@peter_wade That illegal address error is a bug. We're currently hunting that one down. It would help us isolate the bug and make sure it's fixed if you could open a support ticket and attach the scene.

Please create a ticket here: https://support.posersoftware.com/

You can create a scene collection by using the File > Export > Save for sharing ...  That will create a zip file that included the scene and all the assets used by the scene. If the zip file is too big to attach to the support ticket we'll set up a link to send in the files.

This will help us make Poser better.

For a little background on GPU rendering: If a GPU render fails the device needs to be reset before Poser can render to it again. This requires Poser be restarted. If you're getting blank gray renders after a GPU render fails. Exit and restart poser to reset the GPU.

Scenes that are too big can cause this to happen. All the assets (textures and geometry) for a scene need to fit into the GPU memory all at once. GPU's can't do memory swapping the way a CPU can.That's why CPU renders have a much higher capacity for large scenes. That and most systems have much more main memory than dedicated GPU memory.

So how much GPU memory do you need? Roughly 2 GB for each figure in a scene. But, that's all subject to the exact figure. If you're using a character that has a bunch of 4K textures it will need more than 2 GB per figure. Clothes count as a figure too. Less than an actual figure but still probably 1 GB for each clothing item.

Reducing the bucket size and disabling progressive rendering on GPU will save about 200 MB on the GPU memory. So if the scene almost fits that may be enough to squeeze it in.

Other stuff running takes up some of the GPU memory too. If you're trying to maximize the GPU memory for a render close all the windows you're not using. Web browsers and anything the displays pictures will be the biggest RAM pigs. Rendering in the Render Queue will allow you to close Poser too. That will free up even more GPU memory.

What about dual GPU's? Yes but not what you were hoping for. The FireFly implementation of Cycles does use multiple GPUs. But, the entire scene must fit in each GPU's memory. So 2 8GB GPUs won't let you render a 16 GB scene. 8 GB will still be the limit.