Interesting tidbits from HDT folder regarding Oberon...
They had OPN number back in February already. Its looks familiar as well!
Is the second ID number ending in F10 supposed to be in the sequence of PCI IDs?
Seems like a lapse to stop counting in hex.
What happens when you use 16Gbps chip instead of 14Gbps one, which are used as default for theoretical value?
I suppose AMD will tell Sony or MS "Ammm...guys, you cannot do that. We put 448GB/s and 560GB/s as theoretical maximum soo...please remove 16Gbps chips and use 14Gbps ones, because we cannot have it higher then theoretical"
It can be limited by the specifications for the memory controller, or what was negotiated for the design. We don't have external reference for what AMD considers the upper limit before needing an upgrade to the involved units. There have been instances where targeting a lower range yielded area and power savings, or leaving headroom for higher clocks did the opposite (ex. Tahiti's much larger GDDR5 PHY versus Pitcairn/Hawaii).
In other cases, there may be a lack of timing settings or multipliers for a given clock, or the settings exist but AMD did not really implement working silicon for certain combinations (ex. test settings in the BIOS for infinity fabric 1:1 timings in Zen 1 could be selected and would lock up the system.)
Depending on the maximum AMD has validated the controllers for, it may leave the chips the client buys without a guarantee that they will be considered viable if parts are run past the specification agreed to. If AMD is responsible for validating good die, AMD may be able to refuse checking if the chips will fail at 16 Gbps, and the purchase agreement for chips vetted for 14 Gbps wouldn't give the customer grounds to refuse buying them.
I suppose the buyer could then take on the redundant expense of re-validating chips on their own, though discards due to failures to meet the upclock would not be AMD's problem.
Taking the chips out of spec without agreement with AMD may also weaken any claims against AMD if something like premature failure or hardware flaws are encountered later, and AMD can point to the silicon being pushed out of spec.
If the customers do their validation, perhaps it's a question of if there's a larger percentage of lost chips or if their validation is thorough enough if the speeds are significantly beyond what they're set up to test.