Request for Guidance on Multi-GPU Support and Hardware Recommendations for Cinematic Rendering in Unreal Engine 5.6

Dear Unreal Engine Pro Support Team,

Warm greetings. We are reaching out as part of our ongoing efforts to optimize our Unreal Engine 5.6-based cinematic rendering pipeline. We are a production studio deeply invested in creating high-fidelity, non-real-time stereoscopic 3D animated films using Unreal Engine’s Movie Render Queue. Given the scale, complexity, and resolution of our current and upcoming projects, we would greatly appreciate your expert guidance on hardware utilization—specifically with respect to multi-GPU support.

:small_blue_diamond: Project Overview and Rendering Challenges

We are currently engaged in very high quality stereoscopic 3D cinematic immersive movie production. Our render pipeline is built 100% inside Unreal Engine 5.6, and involves:

- Final Output Format:

- Left Eye: 14K resolution

- Right Eye: 14K resolution

- Stereoscopic 3D, 60 FPS, Non-Real-Time Rendering

- Project Scale & Complexity:

- Per project size exceeds 500+ GB (compressed).

- Assets include extensive environments, hyperreal character models, and physically simulated elements such as:

• Large-scale Houdini caches (e.g., crowd systems, fluid sims)

• Ray-traced reflections, global illumination, and volumetrics

• Heavy use of VDB clouds, cloth, and particle dynamics

- Rendering Workflow:

- Using Unreal Movie Render Queue with anti-aliasing, temporal sample accumulation, and full resolution renders

- Target output formats require 14K x 2 (stereo) image sequences for post-production

:small_blue_diamond: Current Hardware Usage

- Workstations with RTX 5090 (32 GB VRAM)

- We are currently utilizing RTX 5090 GPUs, but are facing memory limitations, especially with large scenes involving VDB, groom hair, and complex simulations.

- Scene crashing and performance bottlenecks are occurring during Movie Render Queue tasks when dealing with full-resolution stereo assets and dynamics.

:small_blue_diamond: Planned Upgrade Path – Query for Validation

We are planning to invest in NVIDIA RTX 6000 Ada GPUs (96 GB VRAM) and would like to know:

1. Will Unreal Engine 5.6 natively utilize dual RTX 6000 Ada GPUs installed in the same system during cinematic rendering via Movie Render Queue?

- If not, is there any recommended method to assign stereo rendering or batch sequences to separate GPUs using Unreal-supported configurations?

- Does Movie Render Queue or the rendering pipeline benefit from NVLink on dual RTX 6000 Ada cards?

2. Would it be technically and operationally better to split into two separate machines each with one RTX 6000 Ada card and run eye-specific renders independently?

3. Is the significant investment in a dual 6000 Ada GPU system justified, given our usage is exclusively within Unreal Engine’s cinematic (non-realtime) rendering path?

:small_blue_diamond: Need for Official Recommendation

We would be grateful if your engineering or technical advisory team could:

- Review our rendering pipeline use-case

- Confirm the GPU utilization limits and possibilities within Unreal 5.6

- Recommend the most efficient and scalable configuration (one powerful system with dual RTX 6000 Ada vs two separate render nodes with single GPUs)

We sincerely appreciate the incredible technology that Unreal Engine offers, and we are fully committed to pushing the boundaries of cinematic storytelling using your platform. Your expert advice on this investment decision will have a direct impact on our upcoming production capacity and delivery pipeline.

Please feel free to request any logs, render settings, or benchmarks if needed to evaluate our use-case better.

Warm regards,

Dr. Bharathy CSS

Founder & Technical Director

Fusion VR

Email: [Content removed]

Phone / Whatsapp : 9962992000

Epic Games Account Email: [Content removed]

Steps to Reproduce

Hello,

Are you using Pathtracer? If you are not using Pathtracer, then it would be best to have two machines with a single card rather than one machine with two cards. Aside from PathTracer, there are not many things that can utilize mGPU in the deferred path in Unreal Engine. The only other system of decent merit would be GPU Lightmass.

If you are using Pathtracer, we have only done internal testing using cards with NVLink. I have heard from one studio that says they got Pathtracing mGPU working without NVLink on a PCIe5 motherboard, but we have not had an opportunity to validate that internally. One thing to note is that even with working mGPU for Pathtracing, it is not a clean 2x boost in performance. Only the path tracing portions of the frame are going to be split between GPUs. Stuff like acceleration structure builds or some denoising is only happening on one GPU. That being said, a number of larger car companies utilize mGPU for their visualization rendering, but they are less concerned with price and more concerned with performance.

We generally recommend having two machines with single cards rather than an mGPU machine. It’s likely a better investment to get more machines.

Happy to help. Best of luck!

Dear Shaun,

Thank you for your detailed and insightful response to our query regarding multi-GPU support and hardware recommendations for cinematic rendering in Unreal Engine 5.6.

To clarify, we are not using Pathtracer in our current rendering workflow. Our pipeline is built entirely on Unreal Engine’s Movie Render Queue for high-resolution (14K stereo), non-real-time cinematic rendering. Your explanation regarding the limited utility of mGPU setups outside of Pathtracer and GPU Lightmass was very helpful in understanding Unreal’s current capabilities in this area.

Your suggestion of opting for two separate machines with single high-performance GPUs instead of one machine with dual GPUs aligns well with our production goals. As we continue to work on large-scale stereoscopic 3D content, stability and predictable performance are critical, and this guidance will be instrumental as we plan our upcoming hardware upgrades.

We sincerely appreciate your support and thank you once again for guiding us toward an optimal rendering setup.