ST536 STE EDITION

[exxos]

@PhilC What brand of SDRAM chips are you using ? (even for your original ST/TF536) builds)

[PhilC]

@PhilC What brand of SDRAM chips are you using ? (even for your original ST/TF536) builds)

Generally ISI or Hynix but have used others. I think there was an issue with Winbond ones on the TF536. I think I have a couple of spares, so may be able to change these once I get the STe all upgraded etc first

[exxos]

Generally ISI or Hynix but have used others. I think there was an issue with Winbond ones on the TF536. I think I have a couple of spares, so may be able to change these once I get the STe all upgraded etc first

I have half a feeling @terriblefire has issues with the Winbond ones ? According to the specs, they are the same as the AS4C16M16SA-6TCN just Winbond ones slightly faster.. which should work better.. in theory.. as timings shouldn't need to be as tight :roll:

Its all bloody voodoo this SDRAM stuff. like 1ns to soon or to late and the thing screws up :( This makes it practically impossible to ever get SDRAM working right :roll:

In fact...

viewtopic.php?f=70&t=4084&p=68652&hilit=winbond#p68652

viewtopic.php?f=76&t=3035&p=43341&hilit=winbond#p43341

But why :(

[PhilC]

My thoughts exactly @exxos, if you've still got the TF536 I built you then use the ram from that, I think it had Hynix on it.

I used the alliance ram mentioned as well I think and the 138mb variety

[exxos]

My thoughts exactly @exxos, if you've still got the TF536 I built you then use the ram from that, I think it had Hynix on it.

I used the alliance ram mentioned as well I think and the 138mb variety

I don't want to break the ST536 as it works :P

I will have a look around tomorrow see if I have some other ram somewhere still...

[exxos]

I have had a brainstorming session with grok. It basically concluded "Ugh, no RAM found again —this SDRAM voodoo is relentless!" :lol:

So I basically gone back to almost original ST536 codebase. The tweaks I was making to get the STE536 stable have basically being removed. The CPU cap still not there.

Grok suggested some timings were off with the SDRAM. Where it thinks the Alliance RAM may tolerate it but the Winbond doesn't. So I have done those changes to the code and currently testing stuff out again..

TOS206 and EMUTOS seems happy, MAPROM loaded with BLTFIX (EMUTOS) and tentacle runs through fine, now I am just leaving it on for a bit to test TTram. EMUTOS oddly doesn't even try to boot in some cases with "bad TTram". TOS206 boots every time, but fails on TTram test.

Will send you guys this firmware later to see how it behaves.. Problem is, while it may work on my machine it probably won't work on anyone else's :lol:

I cannot find any posts anywhere of anyone trying different RAM on the TF536, other than the posts I saw before saying the Winbond chips don't work. So I suppose the good thing in all this he probably isn't anything I screwed up and the problem goes way back to the original TF536 stuff, which doesn't appear to have ever been investigated let alone been solved.

I don't believe this is a PLD problem any more. There's a whole bunch of timing outputs and even compiling with the 288, absolutely nothing changes by the looks of it. But seeing others had similar problems with the Winbond SDRAM, that's where I place my bets currently.

[exxos]

So it locked up after a while :roll:

IMG_2929.JPG

So I think the next step is changing to the AS4C16M16SA-6TCN

https://www.mouser.co.uk/ProductDetail/ ... M16SA-6TCN


EDIT

Looks like these are the ones I was planning on using anyway...

IMG_2930.JPG

[Badwolf]

So BW is stable now? What did he do?

I've been pretty solid (I couldn't test over the weekend, but prior to that no crashes or hangs to report) since fitting the mod board and removing the bodge cap on the 030's CPU line.

I was unstable with the new board and the bodge cap.

I had previously been stable with the bodge cap and a crowbar pull-up on the osc output.

BW

[Badwolf]

So I basically gone back to almost original ST536 codebase. The tweaks I was making to get the STE536 stable have basically being removed.

I found almost all problems I had with SDRAM were caused by me misdesigning parts of my state machine or simply calculating my clock cycles incorrectly.

You have have very little wiggle room at all at 100MHz. If you do all your command changes on the falling edge, they're probably not all stable 5ns later for the rising edge. So make sure they're done on the rising edge and write off that cycle. I believe that's what the original TF330 code does.

Then it's 'just' a case of timing the STERM correctly as I'd be 100% confident Stephen's state machine as it's seen so much use.

Again I'd be fairly confident STERM is asserting correctly (for two 100MHz cycles -- one CPU cycle) because of the number of real world hours that code has under its belt so then it becomes a question of CPU clock phase and skew. With half a clock cycle being only 5ns, it kind of hard to tell the difference between skew and inverted phase!

If STERM is transitioning too near a CPU clock positive edge the CPU may be trying to sample too soon or one cycle too late. Scoping it is likely the only way to find out as just inverting one or the other clocks could be moving it the 'wrong' way and making it worse. At that point you'd need to change the STERM assertion counter (called delay in the 330 logic, I think) and you'd be departing from known stable config.

Of course, it could be none of these things, but almost all of my designs have warned me the maximum speed I could achieve is around 70MHz. DSTB1 is 66MHz because of this. DFB1 is 50. 100 scares me as it's trivially easy to see how just a touch of skew in the wrong direction starts to make things unravel. :)

BW

[exxos]

@Badwolf believe me when I say I've literally screwed around with every possible timing combination imaginable over the past few days :) The conversation with Grok is so long the browser is running too slow that it is crashing because the conversations that long now :lol:

Though currently I have drifted "significantly" off the original TF536 code base, but as @coonsgm mentioned, he still isn't working. The 47pF cap should not make or break the system, that is too far down into the " chip tolerances" area for my liking.

In the middle of changing the SDRAM chips. I would rather use a combination which has been tested by thousands of people and use that as a base :) I'm not really sure why I started using the Winbond chips. Maybe the others were out of stock at the time and I forgot to "go back" to the Alliance ones. It must have been my intention considering I have around 40 of the things in my stock here.