Unfortunately I checked, SLI doesnt work for this situation. Because the program loading the LLM uses CUDA library, which doesnt account for/takes advantage of SLI for this purpose at least.