (VFX), animation, motion graphics, design projects, and large-scale workloads. CoreWeave Cloud accelerates final frame rendering by using NVIDIA RTX GPUs on bare metal instances to render scenes in a fraction of the time compared to a CPU.

CoreWeave offers six different GPU types for Conductor, each with different architecture, memory, and numbers of cores. Each type can be deployed in a single or multiple GPU configuration. Because CoreWeave offers a wide range of GPUs, it’s important to understand how to choose the correct type and number of GPUs to meet the desired price and performance goals.

In this chart, GPUs are listed in order of compute performance, GPU memory, and cores available for ray tracing and AI/ML.

NVIDIA GPUArchitectureMemoryCUDA CoresRT CoresMax GPUs

RTX 4000

Turing

8GB

2304

36

7

RTX 5000

Turing

16GB

3072

48

4

RTX A4000

Ampere

16GB

6144

48 2nd Gen

7

RTX A5000

Ampere

24GB

8192

64 2nd Gen

8

RTX A6000

Ampere

48GB

10752

84 2nd Gen

8

A40

Ampere

48GB

10752

84 2nd Gen

8

Workload configuration

Every workload should consider the GPU type, number, and price versus performance. The decision when to choose a more powerful GPU, or multiple GPUs, should be guided by considering these factors:

  • Jobs with long render times that need maximum performance and VRAM should consider the NVIDIA RTX A40 and RTX A6000. They have the highest capability and perform best for large renders with relatively short scene load times.

  • Jobs with short render times and many frames should consider running several smaller GPUs in parallel to deliver the performance needed with better economics. For jobs that benefit from parallelism, especially when less VRAM is needed, try GPUs at the lower end of the scale like the RTX 4000, RTX 5000, and RTX A4000. Larger GPUs may not deliver the expected performance scaling or efficiency.

When choosing a multi-GPU configuration, consider the number of cores available. Larger GPUs have more CUDA cores and RT cores, which are particularly important for accelerated denoising and ray tracing jobs. Also pay attention to the memory footprint of the scene. If a scene with large amount of 4k or 8k textures has low GPU utilization during renders, two RTX A6000 GPUs may perform better than four RTX A4000s, because the A6000s have more VRAM available.

Real-world performance

When comparing GPUs, note that Turing and Ampere architecture have different performance per core. When deciding the number and type of GPU to select, we recommend setting up a rendering pipeline to quickly evaluate how expected jobs perform on different GPU types.

Conductor makes it simple to set up rendering pipelines on CoreWeave Cloud, so it's easy to run real-world tests with many GPU configurations and compare the results from each. It’s important to understand the relationship between render speed and cost, and how that translates to the type and number of GPUs for specific workloads in different scenarios.

Every situation is unique. It’s a good idea to run benchmarks ahead of time to learn the best option to use during a workload crunch versus more economical routine rendering. For example, a job with a higher-performance GPU configuration may render 50% faster but also cost slightly more overall, but the extra expense may be offset by other requirements.

More resources

To learn more, schedule a call with our team. We can help find the best solutions for each specific workload.

Last updated