Optimal settings for 2 GPUs in LM Studio

Hello everybody. I've got a 5070ti and a 5060ti both 16GB in my system which is a 5900X and 64GB DDR4 RAM. I'm trying to run some 16-18GB models like Qwen, Cydonia, Skyfall.

I'm having problems utilising the VRAM I have to get the best usage out of it. LM Studio sees the 32GB VRAM but regardless of if I use Tensor parallelism, Split evenly or Priority order I always get an error after waiting for about 5 minutes for the model to load.

The pattern is always the same: The loading progress bar for the model starts pff quickly then crawls in the last 5-10%. Then I get an error reporting that the model couldn't load.

Does anybody have any tips for optimal settings to get the best out of my system? I know that having 2 GPUs doesn't magically mean you have double the memory and there's caveats. But nevertheless I've also read that LM Studio does have the capability to leverage those 2 GPUs to improve speed.

(EDIT) I should add that I've been trying context lengths of 16384, 32768 which LM Studio is saying will use 17 GB of VRAM so well within the reported 32GB I have. I've even had it working occasionally but most of the time the model fails to load.

Many thanks for any help.