How do I run Unreal Movie Render Queue on a cloud GPU?
This guide is written by the CloudGPU team. CloudGPU is an independent GPU rental service and is not affiliated with Epic Games.
Short answer: Save a Level Sequence and a Movie Pipeline queue or config asset on your own machine, open an RTX 4090 or RTX 5090 instance, put your Unreal Engine install and project on it, then run the editor binary in -game mode with -MoviePipelineConfig=... and -RenderOffscreen inside tmux. Pull the frames back and destroy the instance. Billing is per minute.
Why render MRQ on a rented GPU
Movie Render Queue is where your Unreal cinematic becomes final frames: high temporal sample counts, high-resolution tiles, EXR passes, sometimes the Path Tracer. It’s also where a working laptop or a mid-range desktop slows to a crawl and becomes unusable for anything else.
Renting a GPU lets you keep building on your own machine while an RTX 4090 or RTX 5090 runs the queue elsewhere. CloudGPU bills by the minute, so a render session lasts as long as the render.
What you need
- A CloudGPU account with balance (card or USDT; see /payments)
- Your Unreal Engine project, with a Level Sequence and a saved Movie Pipeline queue or config preset asset
- Your own Unreal Engine install/account (Epic account sign-in or your own source/installed build)
- An SSH client (
ssh,rsync/scp) - Enough free space on the instance for the engine, the project, caches and frames (see below)
Pick a GPU tier and open an instance
Go to /deploy and pick an RTX 4090 or RTX 5090 tier. Which cards can be started right now is shown there. Choose based on VRAM headroom: high-resolution output, heavy Nanite geometry, large virtual textures and the Path Tracer all use VRAM.
Add your SSH public key when deploying. Never paste private keys into the dashboard or a shared document.
Bring your own Unreal Engine install/account
CloudGPU provides the GPU machine; Unreal Engine is not bundled. Options:
- Linux engine download: Epic provides Linux engine downloads to signed-in Epic accounts, or you can build from source through your linked GitHub access. Copy or build the engine on the instance (for example
~/UE_5.x). - Your own engine build: if your studio uses a custom engine,
rsyncthe installed build directly.
Match the engine version exactly to your project to avoid an upgrade prompt or recompiles. Your use of Unreal Engine stays between you and Epic’s terms.
Plan space before you start
Plan free space for four things:
| Item | Notes |
|---|---|
| Engine | An installed UE5 build is large; source builds are larger still |
| Project | Content + plugins |
| Derived Data Cache (DDC) | Shader and asset caches build on first run. Keep them on the instance while it is running |
| Output frames | EXR sequences at 4K add up quickly; multiply frame count × passes × size per frame |
Check free space and the GPU after connecting:
df -h
nvidia-smi
Move the project to the instance
Pick whatever you already use:
# rsync from your machine (skip local caches)
rsync -avz --progress \
--exclude 'Saved/' --exclude 'Intermediate/' --exclude 'DerivedDataCache/' \
./MyProject/ user@<instance-host>:~/projects/MyProject/
Alternatives: git clone (with Git LFS), a Perforce client sync, or upload a zip and extract. Don’t ship Intermediate/, Saved/ or local DDC, because they’re machine-specific.
Prepare the render in the editor (locally)
Do this on your own machine before uploading:
- Open Window → Cinematics → Movie Render Queue.
- Add your Level Sequence; set output format (EXR / PNG, and others where available), resolution, anti-aliasing samples, and the output directory (use a relative path like
{project_dir}/Saved/MovieRenders/). - Save the queue as an asset (for example
/Game/Cinematics/MRQ_FinalQueue), or save the job’s settings as a config preset (for example/Game/Cinematics/MRQ_Final4K). - Commit or sync so the asset is in what you upload.
Run MRQ headless with -RenderOffscreen
There’s no monitor on the instance, so you run the editor binary in game mode and tell it to render offscreen. Flags can differ between engine versions, so check them against Epic’s documentation for the version you use.
~/UE_5.x/Engine/Binaries/Linux/UnrealEditor \
~/projects/MyProject/MyProject.uproject \
-game \
-MoviePipelineConfig="/Game/Cinematics/MRQ_FinalQueue" \
-RenderOffscreen \
-Unattended -NoSplash -NoSound \
-log -stdout -FullStdOutLogOutput \
2>&1 | tee ~/mrq_render.log
Notes:
-MoviePipelineConfigaccepts a saved queue asset. To render one sequence with a config preset instead, pass the map and-LevelSequence="/Game/Cinematics/Shot010"together with-MoviePipelineConfig="/Game/Cinematics/MRQ_Final4K".-RenderOffscreenlets the renderer run without a window or display server.- Run long renders inside
tmuxorscreenso a dropped SSH connection doesn’t kill the job:
tmux new -s mrq
# run the command above, then detach with Ctrl+b d
tmux attach -t mrq # later
The first run compiles shaders, so it takes longer than later runs while the same instance is running.
Watch progress, pull frames back
# progress
tail -f ~/mrq_render.log | grep -i -E "MoviePipeline|Error|Warning"
watch -n 5 nvidia-smi
# pull frames to your machine
rsync -avz --progress \
user@<instance-host>:~/projects/MyProject/Saved/MovieRenders/ ./renders/
For long sequences, pull in batches while the render continues.
If the render fails
| Symptom | Likely fix |
|---|---|
| Crash on start mentioning display / X11 / Wayland | Confirm -RenderOffscreen is present; don’t use -windowed without a display |
| Very long first render | Shader compilation; keep the DDC on the instance and render the rest on the same instance |
| “Asset not found” for queue/config | The path must be the asset path (/Game/...), not a file path; confirm the asset was uploaded |
| Out-of-memory / device lost | Lower resolution or tile count, reduce the texture pool, or choose the tier with more VRAM |
| Path Tracer / hardware ray tracing not available | Check your UE version’s ray tracing support for your OS and RHI |
| Plugin missing on Linux | Some marketplace plugins ship Windows-only binaries; rebuild the plugin for Linux or render those shots elsewhere |
| Render stops when SSH drops | Use tmux/screen or nohup |
Before you destroy the instance
Copy the frames and logs to your own machine first. Destroying the instance stops billing, and you should treat anything left on it as gone. See save your work before destroying a cloud GPU.
Next
- Show the build live in a browser: Pixel Streaming on a cloud GPU
- Mix UE greybox renders with AI video: Cloud GPU for AI video production
- Overview: Cloud GPU for Unreal Engine
Open a GPU · Card or USDT; see /payments.