The same compute resources, but not the same VRAM. It will more or less get you the same tokens per second as a ~13B model but should have significantly "higher

The same compute resources, but not the same VRAM. It will more or less get you the same tokens per second as a ~13B model but should have significantly "higher quality output" than a single 13B model.