>>105801899
batching will potentially use your vram more efficiency across gens, but there's a sweet spot before the batch size starts causing negative returns. its pretty easy to test, just run batches of various sizes then divide the run time by the batch count.