My general rule of thumb is that if the board POSTs but is known to use garbage/ low quality caps, an complete recap is worth the effort. Especially as I see one bulger among those 1500s.

I think its gonna be an issue with the chipsets, although not sure which one. You could also check for any broken/missing components. Seems like theres also some caps bulging near the gameport.

FYI, TheRetroWeb still has no photo of this board. https://theretroweb.com/motherboards/s/soyo-sy-7vma-b

Deksor wrote:

Oh BTW, you might want to indicate which caps you already replaced. Looking at that photo there still seem to be some that should be replaced and likely those parallel to it as well.

That photo is from before it was recapped.

weedeewee wrote on 2024-04-17, 17:32:

FYI, TheRetroWeb still has no photo of this board. https://theretroweb.com/motherboards/s/soyo-sy-7vma-b

Deksor wrote:

Oh BTW, you might want to indicate which caps you already replaced. Looking at that photo there still seem to be some that should be replaced and likely those parallel to it as well.

Okay, I'll take a better photo and submit it to them.

There are the five 6.3 1500 caps near the socket that are original, four 16 volt 470s (all Teapo), and then the tiny caps - I've never gone so far as to replace the tiny capacitors on any motherboard so far. There were actually eleven of the 6.3 volt 1000 caps but one was physically damaged, so I replaced it when I replaced the 6.3 volt 2200 caps by the socket months ago.

Why I suspect it is a power issue: while it works fine with 512 megabytes, that's only if a single 512 is installed - if I install a pair of 256 DIMMs (both of which passed alone) it will fail at the end of test five. I'm assuming that a pair of DIMMs draws more power than a single of the same capacity, but I could be wrong.

Probably no the same thing, but I had a problem that drove me mad trying to track down, with sticks working fine on their own, but not together. Changed caps, rejigged the RAM power supply voltage to bypass the FET that switched between normal and standby, changed power supplies, carried on like this for hours. Errors appeared to be mostly random addresses, and only on the slow memtest tests (mod20 I think) but exactly which bit was always clustered in the same area. Turned out to be bad airflow and one chip overheating. Put 4 sticks in, pointed a case fan at them, left it for 24 hours and no errors.

I dont think old boards like that are effected but I had a socket 1366 system that got/gets memory errors using mixed manufacturers memory of same spec. Since its triple channel if I put 3 matched sticks from one vendor in it and 3 matched sticks from another vendor in the other 3 slots - all the exact same specs timings and everything I get errors in memcheck.

If I test the 3 sticks from one manufacturer alone they check fine, do the same thing with the other they are fine but combined they error and its not the motherboard mem slots. I later bought a set of 6 sticks of match memory from one manufacturer and it works perfect. It just didnt like mixed manufacturers even if the sticks were all the same spec.

I never had that problem on older motherboards but thought I would mention it.

I tried downclocking the memory from 133 MHz to 100 and et voila 768 megabytes with no Memtest errors! But I still want to run it at 133 (this is all PC133)...

I had some 47 and 100uF caps go nasty on one board a while back, think it was a Shuttle though.

BitWrangler wrote on 2024-04-18, 03:58:

I had some 47 and 100uF caps go nasty on one board a while back, think it was a Shuttle though.

What was the symptom that caused you to decide that they were at fault?

I didn't actually get as far as memtest, I recapped the main caps then the board got going, but it was 50/50 it got to a dos prompt, freezing in a random place after "Starting MSDOS" message, backed off some memory timing in CMOS and it got to prompt reliably, then it was crashing out of doom timedemo with either early termination or lockup... and the timings were as slack as they got, so examined board super carefully again and there were just hints of bulge and a very tiny brown speck on two of the smaller caps, but after I pulled them all, only 2 out of 12 still seemed functional on the component tester. .... After fresh ones went in, RAM timings could be set to something reasonable, got to prompt, ran Doom correctly, ran quake, and passed memtest.

I have replacements for the nine remaining Teapo caps on the way, but I don't feel at all confident that replacing them is going to solve this issue.

Quickest diagnosis is with scope on all voltage rails.

rasz_pl wrote on 2024-04-18, 19:21:

Quickest diagnosis is with scope on all voltage rails.

Sadly, I do not have one. My electronics repair experience seems to be quite limited in comparison to some of the members of this forums that's for sure.

Repo Man11 wrote on 2024-04-18, 15:36:

I have replacements for the nine remaining Teapo caps on the way, but I don't feel at all confident that replacing them is going to solve this issue.

I hate when life justifies my pessimism. All but the tiniest caps on this board have now been replaced with Nichicon ones, but the Memtest error remains. I did find one combination of a 512 and a 256 that will pass Memtest, but no combination of 512 megabyte sticks will pass at 133 MHz. Heavy sigh.

Repo Man11 wrote on 2024-05-05, 19:24:

All but the tiniest caps on this board have now been replaced with Nichicon ones, but the Memtest error remains.

Might be signal integrity issue due to how the PCB is routed, nothing you can do about that. Might be the sticks you have load the lines more or cause more reflections, so one still works but two is just too much for the memory controller.

One more thing you can try, though I doubt that will help, is to add 10-100 nF capacitor to the DIMM power supply rail / regulator. Just solder them to the stubs of the electrolytic caps on the bottom side of the PCB, if you need to add some wire to make it reach then use thick one. Solid core or braid are both fine just nothing very thin. It's important these caps are ceramic and not foil type. SMD MLCC would be best but will require some extra wire most likely.

Deunan wrote on 2024-05-05, 19:54:

Repo Man11 wrote on 2024-05-05, 19:24:

All but the tiniest caps on this board have now been replaced with Nichicon ones, but the Memtest error remains.

Might be signal integrity issue due to how the PCB is routed, nothing you can do about that. Might be the sticks you have load the lines more or cause more reflections, so one still works but two is just too much for the memory controller.

One more thing you can try, though I doubt that will help, is to add 10-100 nF capacitor to the DIMM power supply rail / regulator. Just solder them to the stubs of the electrolytic caps on the bottom side of the PCB, if you need to add some wire to make it reach then use thick one. Solid core or braid are both fine just nothing very thin. It's important these caps are ceramic and not foil type. SMD MLCC would be best but will require some extra wire most likely.

I appreciate the advice, but that's a bit too far for me. A limit of 768 megabytes of memory isn't the end of the world, it's still a useful motherboard.