Latent diffusion and Stable Diffusion

SD 1.5 encodes a $512 \times 512 \times 3$ image into a $64 \times 64 \times 4$ latent. How many times fewer values does the diffusion model process?