Dec. 23 at 5:26 PM
$GSIT is not even serious about Plato, a chip with unknown specs or architecture to be allegedly taped out in 2027.
All we know is that it would target 12-15 W and medium to high performance for edge applications.
Note that no PIM/INT1 can compete in this range with Dataflow.
So the claim that it would be built on APU foundations is BS. It'd require a completely different architecture, not even PIM, on a pricey node (<16nm), w/ heavy licensing/IP.
Costing north of
$200M they don't have, &
$GSIT lacks the knowledge or expertise to design processors, apart from the SRAM hack that came from the MikaMonu acquisition.