On 8/30/22 18:01, Tim Harvey wrote: Hi,
Marek, Might I ask how you ran into the issue
There were a couple of boards which were unstable and kept crashing at runtime, often when using GPU. Either the machine locked up completely or there were rendering artifacts. Eventually also memtester triggered on errors during walking ones test.
and how you came up with the new numbers?
The hardware experts ran simulation of the bus in I think hyperlynx and come up with new values for the RPA.
Were the new register settings obtained by simply using a newver version of the NXP DDR tool or a new version of the RPA Spreadsheet used with the tool or some other means?
Nope, the RPA spreadsheet version was the same, it was the simulation the hardware people ran which exposed the sub-optimal inputs into the RPA.
