No reason to be hurry, but if you are willing to measure the values =>Quote:
Originally Posted by xan
:toast:
Power supply DO have effect ! I have put most of my power supply thoughts to post #1, but here is short summary:Quote:
Originally Posted by xan
- I tested two supplies: very bad/cheapo and good zalman (very low ripple voltages)
- With good supply I was able to prevent BS in 'SkiAlpin' with win98 completely but not with w2k (I still got BS with win98, some cases in HL2)
- Added capacitors into MOLEX-connector helped considerably with poor supply (not tested with good one, since it has capacitors inside :D :D)
- I have allways (?) used a dedicated (own) molex-line for 6800le to have better current sourcing capability from supply. Good point 'xan' to mention this one :up:, I will add it to document..
When I was thinking BS-feature in *deep*, I found that the one of the most important fundamental problems with this design is the usage of 5 V supply for memory voltage generation. Vcore is generated from 12 V and I think the 'overhead' between voltages make Vcore fundamentally more stabile. The regulator converts Vin -> Vout (5 V -> 2.72 v, for Vmem), and here the ratio is far too low Vout/Vin > 0.5 . This makes the operation highly sensitive to all parameter changes as we have seen/proofed. I suppose (not 100 % proofed, yet) that having a SMALL over-voltage on 5 V Molex (5.25, 5.5 ?) the BS-propability would be reduced considerably in all cases. That would give clean/simple/sensible explantion why other CARDS are working better, since voltages varies between different PSU's.
--
I'll will be out-of-www some days - no fast comments until back.