CPS1/1.5 at 48MHz SDRAM

Happy new year and please, run the update script (update_all).

Two weeks ago, the memory poll results were that 20% of users of 128MB modules had issues with the CPS1.5 core. That was too many people.

I spent many hours testing the memory modules, and I summarized most of my findings here. I know that at least two SDRAM module builders are going to try to improve the modules. From the FPGA side, there is little that can be done. The approach of finding a lucky way of sending signals that seem to work for many people is just not acceptable to me.

What I can do to tackle the electrical issues around the memory is to slow down the edges of the FPGA signals. That is a solution with a strong theoretical background that fits the data. I have already proven it helps make the memory modules more reliable. The downside is that if the edges are slower, the signals need more time to toggle. Thus, 96MHz operation is off the table. 

I have re-worked the cores to operate off a 48 MHz SDRAM. I do not think accuracy has been affected. Internally, part of the core logic is still off a 96MHz clock.

I hope this update will work for much more people. Please run the update script and let me know how it goes. Please do check that your setup is correct before answering the poll. You should be running the 210101 version of the CPS1 and CPS15 cores after the update. Please note that Slam Masters/Muscle Bomber hanging up after one combat is not related to the SDRAM.

I had more updates for this week, but I have finally decided to keep it focused on this issue. As those 20% of users affected probably care about this more than any other update.

I hope to close the SDRAM chapter -at least from my side- here. I have plenty of exciting things in the oven and I plan to work on those next.