Sure, just offload to system RAM, and don't use your system driver's fallback, but a specific implementation like MultiGPU.

It won't speed it up, but using a quantization that fits in VRAM will prevent the offload penalty.