nVidia had some low-end GPUs in the Fermi era with mixed density chips.
Problem with mixing memory chip sizes is you have non-uniform bandwidth access. Say you have a hypothetical 256-bit card with 6x 8Gbit modules and 2x 16Gbit modules for a 10GB total running at 10GT/s.
You can access 8GB of RAM at 320GB/s but the remaining 2GB can only be accessed at 80GB/s. Might be fine on a console since you use the memory for non-graphics tasks.
Problem with mixing memory chip sizes is you have non-uniform bandwidth access. Say you have a hypothetical 256-bit card with 6x 8Gbit modules and 2x 16Gbit modules for a 10GB total running at 10GT/s.
You can access 8GB of RAM at 320GB/s but the remaining 2GB can only be accessed at 80GB/s. Might be fine on a console since you use the memory for non-graphics tasks.
Last edited: