I know that you can't mix different spec cards for SLI gaming, but I suspect that when it comes to 3D rendering, the renderer will simply divide the work up automatically.

Choosing the optimum tile size is important for efficient usage of the core's in our processors, I suspect that the tiles will simply be distributed to the difference in power of graphic cards. So the key to getting it to work efficiently (I suspect) is finding a tile size which allows the renderer to effortlessly divide the work to graphics cards with different cuda core sizes.

In my e specific case, I have a GTX 970 and a TITAN X; can I use them both for rendering?

Yes, you can use multiple GPUs for cycles rendering even if they are not the same kind (as long as they are not a mix of Nvida and AMD/ATI based cards).

Adding additional GPUs is the easiest way to speed rendering in cylces. Two GPUs will render an image in almost half the time it would take to render with only one. 4 GPUs will cut rendering times to almost 1/4...

In the Compute Device section of the User preferences you can choose the GPUs to be used.

When rendering in cycles, each GPU will render one tile and move on to the next unrendered tile until there are no more tiles left to render. Blender will render simultaneously as many tiles as there are GPUs available.

Each GPU will render it's assigned tiles, but if one of the GPUs is substantially slower it will just not render as may tiles within a frame as a faster GPU, and that might even cause the rest of the GPUs to idle until the rendering of the last tile is done.

The available memory for rendering will not be the sum of all the memory installed on the cards, but will be limited to the vRAM of the smallest card. In other words: adding a new GPU with 8GB to a system that has an old GPU with only 2GB, will make the new GPU only use 2GB of vRAM.

To find the optimal tile size you'll have to experiment a bit, as it varies depending on the the geometry, the size and complexity of the textures used. A good starting size is 256x256 but I suspect the cards you mention might be able to deal with larger tiles.

Other considerations:

Keep in mind that GPUs that are connected to a monitor (or monitors) will not work as efficiently as those that are just used for rendering. Refreshing the screen(s) or using other applications might also take away from the resources available for rendering.

If one of the GPUs is substantially slower or has a very limited amount of vRAM, it might be more efficient to have the less powerful card just take care of refreshing the monitors and not take part in the rendering

Rendering using GPU is generally faster than rendering on CPU, but there are still a lot of processes (ie. physics, particles, baking, etc) that are not dependent on the GPU, so having two or more GPUs will not make simulations cache or bake any faster.

If your scene has very large textures or needs to use more RAM than what's available on the GPUs, your scene will fail and not render, no matter how many cards you put in. For very big scenes with large textures that require a olt of RAM, you might still be limited to rendering on CPU.

Check that your power supply is adequate to handle both cards.

With two or GPUs your machine will generate a lot more heat. Make sure you have an adequate cooling system. If the machine overheats it will slow down and eventually shutdown or fail.

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.