Page 124 of 124

Re: REV 3 - REV 5 - The beginning (ST536)

Posted: 20 Nov 2026 15:11
by exxos
Quick update on the SDRAM side of things, as there has been a fair bit of testing over the last couple of days.

I have reworked how the SDRAM side is timed. The SDRAM controller now runs slower than the original TF536 code did, which costs a little bit of speed against my previous builds, but gives it the headroom I needed. Even so, the GemBench TT-RAM score is 942%, against 847% I had with the original controller, so it is still about 11% faster than where I started.

The main thing I wanted to know was whether it would survive the fitter rearranging the logic, as that is what has been biting me for a very long time. Change one thing, the fitter shuffles everything around, and something which has nothing to do with the change stops working. That has been one of the circles I have been going around in for a long time.

So this time I deliberately forced the fitter to rearrange things.

First I checked that rebuilding the same code gives exactly the same JED, which it does, so any difference after a change really is down to the change and not the build itself. I also found that exhaustive fit mode was switched on in the project, which means the fitter picks its own limits and ignores the input limit settings. So any changes to those settings in the past may never have done anything at all.

With exhaustive mode off and the input limit forced well down, the logic stayed exactly the same, but most of the design got moved around inside the PLD, including the SDRAM state machine. That build passed GB6 and YAARTTT with freezer spray and with the board heated up.

Then I made a real code change, a small tidy up of the IDE logic, and let the fitter do its own thing as normal. Again most of the design moved, and this time even the SDRAM clock output from the PLD moved. That has passed GB6, and 7 full YAARTTT passes so far, including freezer spray and heat on the PLD and SDRAM chips.

Before that the same firmware ran 102 passes overnight on my 5.66 board, which drives the SDRAM clock from the PLD like Greg's, and 151 GB6 passes on my other board which uses the direct clock.

So that is three different fits, with the logic moved around inside the PLD each time, on both board types, hot and cold, and nothing has failed. That is the first time I have had a build survive the fitter like this, and it backs up what I have suspected for a long time, which is that the old timing simply did not have enough margin to cope with the fitter moving things around.

I will run a couple more checks to confirm that last part properly, but so far it is looking very good.

Re: REV 3 - REV 5 - The beginning (ST536)

Posted: 20 Nov 2026 15:27
by PhilC
That's a great update Chris, well done.

Let me know if you need a second test bed at some point and I'll hook the Mega up ready.

Re: REV 3 - REV 5 - The beginning (ST536)

Posted: 20 Nov 2026 16:01
by exxos
PhilC wrote: 20 Nov 2026 15:27 That's a great update Chris, well done.

Let me know if you need a second test bed at some point and I'll hook the Mega up ready.
:thumbup: will definitely need more testing likely after Christmas sometime.

Re: REV 3 - REV 5 - The beginning (ST536)

Posted: 20 Nov 2026 16:06
by exxos
Routing is still going on in the background.. All bit of a tight fit..

Currently it is on stage 39 of the changes. I really hope this FPGA works, else this new reroute will have been a colossal waste of time :lol: the saving grace here is I'm not actually routing much myself. Mostly I've just been telling off A.I. for doing stuff I told it NOT to do or not linking stuff up right etc !

Capture.PNG
Capture.PNG (256.53 KiB) Viewed 22 times

Those 2 IDE buffers I did have much further down the board to reduce the track length on the higher data bits as they were causing part of the issue. But then I had to move it back up again because there wasn't any way to actually hook it up to the power rails without starting to mess around with the SDRAM routing..

There's still several signals which are not rooted at all yet is things like reset go everywhere and not easy to root without 20+ vias currently ! So I may have to sit and stare at the thing for a while at some point very soon !

I was using 0.154mm tracks, were I started converting down to 0.127, and later moved to 0.120 as things are just getting tighter and tighter. Problem is not just the tracks is the gaps between them and everything else as well.