That depends on the packaging tech. If they use 2.5D TSV, SerDes is likely unnecessary since it can be dense enough to carry all the links as if they are on-chip 32B/64B data buses. Though having a read at the Zen 2 ISSCC presentation, there seem to be an argument against interposers (?).The savings wouldn't be also exactly 45% as you'd need to re-add the SerDes with enough bandwidth to the IO die, which would be significantly wider than what we see in Zen2 to be able to feed enough bandwidth.
Zen 2 CCD is 74 mm2, with the IFOP taking ~8% of the space, meaning ~6 mm^2. Each gives 32B read + 16B write per clock. Assuming one IFOP per two GDDR6 channels, that'd gives 8 IFOPs for ~48 mm^2, providing 384-768 GB/s aggregate at 1-2 GHz.
Seem like a considerable saving e.g. for a hypothetical 2X Navi 10 product, shaving off perhaps >100 mm2 in the compute die. It seems to be even more relevant to CDNA products, which are touted to support advanced interconnect topologies (= more SerDes).
Last edited: