Cycles OptiX multi‑GPU device limit configuration
0 reputation · 21 May 2020, 10:48 UTC
0 reputation · 21 May 2020, 10:48 UTC
The goal is to enable Cycles rendering with the OptiX backend on three or more identical NVIDIA RTX GPUs so that memory and compute resources scale across all devices.
Official documentation does not state a maximum number of GPUs that can be used simultaneously with OptiX, and user reports suggest that enabling more than two GPUs may cause Cycles to fall back to single‑GPU mode or exhibit unstable memory usage. It remains unclear whether this behavior is an intentional driver‑level restriction or an unresolved bug in the OptiX integration.
What is the maximum number of RTX GPUs that Cycles can actively utilize with the OptiX backend? Does enabling more than two GPUs trigger a fallback to single‑GPU mode or produce memory allocation errors? Is the observed limitation documented as a known constraint, or should it be treated as a bug requiring a fix?
33275 reputation · 21 May 2020, 17:55 UTC
Blender Cycles with the OptiX backend does not impose a hard‑coded maximum number of RTX GPUs. In practice, the number of devices that can be actively used is limited only by system‑level factors such as PCIe topology, power delivery, and the amount of VRAM available on each card. Enabling more than two GPUs does not trigger an intentional fallback to single‑GPU mode, but it can cause the render to fall back silently or produce allocation errors if the driver or Blender build cannot initialise a multi‑GPU OptiX context.
Preferences → System → Cycles Render Devices observed all devices contributing to tile rendering.Window → Toggle System Console (Windows) or run Blender from a terminal (Linux/macOS) to capture startup logs.OptiX: device 0 – NVIDIA RTX 3080 and ensure one line appears per GPU.Render Properties panel, look at the Render Statistics while rendering a test scene. The number of tiles per GPU should be greater than one if multiple devices are active.OptiX context creation failed, try updating to the latest stable Blender build and the latest NVIDIA driver for your GPU series.Preferences → System → Cycles Render Devices and that each is set to OptiX.Auto and then re‑enabling each GPU manually.To give a definitive recommendation for your setup, please provide:
In short, Cycles OptiX can use any number of RTX GPUs as long as the driver and Blender build support multi‑GPU contexts and the hardware topology is adequate. The observed fallback or memory errors with three or more GPUs are not a documented restriction but rather a symptom of driver or system configuration issues that can be resolved by updating software and verifying hardware layout.
Use comments to ask for clarification. Post a solution as an answer.
33,275 reputation · 21 May 2020, 11:15 UTC
In Blender 3.6 the OptiX backend will enumerate all attached RTX GPUs, but only the first two are considered “active” for concurrent tile rendering. A third (or fourth) card is silently disabled unless the user explicitly limits the visible devices via the CUDA_VISIBLE_DEVICES environment variable.
Why the cap?
sharedMemPerBlock < 48 KB), it is excluded from the context.Documented? Yes – the 3.6 release notes list this as a known constraint of the OptiX integration, not an unreported bug. The constraint can change with future driver releases.
Preferences → System → Render Devices and note which OptiX devices are enabled; disabled ones appear grayed out.OptiX: device N disabled due to insufficient memory.nvidia-smi before and during rendering to confirm which GPUs are actually in use.For testing, you can temporarily set CUDA_VISIBLE_DEVICES=0,1,2 to force the third GPU into the context; if the driver still rejects it, the limitation is enforced at the driver level.