Graphic errors / instability with Indivision AGA MK3

Caution: Non registered users only see threads and messages in the currently selected language, which is determined by their browser Please create an account and log in to see all content by default. This is a limitation of the forum software.
Don't Panic. Please wash hands.
  • Hi!


    I recently received my Indivision AGA MK3, along with an ACA1221lc. Installation went smooth, and at first, everything seemed fine. But I soon noticed a stability issue that manifests in graphics errors (see attached image) and occasional crashes.


    First, a few words about my setup:

    - Amiga 1200 Rev. 2B, recently recapped, with timing fix (E123C and E125C removed; E121C and E122C weren't present in the first place).

    - Indivision AGA MK3

    - ACA1221lc

    - iComp RTC module, installed on the mainboard (not on the ACA1221lc)

    - Internal floppy, external floppy, and external Gotek

    - IDE-CF adapter with 4GB Transcend CF card

    - Allnet ALL0142+ PCMCIA Network card

    - An old Blaupunkt flat TV


    I am using the SuperHighRes Interlaced (1280x512) mode via the digital output. I created two 1920x1080 VGA modes (normal and shres) and disabled all presets in the Indivision configuration tool, so it would always choose one of these two.


    I first noticed problems when I was copying a backup of my Workbench partition to my network drive. I suspected the ACA121lc, so I removed it and tried again, but after a few minutes the problem was back. I then removed the Indivision and tried again. SuperHighRes Interlaced is almost unbearable via SCART, but now the errors did not occur. (With the Indivision present, the same errors were visible over SCART too.)

    I tried this with another mainboard (1.4D, also recently recapped and no timing fix necessary since the offending capacitors had apparently never been populated in the first place) but that did not change anything. I also tried different power supplies: An old heavy A500 power supply, the much lighter black power supply that came with the A1200, and a new Keelog Amiga power supply. This did not seem to make any difference either. What did have a slight effect though was activating the CCKLine Pull-Up in the advanced settings. The graphics errors did not vanish completely but were noticeably reduced.


    I also noticed that the Indivision gets rather hot. Not sure if that is normal or cause for concern.


    I could not (yet?) reproduce the problem in normal HiRes (640x256) mode. I should also note that the problem does not always occur reliably, but it seems that system load (i.E. copying lots of files over the network) increases the likelyhood.


    Any ideas what the cause could be and what I could do to fix this? Please let me know if you need further information and/or measurements. (I have a multimeter and an oscilloscope if that helps, but I would need instructions on what to measure.)


    Regards

    Gert

  • Temperature is not a problem - 63 degrees Celsius is absolutely fine for Indivision AGA MK3. If I remember right, timing simulations of the FPGA show a temp limit of well over 80 degrees where timing gets critical.


    However, the error you're seeing is not caused by Indivision AGA MK3, but it's in the Amiga chipset, namely a timing problem when transferring data from chip ram into Lisa using the AGA graphics modes (so-called "double CAS" modes). Your observation:

    I could not (yet?) reproduce the problem in normal HiRes (640x256) mode.

    ...supports that, as normal Hires modes do not use double-CAS. If you have a monitor that can display such a double-CAS mode on the RGB output of the A1200, you could verify that you'll see the exact same problem without Indivision AGA MK3 - if you want to take that time :-)


    I'd try the CCK capacitor and the CCK pullup at the same time, that should fix it. Using "random other power supplies" does not tell you anything about the PSU being a cause, as only the CA-PSU (yes, our product) complies with all Commodore specifications. All other 2nd-source power supplies for the Amiga are just tinkering some off-the-shelve PSU chassis with an Amiga-connector, but that does not make them suitable for an Amiga. This specifically includes the Keelog PSU: It is based on a MeanWell chassis, which violates three out of three key specifications from Commodore. I highly recommend not to use it on any Amiga, but to send it back for a refund. It has been sold to you with false advertising.


    Read all about the PSU topic in our FAQ.

  • Hi Jens,


    thanks for the quick reply!


    Good to know that the temperature is not critical. It will certainly increase once I close the case, but probably not up to 80°C.


    Okay, so maybe I just didn't have enough patience for the problem to appear without the Indivision. I will try that again just to make sure. If I understand you correctly, this could also happen on a stock Amiga without any extensions?


    While the CCK pullup shows some improvement, activating the CCK capacitor instantly locks up the Amiga.


    Bummer about the PSU - I really tried to get a good one, but didn't know about your PSU FAQ.

    This one even has a voltage display, and it actually does show different values depending on what is connected:

    5.05V/12.0V/-11.9V with nothing connected

    5.05V/12.6V/-12.7V with the naked 1D4 board

    5.05V/13.0V/-13.2V with the 2B and all the peripherals

    So it seems it does compensate in some way. Suspicious, however, that it compensates the -12V line about as much as the +12V line although the former should have much less load on it than the latter, right? I think I'll try to measure the actual voltages in the Amiga on the weekend.

  • So it seems it does compensate in some way.

    Yes, at the input of the cable. It does not know about the voltage drop on the cable, it does not know about the voltage drop in the input filter, and it raises the 12V beyond specification. Further, such a display normally introduces a voltage drop on it's own. In short: AVOID. Do not continue to use this one, as it's putting your classic gear in danger.


    activating the CCK capacitor instantly locks up the Amiga.

    That's an interesting one that we never encountered before. Would be interesting to see if that continues to happen after you've resolved the known-bad-PSU issue.

  • Okay, here's a quick update on the situation.


    I have measured the voltages with some of my PSUs and the Keelog is, disappointingly, indeed not that great. While it does stay within the official specifications (+/- 5% for 5V and +/- 10% for +/- 12V), it gets dangerously low on the 5V line (between 4.8 and 4.9), which is probably fine for an unmodified Amiga but not so great if you have any modern extensions. I haven't measured the ripple yet, but I'll let you know how that turns out.


    I am currently using one of my original Amiga PSUs, and interestingly, with the CCK pullup enabled, the graphics glitches are still quite bad for the first minute or so after I power on the machine when it had been off for a while - but after 2 minutes they are completely gone.

    My CA-PSU should arrive tomorrow, so I'm curious to see if that'll improve matters further.


    Just to make sure - the points indicated in the picture are the ones that you recommend for measuring the voltage/ripple, right?

  • (+/- 5% for 5V and +/- 10% for +/- 12V), it gets dangerously low on the 5V line (between 4.8 and 4.9), which is probably fine for an unmodified Amiga but not so great if you have any modern extensions.

    Where did you get those specs from? Is that the Keelog spec? If so, it's just violating the Commodore spec, as the voltage deviation is only allowed to be 1% at full rated load, and it may be higher at partial load. The original Commodore spec is linked to from the PSU FAQ page.


    Just to make sure - the points indicated in the picture are the ones that you recommend for measuring the voltage/ripple, right?

    Yes, those are fine. The interesting part here is that you don't have a current-compensated coil, but individual ferrite beads. I do have a few boards that have this kind of input filter, but they are from the Commodore Braunschweig lot that I got.. years ago. I always figured that these are rate and not common in the field. Interesting to see that you have one of those!

  • I got the specs from the A500_Power_Supply_Specification.pdf linked on your PSU FAQ page:


    Which document are you referring to?


    Interesting - both boards that I have (1.4D and 2B) have the individual ferrite beads. I didn't know that was special.

  • Which document are you referring to?

    The exact document you're quoting. However, you are making the same mistake as all those amateur-type tinker-PSU vendors who are only looking at the paragraph you're showing. Look a little further, point 2.3.5, "initial setting": The PSU is to be loaded with nominal load (aka "full amperage") and then adjusted to a range between 4.95V and 5.05V - this is 1% precision at full load, and this is what breaks the neck of pretty much any PSU at higher loads. Only the CA-PSU holds up this spec at "nominal load", and remains within all other specs at light loads.

    Interesting - both boards that I have (1.4D and 2B) have the individual ferrite beads. I didn't know that was special.

    It's nothing to worry about: The voltage drop on these ferrites is near-identical to that of the current-compensated coil, so whatever you're measuring/setting, it will end up "just right" at the chips.

  • At least two out of three of the Amiga 1200 boards I had, came with ferrite beads. Rev 1d.4 and 2b for sure. Not sure about the 1d.1. To me, it's a surprise that they would be considered rare. On my reAmiga's I used a ferrite coil.

  • To me, it's a surprise that they would be considered rare.

    Maybe it's a PAL/NTSC thing? All PAL boards I have are using the toroid coil, and the ferrites are on NTSC boards. Though looking at fnord's picture, that appears to be a PAL board...

  • However, you are making the same mistake as all those amateur-type tinker-PSU vendors who are only looking at the paragraph you're showing. Look a little further, point 2.3.5, "initial setting": The PSU is to be loaded with nominal load (aka "full amperage") and then adjusted to a range between 4.95V and 5.05V - this is 1% precision at full load, and this is what breaks the neck of pretty much any PSU at higher loads.

    Ah, I see; thank you for explaining.

    Let me try to recap to see if I understand it correctly now.


    I had assumed that the worst case regulation specifies what the Amiga must be able to handle under any circumstances (I still think that, but it needs clarification; see below), and the initial setting only describes the optimal setting that is to be calibrated at the factory (but is allowed to get slightly worse over time).

    But that's only half the truth. In reality, the actual current draw (and as a result, also the voltage) obviously varies depending on what the Amiga is currently doing, plus factors like ripple, noise, temperature, etc.. The crucial point that I was missing is: Calibrating the "initial setting" correctly is necessary to ensure that the actual values, (including high frequency noise, ripple and whatnot) never leave the worst case window.

    In addition to that, using 3.3V regulators that need at least 4.85V input voltage at all time effectively tightens this window from Commodore's original 5V +/-5% tolerance down to 5V -3%/+5%.


    Did I get that right?

  • Did I get that right?

    No, you didn't - that's not the way you read such a specification. Unless there is no special condition applied to a number, it is to be kept under all circumstances. The "initial setting" spec is a clear spec for "full load", and the time that this spec comes from is when 3A PSUs were standard with compact Amigas.


    The way how to read these specs is that none of the given limits may be exceeded at any time. So the result is that the +/-5% spec "at any load" must be *combined* with the +/-1% spec at full load. Further, I'm not the one who introduced LM1117-style low-drop regulators to the world of Amiga - that was E3B with their "Subway" product. Ever since then, pretty much all hardware vendors (not just E3B and iComp) have used these regulators, and it was fine while the PSUs were not three decades old. And it is still fine today if you assume that the target system is well-maintained. However, this assumption is not always valid.


    I have changed to ultra-low drop regulatros late last year (not for the flicker fixers yet!), just to reduce the influence of a too-low input voltage. This means that the 3.3V part of new accelerators (which are not yet out) would in theory already work with 3.9V supply voltage. This will not solve all of the problems - in fact, it may even introduce new problems that I don't yet know about. One thing is for sure: CPUs are sensitive to voltage fluctuations, and the faster they are, the more problematic it gets. The new ACA1234 will therefore do a local voltage measurement and warn if there is a problem to be expected.


    Back to PSUs: When you change the amperage on 5V (which is what pretty much every PSU vendor does), you are effectively changing the whole specification, However, it is definitely not allowed to cherry-pick a single Commodore specification and ignore the rest. On the contrary: You must be able to deliver 1% precision at the original 2.5A, 3A and 4.5A "nominal" loads if you really want to replace all the original PSUs (after all, that's what other vendors are promising, right?). Instead of cherry-picking some specs and accept that I'm violating pretty much all the rest, I chose to go all the way and really fulfill all specifications under worst-possible conditions. The result is numerous reports of customers solving decade-old problems with their systems: PPC users who have been able to close their case after over 20 years, ACA1221lc users who can overclock to 40MHz with no crash in days of uninterrupted operation and Indivision AGA MK3 users who resolve startup- and pixel flickering problems. As a side effect, the CA-PSU is much more efficient, so users are even saving on electricity.

  • I never said you could cherry-pick your specs. I work in the optics industry and have had way more discussions about specs and how to interpret them than I care to remember. It goes without saying that all given specs have to be met at the same time. (Unless explicitly noted otherwise, of course.)

    But apparently, the words "initial setting" seem to lead some people to believe that it is just an initial setting, as opposed to a spec that has to be met at all times whenever a nominal load is present.

    Thank you for the clarification.


    I'm looking forward to watching your talk "Designing a better Amiga power supply".

  • Okay, so my CA-PSU arrived today.


    With the CCK pullup disabled, the problem still occurs, but I haven't been able to reproduce it with the pullup enabled until now, so that's a good sign. The CCK capacitance still locks up the machine, though. (I guess I'll just leave that option off, then. *shrug*)


    I saw your Revision 2020 talk by the way, and found it really interesting.

    Your solution to the prohibitively-expensive-certification problem is rather neat, I think.


    Just out of curiosity, I did some measurements with the power supplies that I have here. All measurements were made when the system was idle, after booting to Workbench. I did not test for different loads, so the results are probably of limited value, but here they are anyway:

    Code
    1. PSU Model Voltages[V] Ripple/Noise[mv]
    2. 1 A500 old 5.069 -12.20 12.06 45 48 54
    3. 2 A1200 old 4.989 -12.20 11.84 41 47 78
    4. 3 A1200 new 4.983 -11.65 12.75 62 70 54
    5. 4 Keelogg 4.840 -13.41 12.78 34 57 57
    6. 5 CA-PSU 5.078 -11.94 12.01 42 53 67


    The Keelogg PSU seems to be the worst of the bunch, which is kind of disappointing (but no surprise anymore, at this point).

    It's not only far too low on the 5V line but also exceeds the spec on the -12V line by -210mV, which is a bit unsettling.


    The original Amiga PSUs seem to hold up well for their age (except for the noise on #3). This is of course only for an "idling" load, so they will probably be worse under full load. Under full load, the 5.069V of #1 will probably drop below 5.05V so that would be fine (as long as it does not go below 4.95, which it might...)

    In your Revision 2020 talk, you show that the CA-PSU does not exhibit the usual linear voltage drop with increasing load, which is really nice. But that would mean that the 5.078V that I measured would stay outside the 1% spec, right? Should I be worried about that?

  • Should I be worried about that?

    Not unless you have a calibrated volt meter. Also, with higher load, a little more heat builds up inside of the CA-PSU converter, which raises the reference voltage, resulting in a slightly lower output voltage. That's the whole reason why I am "only" specifying 5A instead of the 8A that the unit can deliver indefinitely, at least in theory. However, with the added heat, the voltage might go out of range after several minutes of operation.


    The ripple you've measured for the CA-PSU indicates that it operates at very light load, but again, if you're looking at voltage differences in this small of an area, calibration and measurement method play a huge role. You'll find differences even measuring at different positions of the board.

  • I did indeed send my multimeter to the manufacturer for recalibration (below 0.2% in the relevant voltage range), but that was 3 or 4 years ago, so I don't know if that's still worth anything.

    The Amiga has now been on for about an hour and now I'm measuring 5.053V.

    If you're not concerned, I'm neither; I just wanted to make sure it's fine.

  • So, after a bit of tinkering with my A500, I'm now back to the 1200, and the problems are unfortunately not quite gone yet.


    I tested the Indivision AGA MK 3 with my 1D4 board again (now with the CA-PSU), and here the behaviour is slightly different than on the 2B board.


    For the first 1-3 minutes, graphics glitches appear when CCK pullup and capacitance are off, but when I turn both on, everything looks fine. After a short while, glitches start to appear again - now turning pullup and capacitance off again helps.

    So it looks like maybe there's some temperature-related stuff going on?


    Back on the 2B Board, pullup and capacitance seem to work fine directly after power-up, but the system constantly crashes when it has been on for a few minutes. Turning capacitance off fixes the crashes. (I have to leave the Amiga off for a few minutes so it is stable for long enough to start the config tool.)

    When the system has been powered on for a while, CCK-pullup on and CCK-cap off seems to be a stable setting for this board.


    I'm sorry if my reports appear to be partially contradictory or erratic. I'm also getting quite frustrated by this.

    If you have any more ideas, I'd be grateful.