I suspect these problems to be related to initialization of the SDRAM.
Our (free) SDRAM controller does not include the initialization sequence to the SDRAM as recommended in the datasheet. We will add this soon, but I suspect the problems you’re having are due to the lack of init.
If you are pressed for time, you can add the sequence initialization to the SDRAM controller. It would only require the addition of a few states and the sequence is documented in the Micron SDRAM datasheet.
This init sequence is included in the latest release of our sample SDRAM controller code (RAMTester). FrontPanel 3.0.9 should include this in the source code.
I was getting failures with 3.0.8 and the pre-built .bit RAMtester file for the XEM3005. I’d get 2-3 failures out of 100 with the old Verilog code. It appears this was indeed due to the lack of proper SDRAM init (as per Micron data sheet). With 3.0.9 I haven’t seen any errors (at 100MHz) after several runs.
The fastest I can reliably get RAMTester to pass is with an (approx.) 155MHz clock, at room temp, in default system sync. mode. The SI isn’t bad on the DRAM_CLK (scope measured, 500MHz BW) even on the BRK board pin. I know from past experience that a MicroBlaze will run at 75MHz in a S3E, so a 2x DRAM clock just might work…
Perhaps we need to start an Over-Clockers forum with folks using liquid coolers and copper heatsinks. Something tells me Zalman doesn’t really recognize the XEM market yet, though.
When running RAMTester.exe after loading the bitstream through frontpanel, I get the following error message:
C:\Program Files\Opal Kelly\FrontPanel\Samples\RAMTester>RAMTester.exe
---- Opal Kelly ---- RAMTester Application v1.0 ----
FrontPanel DLL could not be loaded.