GPU Rendering¶

GPU rendering makes it possible to use your graphics card for rendering, instead of the CPU. This can speed up rendering because modern GPUs are designed to do quite a lot of number crunching. On the other hand, they also have some limitations in rendering complex scenes, due to more limited memory, and issues with interactivity when using the same graphics card for display and rendering.

To enable GPU rendering, go into the Preferences ‣ System ‣ Cycles Render Devices, and select either CUDA, OptiX, HIP, oneAPI, or Metal. Next, you must configure each scene to use GPU rendering in Properties ‣ Render ‣ Device.

Rendering Technologies¶

Blender supports different technologies to render on the GPU depending on the particular GPU manufacturer and operating system.

CUDA – NVIDIA¶

CUDA is supported on Windows and Linux and requires a Nvidia graphics cards with compute capability 3.0 and higher. To make sure your GPU is supported, see the list of Nvidia graphics cards with the compute capabilities and supported graphics cards.

OptiX – NVIDIA¶

OptiX is supported on Windows and Linux and requires a Nvidia graphics cards with compute capability 5.0 and higher and a driver version of at least 495.89. To make sure your GPU is supported, see the list of Nvidia graphics cards.

OptiX takes advantage of hardware ray-tracing acceleration in RTX graphics cards, for improved performance.

L’acceleració de GPU per a OpenImageDenoise està disponible per a la capacitat de càlcul 7.0 i superior, que inclou totes les targetes NVIDIA RTX.

HIP – AMD¶

HIP és compatible amb Windows i Linux i requereix una targeta gràfica AMD amb arquitectura de RDNA1 o més recent. Tant les GPUs discretes com les APUs són compatibles.

Supported GPUs include:

Radeon RX 5000 Series

Radeon RX 6000 Series

Radeon RX 7000 Series

Radeon Pro W6000 Series

Minimum driver versions:

Windows: Radeon Software 21.12.1 or Radeon PRO Software 21.Q4

Linux: Radeon Software 22.10 or ROCm 5.3

Please refer to AMD’s website for more information about AMD graphics cards and their architectures.

Experimental hardware ray-tracing support is available with the most recent drivers. This can be enabled in the preferences. However there are currently known issues regarding motion blur, hair and point cloud rendering, and degenerate triangle shapes.

El dessorollat accelerat per GPU està disponible concretament a les GPU Radeon RX 6000 i Radeon RX 7000.

oneAPI – Intel¶

oneAPI és una biblioteca de càlcul compatible amb Windows i Linux i requereix una targeta gràfica Intel® Arc™ amb l’arquitectura Xe HPG. Admet l’acceleració de maquinari per al radiotraçat i el dessorollat.

Supported GPUs include:

Intel® Arc™ A-Series

Minimum driver versions:

Windows: Intel Graphics Driver XX.X.101.5518

Linux: paquet intel-level-zero-gpu 1.3.27642, normalment disponible mitjançant el paquet intel-compute-runtime XX.XX.27642.38

Please refer to Intel’s website for more information about Intel graphics cards and their architectures.

El dessorollat accelerat per GPU està disponible a totes les GPU compatibles.

Metal – Apple (macOS)¶

Metal is supported on Apple computers with Apple Silicon. macOS 13.0 or newer is required to support all features.

El dessorollat accelerat per GPU està disponible a Apple Silicon.

Limitacions¶

Path Guiding is not supported on any GPU.

Open Shading Language is only supported for OptiX, with some limitations listed in the documentation.

Frequently Asked Questions¶

Why is Blender unresponsive during rendering?¶

En les generacions de GPU més antigues, les targetes gràfiques només poden revelar o dibuixar la interfície d’usuària. Això pot fer que el Blender no respongui mentre fa el revelar. Les escenes pesades també poden fer que Blender no respongui en les GPU més noves, quan s’utilitza molta memòria o s’executen aspectors costosos, però això és generalment un problema menor.

The only complete solution for this is to use a dedicated GPU for rendering, and another for display.

Why does a scene that renders on the CPU not render on the GPU?¶

There may be multiple causes, but the most common one is that there is not enough memory on your graphics card. Typically, the GPU can only use the amount of memory that is on the GPU (see Would multiple GPUs increase available memory? for more information). This is usually much smaller than the amount of system memory the CPU can access. With CUDA, OptiX, HIP and Metal devices, if the GPU memory is full Blender will automatically try to use system memory. This has a performance impact, but will usually still result in a faster render than using CPU rendering.

Can multiple GPUs be used for rendering?¶

Yes, go to Preferences ‣ System ‣ Compute Device Panel, and configure it as you desire.

Would multiple GPUs increase available memory?¶

Typically, no, each GPU can only access its own memory.

The exception is NVIDIA GPUs connected with NVLink, where multiple GPUs can share memory at a small performance cost. This is can be enabled with Distributed Memory Across Devices in the preferences.

What renders faster?¶

This varies depending on the hardware used. Different technologies also have different compute times depending on the scene tested. For the most up to date information on the performance of different devices, browse the Blender Open Data resource.

Error Messages¶

In case of problems, be sure to install the official graphics drivers from the GPU manufacturers website, or through the package manager on Linux. The graphics drivers provided by the computer manufacturer can sometimes be outdated or incomplete.

Error: Out of memory¶

This usually means there is not enough memory to store the scene for use by the GPU.

Nota

One way to reduce memory usage is by using smaller resolution textures. For example, 8k, 4k, 2k, and 1k image textures take up respectively 256MB, 64MB, 16MB and 4MB of memory.

The NVIDIA OpenGL driver lost connection with the display driver¶

If a GPU is used for both display and rendering, Windows has a limit on the time the GPU can do render computations. If you have a particularly heavy scene, Cycles can take up too much GPU time. Reducing Tile Size in the Performance panel may alleviate the issue, but the only real solution is to use separate graphics cards for display and rendering.

Una altra solució pot ser augmentar el temps d’espera, encara que això farà que la interfície d’usuària sigui menys responsiva a l’hora de revelar escenes pesades.`Més info aquí <https://learn.microsoft.com/en-us/windows-hardware/drivers/display/timeout-detection-and- recovery>`__.

Unsupported GNU version¶

On Linux, depending on your GCC version you might get this error. See the Nvidia CUDA Installation Guide for Linux for a list of supported GCC versions. There are two possible solutions to this error:

Use an alternate compiler

If you have an older GCC installed that is compatible with the installed CUDA toolkit version, then you can use it instead of the default compiler. This is done by setting the CYCLES_CUDA_EXTRA_CFLAGS environment variable when starting Blender.

Launch Blender from the command line as follows:

CYCLES_CUDA_EXTRA_CFLAGS="-ccbin gcc-x.x" blender

(Substitute the name or path of the compatible GCC compiler).

Remove compatibility checks

Si el de més amunt no té èxit, elimineu la línia següent a /usr/local/cuda/include/host_config.h:

#error -- unsupported GNU version! gcc x.x and up are not supported!

This will allow Cycles to successfully compile the CUDA rendering kernel the first time it attempts to use your GPU for rendering. Once the kernel is built successfully, you can launch Blender as you normally would and the CUDA kernel will still be used for rendering.

CUDA Error: Kernel compilation failed¶

This error may happen if you have a new NVIDIA graphics card that is not yet supported by the Blender version and CUDA toolkit you have installed. In this case Blender may try to dynamically build a kernel for your graphics card and fail.

In this case you can:

Check if the latest Blender version (official or experimental builds) supports your graphics card.
If you build Blender yourself, try to download and install a newer CUDA developer toolkit.

Normally users do not need to install the CUDA toolkit as Blender comes with precompiled kernels.